port/master-into-v4 (#1176)

* 3.7.0 (#1175) * mark as next, introduce dynamic scale for face debugger * use latest onnxruntime * update within Gradio 5 * Remove system memory limit (#986) * remove system memory limit from ui * remove system memory limit from args.py * flatten the face store * prevent countless importlib.import_module calls * remove --onnxruntime from install.py * remove --onnxruntime from install.py * resolve static inference providers to fix macos (#1127) * resolve static inference providers to fix macos * fix lint * restore old behaviour * restore old behaviour * handle ghost and uniface as well * adjust condition for ghost and uniface * fix Gradio gallery styles * remove face store (#1132) * fix dataflow in streamer * Face selector auto mode (#1137) * introduce face selector auto mode * introduce face selector auto mode * introduce face selector auto mode * correct way is to pass source_vision_frames * make the world a better place * fix dataflow in faceswapper, no read of files withing inner methods (#1148) * fix dataflow in faceswapper, no read of files withing inner methods * fix lint * adjust code more * adjust code more * bring back the face store but for source and reference only (#1149) * bring back the face store but for source and reference only * fix ci * minor improvement * guard for tobytes() * drop condition in select_faces() * Replace CONFIG_PARSER global with @lru_cache (#1147) * remove global config_parser * fix import order * remove lambda * remove unused block * optimize app context detection * decouple common modules from core (#1152) * decouple common modules from core * remove that nonsense * remove that nonsense * minor adjustment to workflows * Tag HEVC output as hvc1 and move moov atom to the front (#1153) * Tag HEVC output as hvc1 and move moov atom to the front ffmpeg defaults HEVC in MP4 to the 'hev1' sample entry and leaves the moov atom at the tail. Apple players (QuickTime, Finder QuickLook) refuse to decode 'hev1' and stall reading a tail-placed moov on large files, so hevc_nvenc / libx265 renders cannot be previewed on macOS. - add ffmpeg_builder.set_video_tag(): emit `-tag:v hvc1` for every HEVC encoder (libx265, hevc_nvenc, hevc_amf, hevc_qsv, hevc_videotoolbox). Applied in merge_video where the encoder is known; `-c:v copy` in the audio mux / concat steps preserves the tag. - add ffmpeg_builder.set_faststart(): emit `-movflags +faststart`, applied in restore_audio / replace_audio / concat_video which write the final output. H.264 and other codecs are left untouched. Verified on a real hevc_nvenc render: hev1 hung QuickLook (no thumbnail); after the patch the file is hvc1 with a front-placed moov and QuickLook generates a thumbnail. * Restrict hvc1 tag and faststart to quicktime containers Gate set_video_tag / set_faststart on the output container format (m4v, mov, mp4) via get_file_format(), so non-quicktime muxers no longer receive -tag:v hvc1 / -movflags +faststart. Trim test_set_video_tag to a single positive and negative assertion. Addresses review on #1153. * Move hvc1 tag and faststart gates into ffmpeg_builder Rename set_video_tag / set_faststart to conditional_* and push the container-format gate (m4v, mov, mp4) inside the builders, keeping ffmpeg.py free of inline conditionals. Matches the set_image_quality pattern. Addresses review on #1153. * post cleanup after merge * Pack target frames (#1158) * pack target frames * add todos * add todos, resolve todos * resolve todos * change names * revert to single target frame for select faces * fix lint * return empty frame * get() have no default * Fix trim (#1162) * fix trim * fix trim * rename ffmpeg builder method * rename to temp_frame_set and temp_frame_pattern --------- Co-authored-by: harisreedhar <h4harisreedhar.s.s@gmail.com> Co-authored-by: Harisreedhar <46858047+harisreedhar@users.noreply.github.com> * Implement face tracker (#1163) * add face tracker * change get_nearest_track_face -> get_nearest_track_index * create face_creator.py and move methods around * add type FaceTrack * naming * remove iou test, don't belong there * fix spaces * rename to interpolate_points * rename to find_best_face_track * just track_faces * cleanp * previous next naming * remove >= and >= * rename * remove helper from test and use face from source.jpg * make get_anchor_indices more readable * track_faces() call before and is forwarded to select_faces * change to interpolate_faces * rename methods * rename methods * rename variables * remove dtype * move face_anlyser -> face_creator * claenup face_creator.py * move tests to dedicated test face detector * move tracking inside select_faces * simplify face_tracker (#1165) * minor renaming * improve face_tracker test (#1166) * improve face_tracker test * cleanup * Add target frame amount (#1167) * introduce --target-frame-amount * add ui * make track_faces conditional * update choices.py * fix [] * rename component file to frame_process.py * fix track preview (#1168) * introduce face origin (#1169) * add guard to prevent failure * show and hide voice extractor according to lip syncer * rename average_face_coordinates to average_face_geometry * use static faces for select_faces() * face store with lock (#1171) * face store with lock * face store with lock * remove refill color from bbox * adjust tests and handle frame_position proper way * enforce similar naming * introduce face tracker score * introduce face tracker score * fix/audio-trim-alignment (#1173) * fix audio offset * fix audio offset * remove reference_frame_number check --------- Co-authored-by: harisreedhar <h4harisreedhar.s.s@gmail.com> * reduce face tracker score from 0 to 0.5 * mark as 3.7.0 * make face tracker stateless --------- Co-authored-by: Harisreedhar <46858047+harisreedhar@users.noreply.github.com> Co-authored-by: kazuki nakai <kazuki.nakai@agiletec.net> Co-authored-by: harisreedhar <h4harisreedhar.s.s@gmail.com> * update preview * fix wording * fix wording * last minute change to frame distribution --------- Co-authored-by: Harisreedhar <46858047+harisreedhar@users.noreply.github.com> Co-authored-by: kazuki nakai <kazuki.nakai@agiletec.net> Co-authored-by: harisreedhar <h4harisreedhar.s.s@gmail.com>
2026-07-15 06:27:21 +02:00 · 2026-07-01 11:51:54 +02:00
parent 5f403ada65
commit 300470e7c7
47 changed files with 1148 additions and 345 deletions
@@ -35,6 +35,9 @@ reference_face_position =
 reference_face_distance =
 reference_frame_number =

+[face_tracker]
+face_tracker_score =
+
 [face_masker]
 face_occluder_model =
 face_parser_model =
@@ -52,6 +55,9 @@ trim_frame_start =
 trim_frame_end =
 temp_frame_format =

+[frame_distribution]
+target_frame_amount =
+
 [output_creation]
 output_image_quality =
 output_image_scale =
@@ -36,6 +36,7 @@ def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
 	apply_state_item('reference_face_position', args.get('reference_face_position'))
 	apply_state_item('reference_face_distance', args.get('reference_face_distance'))
 	apply_state_item('reference_frame_number', args.get('reference_frame_number'))
+	apply_state_item('face_tracker_score', args.get('face_tracker_score'))
 	apply_state_item('face_occluder_model', args.get('face_occluder_model'))
 	apply_state_item('face_parser_model', args.get('face_parser_model'))
 	apply_state_item('face_mask_types', args.get('face_mask_types'))
@@ -47,6 +48,7 @@ def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
 	apply_state_item('trim_frame_start', args.get('trim_frame_start'))
 	apply_state_item('trim_frame_end', args.get('trim_frame_end'))
 	apply_state_item('temp_frame_format', args.get('temp_frame_format'))
+	apply_state_item('target_frame_amount', args.get('target_frame_amount'))
 	apply_state_item('output_image_quality', args.get('output_image_quality'))
 	apply_state_item('output_image_scale', args.get('output_image_scale'))
 	apply_state_item('output_audio_encoder', args.get('output_audio_encoder'))
@@ -2,7 +2,7 @@ import logging
 from typing import List, Sequence, get_args

 from facefusion.common_helper import create_float_range, create_int_range
-from facefusion.types import Angle, ApiSecurityStrategy, AudioEncoder, AudioFormat, AudioSet, BenchmarkMode, BenchmarkResolution, BenchmarkSet, DownloadProvider, DownloadProviderSet, DownloadScope, ExecutionProvider, ExecutionProviderSet, FaceDetectorModel, FaceDetectorSet, FaceLandmarkerModel, FaceMaskArea, FaceMaskAreaSet, FaceMaskRegion, FaceMaskRegionSet, FaceMaskType, FaceOccluderModel, FaceParserModel, FaceSelectorMode, FaceSelectorOrder, Gender, ImageEncoder, ImageFormat, ImageSet, JobStatus, LogLevel, LogLevelSet, Race, Score, TempFrameFormat, VideoEncoder, VideoFormat, VideoMemoryStrategy, VideoPreset, VideoSet, VoiceExtractorModel, WorkFlow
+from facefusion.types import Angle, ApiSecurityStrategy, AudioEncoder, AudioFormat, AudioSet, BenchmarkMode, BenchmarkResolution, BenchmarkSet, DownloadProvider, DownloadProviderSet, DownloadScope, ExecutionProvider, ExecutionProviderSet, FaceDetectorModel, FaceDetectorSet, FaceLandmarkerModel, FaceMaskArea, FaceMaskAreaSet, FaceMaskRegion, FaceMaskRegionSet, FaceMaskType, FaceOccluderModel, FaceParserModel, FaceSelectorGender, FaceSelectorMode, FaceSelectorOrder, FaceSelectorRace, Gender, ImageEncoder, ImageFormat, ImageSet, JobStatus, LogLevel, LogLevelSet, Race, Score, TempFrameFormat, VideoEncoder, VideoFormat, VideoMemoryStrategy, VideoPreset, VideoSet, VoiceExtractorModel, WorkFlow

 face_detector_set : FaceDetectorSet =\
 {
@@ -16,8 +16,10 @@ face_detector_models : List[FaceDetectorModel] = list(get_args(FaceDetectorModel
 face_landmarker_models : List[FaceLandmarkerModel] = list(get_args(FaceLandmarkerModel))
 face_selector_modes : List[FaceSelectorMode] = list(get_args(FaceSelectorMode))
 face_selector_orders : List[FaceSelectorOrder] = list(get_args(FaceSelectorOrder))
-face_selector_genders : List[Gender] = list(get_args(Gender))
-face_selector_races : List[Race] = list(get_args(Race))
+genders : List[Gender] = list(get_args(Gender))
+races : List[Race] = list(get_args(Race))
+face_selector_genders : List[FaceSelectorGender] = list(get_args(FaceSelectorGender))
+face_selector_races : List[FaceSelectorRace] = list(get_args(FaceSelectorRace))
 face_occluder_models : List[FaceOccluderModel] = list(get_args(FaceOccluderModel))
 face_parser_models : List[FaceParserModel] = list(get_args(FaceParserModel))
 face_mask_types : List[FaceMaskType] = list(get_args(FaceMaskType))
@@ -159,6 +161,8 @@ face_mask_blur_range : Sequence[float] = create_float_range(0.0, 1.0, 0.05)
 face_mask_padding_range : Sequence[int] = create_int_range(0, 100, 1)
 face_selector_age_range : Sequence[int] = create_int_range(0, 100, 1)
 reference_face_distance_range : Sequence[float] = create_float_range(0.0, 1.0, 0.05)
+face_tracker_score_range : Sequence[Score] = create_float_range(0.0, 0.5, 0.05)
+target_frame_amount_range : Sequence[int] = create_int_range(0, 10, 1)
 output_image_quality_range : Sequence[int] = create_int_range(0, 100, 1)
 output_image_scale_range : Sequence[float] = create_float_range(0.25, 8.0, 0.25)
 output_audio_quality_range : Sequence[int] = create_int_range(0, 100, 1)
@@ -78,6 +78,12 @@ def get_first(__list__ : Any) -> Any:
 	return None


+def get_middle(__list__ : Any) -> Any:
+	if isinstance(__list__, Sequence) and __list__:
+		return __list__[len(__list__) // 2]
+	return None
+
+
 def get_last(__list__ : Any) -> Any:
 	if isinstance(__list__, Reversible):
 		return next(reversed(__list__), None)
@@ -3,10 +3,10 @@ from typing import List, Optional
 import numpy

 from facefusion import face_store, state_manager
-from facefusion.common_helper import get_first
+from facefusion.common_helper import get_first, get_middle
 from facefusion.face_classifier import classify_face
 from facefusion.face_detector import detect_faces, detect_faces_by_angle
-from facefusion.face_helper import apply_nms, convert_to_face_landmark_5, estimate_face_angle, get_nms_threshold
+from facefusion.face_helper import apply_nms, average_points, convert_to_face_landmark_5, estimate_face_angle, get_nms_threshold
 from facefusion.face_landmarker import detect_face_landmark, estimate_face_landmark_68_5
 from facefusion.face_recognizer import calculate_face_embedding
 from facefusion.types import BoundingBox, Face, FaceLandmark5, FaceLandmarkSet, FaceScoreSet, Score, VisionFrame
@@ -46,7 +46,9 @@ def create_faces(vision_frame : VisionFrame, bounding_boxes : List[BoundingBox],
 		}
 		face_embedding, face_embedding_norm = calculate_face_embedding(vision_frame, face_landmark_set.get('5/68'))
 		gender, age, race = classify_face(vision_frame, face_landmark_set.get('5/68'))
+
 		faces.append(Face(
+			origin = 'detect',
 			bounding_box = bounding_box,
 			score_set = face_score_set,
 			landmark_set = face_landmark_set,
@@ -67,48 +69,6 @@ def get_one_face(faces : List[Face], position : int = 0) -> Optional[Face]:
 	return None


-def get_average_face(faces : List[Face]) -> Optional[Face]:
-	face_embeddings = []
-	face_embeddings_norm = []
-
-	if faces:
-		first_face = get_first(faces)
-
-		for face in faces:
-			face_embeddings.append(face.embedding)
-			face_embeddings_norm.append(face.embedding_norm)
-
-		return Face(
-			bounding_box = first_face.bounding_box,
-			score_set = first_face.score_set,
-			landmark_set = first_face.landmark_set,
-			angle = first_face.angle,
-			embedding = numpy.mean(face_embeddings, axis = 0),
-			embedding_norm = numpy.mean(face_embeddings_norm, axis = 0),
-			gender = first_face.gender,
-			age = first_face.age,
-			race = first_face.race
-		)
-	return None
-
-
-def get_static_faces(vision_frames : List[VisionFrame]) -> List[Face]:
-	many_faces : List[Face] = []
-
-	for vision_frame in vision_frames:
-		faces = face_store.get_faces(vision_frame)
-
-		if not faces:
-			faces = get_many_faces([ vision_frame ])
-
-			if faces:
-				face_store.set_faces(vision_frame, faces)
-
-		many_faces.extend(faces)
-
-	return many_faces
-
-
 def get_many_faces(vision_frames : List[VisionFrame]) -> List[Face]:
 	many_faces : List[Face] = []

@@ -136,6 +96,100 @@ def get_many_faces(vision_frames : List[VisionFrame]) -> List[Face]:
 	return many_faces


+def get_static_faces(vision_frames : List[VisionFrame]) -> List[Face]:
+	many_faces : List[Face] = []
+
+	for vision_frame in vision_frames:
+		faces = face_store.get_faces(vision_frame)
+
+		if not faces:
+			with face_store.resolve_lock(vision_frame):
+				faces = face_store.get_faces(vision_frame)
+
+				if not faces:
+					faces = get_many_faces([ vision_frame ])
+
+					if faces:
+						face_store.set_faces(vision_frame, faces)
+
+		many_faces.extend(faces)
+
+	return many_faces
+
+
+def refill_faces(faces : List[Optional[Face]]) -> List[Face]:
+	fill_faces = []
+	anchor_index_previous = -1
+
+	for index, face in enumerate(faces):
+		if face:
+			for gap_index in range(anchor_index_previous + 1, index):
+				average_factor = (gap_index - anchor_index_previous) / (index - anchor_index_previous)
+				average_face = average_face_geometry([faces[anchor_index_previous], face], average_factor)
+				fill_faces.append(average_face)
+
+			fill_faces.append(face)
+			anchor_index_previous = index
+
+	return fill_faces
+
+
+def average_face_geometry(faces : List[Face], average_factor : float) -> Face:
+	face_first = get_first(faces)
+	face_middle = get_middle(faces)
+	face_anchor = face_middle
+
+	if average_factor < 0.5:
+		face_anchor = face_first
+
+	landmark_set : FaceLandmarkSet =\
+	{
+		'5': average_points(face_first.landmark_set.get('5'), face_middle.landmark_set.get('5'), average_factor),
+		'5/68': average_points(face_first.landmark_set.get('5/68'), face_middle.landmark_set.get('5/68'), average_factor),
+		'68': average_points(face_first.landmark_set.get('68'), face_middle.landmark_set.get('68'), average_factor),
+		'68/5': average_points(face_first.landmark_set.get('68/5'), face_middle.landmark_set.get('68/5'), average_factor)
+	}
+
+	return Face(
+		origin = 'refill',
+		bounding_box = average_points(face_first.bounding_box, face_middle.bounding_box, average_factor),
+		score_set = face_anchor.score_set,
+		landmark_set = landmark_set,
+		angle = estimate_face_angle(landmark_set.get('68/5')),
+		embedding = face_anchor.embedding,
+		embedding_norm = face_anchor.embedding_norm,
+		gender = face_anchor.gender,
+		age = face_anchor.age,
+		race = face_anchor.race
+	)
+
+
+def average_face_identity(faces : List[Face]) -> Optional[Face]:
+	face_embeddings = []
+	face_embeddings_norm = []
+
+	if faces:
+		first_face = get_first(faces)
+
+		for face in faces:
+			face_embeddings.append(face.embedding)
+			face_embeddings_norm.append(face.embedding_norm)
+
+		return Face(
+			origin = first_face.origin,
+			bounding_box = first_face.bounding_box,
+			score_set = first_face.score_set,
+			landmark_set = first_face.landmark_set,
+			angle = first_face.angle,
+			embedding = numpy.mean(face_embeddings, axis = 0),
+			embedding_norm = numpy.mean(face_embeddings_norm, axis = 0),
+			gender = first_face.gender,
+			age = first_face.age,
+			race = first_face.race
+		)
+	return None
+
+
 def scale_face(target_face : Face, target_vision_frame : VisionFrame, temp_vision_frame : VisionFrame) -> Face:
 	scale_x = temp_vision_frame.shape[1] / target_vision_frame.shape[1]
 	scale_y = temp_vision_frame.shape[0] / target_vision_frame.shape[0]
@@ -254,3 +254,23 @@ def merge_matrix(temp_matrices : List[Matrix]) -> Matrix:
 		matrix = numpy.dot(temp_matrix, matrix)

 	return matrix[:2, :]
+
+
+def calculate_bounding_box_overlap(bounding_box_a : BoundingBox, bounding_box_b : BoundingBox) -> float:
+	intersection_x1 = max(bounding_box_a[0], bounding_box_b[0])
+	intersection_y1 = max(bounding_box_a[1], bounding_box_b[1])
+	intersection_x2 = min(bounding_box_a[2], bounding_box_b[2])
+	intersection_y2 = min(bounding_box_a[3], bounding_box_b[3])
+	intersection = max(0, intersection_x2 - intersection_x1) * max(0, intersection_y2 - intersection_y1)
+	bounding_box_area = (bounding_box_a[2] - bounding_box_a[0]) * (bounding_box_a[3] - bounding_box_a[1])
+	reference_bounding_box_area = (bounding_box_b[2] - bounding_box_b[0]) * (bounding_box_b[3] - bounding_box_b[1])
+	union = bounding_box_area + reference_bounding_box_area - intersection
+
+	if union > 0:
+		return intersection / union
+
+	return 0.0
+
+
+def average_points(points_previous : Points, points_next : Points, average_factor : float) -> Points:
+	return points_previous * (1 - average_factor) + points_next * average_factor
@@ -2,26 +2,35 @@ from typing import List

 import numpy

+import facefusion.choices
 from facefusion import state_manager
-from facefusion.face_analyser import get_many_faces, get_one_face, get_static_faces
+from facefusion.common_helper import get_first, get_middle
+from facefusion.face_creator import get_one_face, get_static_faces
+from facefusion.face_tracker import track_faces
 from facefusion.types import Face, FaceSelectorOrder, Gender, Race, Score, VisionFrame


-def select_faces(reference_vision_frame : VisionFrame, target_vision_frame : VisionFrame) -> List[Face]:
-	target_faces = get_many_faces([ target_vision_frame ])
+def select_faces(reference_vision_frame : VisionFrame, source_vision_frames : List[VisionFrame], target_vision_frames : List[VisionFrame]) -> List[Face]:
+	source_faces = get_static_faces(source_vision_frames)
+
+	if state_manager.get_item('face_tracker_score') > 0:
+		target_faces = track_faces(target_vision_frames, state_manager.get_item('face_tracker_score'))
+	else:
+		target_faces = get_static_faces([ get_middle(target_vision_frames) ])

 	if state_manager.get_item('face_selector_mode') == 'many':
-		return sort_and_filter_faces(target_faces)
+		return sort_and_filter_faces(source_faces, target_faces)

 	if state_manager.get_item('face_selector_mode') == 'one':
-		target_face = get_one_face(sort_and_filter_faces(target_faces))
+		target_face = get_one_face(sort_and_filter_faces(source_faces, target_faces))
 		if target_face:
 			return [ target_face ]

 	if state_manager.get_item('face_selector_mode') == 'reference':
 		reference_faces = get_static_faces([ reference_vision_frame ])
-		reference_faces = sort_and_filter_faces(reference_faces)
+		reference_faces = sort_and_filter_faces(source_faces, reference_faces)
 		reference_face = get_one_face(reference_faces, state_manager.get_item('reference_face_position'))
+
 		if reference_face:
 			match_faces = find_match_faces([ reference_face ], target_faces, state_manager.get_item('reference_face_distance'))
 			return match_faces
@@ -53,17 +62,33 @@ def calculate_face_distance(face : Face, reference_face : Face) -> float:
 	return 0


-def sort_and_filter_faces(faces : List[Face]) -> List[Face]:
-	if faces:
+def sort_and_filter_faces(source_faces : List[Face], target_faces : List[Face]) -> List[Face]:
+	if target_faces:
 		if state_manager.get_item('face_selector_order'):
-			faces = sort_faces_by_order(faces, state_manager.get_item('face_selector_order'))
-		if state_manager.get_item('face_selector_gender'):
-			faces = filter_faces_by_gender(faces, state_manager.get_item('face_selector_gender'))
-		if state_manager.get_item('face_selector_race'):
-			faces = filter_faces_by_race(faces, state_manager.get_item('face_selector_race'))
+			target_faces = sort_faces_by_order(target_faces, state_manager.get_item('face_selector_order'))
+
+		face_selector_gender = state_manager.get_item('face_selector_gender')
+		face_selector_race = state_manager.get_item('face_selector_race')
+
+		if source_faces and face_selector_gender == 'auto' or face_selector_race == 'auto':
+			source_face = get_first(sort_faces_by_order(source_faces, 'large-small'))
+
+			if source_face:
+				if face_selector_gender == 'auto':
+					face_selector_gender = source_face.gender
+				if face_selector_race == 'auto':
+					face_selector_race = source_face.race
+
+		if face_selector_gender in facefusion.choices.genders:
+			target_faces = filter_faces_by_gender(target_faces, face_selector_gender)
+
+		if face_selector_race in facefusion.choices.races:
+			target_faces = filter_faces_by_race(target_faces, face_selector_race)
+
 		if state_manager.get_item('face_selector_age_start') or state_manager.get_item('face_selector_age_end'):
-			faces = filter_faces_by_age(faces, state_manager.get_item('face_selector_age_start'), state_manager.get_item('face_selector_age_end'))
-	return faces
+			target_faces = filter_faces_by_age(target_faces, state_manager.get_item('face_selector_age_start'), state_manager.get_item('face_selector_age_end'))
+
+	return target_faces


 def sort_faces_by_order(faces : List[Face], order : FaceSelectorOrder) -> List[Face]:
@@ -1,3 +1,4 @@
+import threading
 from typing import List, Optional

 import numpy
@@ -11,14 +12,30 @@ FACE_STORE : FaceStore = {}
 def get_faces(vision_frame : VisionFrame) -> Optional[List[Face]]:
 	if numpy.any(vision_frame):
 		vision_hash = create_hash(vision_frame.tobytes())
-		return FACE_STORE.get(vision_hash)
+
+		if FACE_STORE.get(vision_hash):
+			return FACE_STORE.get(vision_hash).get('faces')
+
 	return None


 def set_faces(vision_frame : VisionFrame, faces : List[Face]) -> None:
 	if numpy.any(vision_frame):
 		vision_hash = create_hash(vision_frame.tobytes())
-		FACE_STORE[vision_hash] = faces
+		FACE_STORE.setdefault(vision_hash,
+		{
+			'lock': threading.Lock()
+		})['faces'] = faces
+
+
+def resolve_lock(vision_frame : VisionFrame) -> threading.Lock:
+	if numpy.any(vision_frame):
+		vision_hash = create_hash(vision_frame.tobytes())
+		return FACE_STORE.setdefault(vision_hash,
+		{
+			'lock': threading.Lock()
+		}).get('lock')
+	return threading.Lock()


 def clear_faces() -> None:
@@ -0,0 +1,61 @@
+from typing import List
+
+from facefusion.common_helper import get_first, get_last
+from facefusion.face_creator import get_static_faces, refill_faces
+from facefusion.face_helper import calculate_bounding_box_overlap
+from facefusion.types import Face, FaceTrack, Score, VisionFrame
+
+
+def track_faces(vision_frames : List[VisionFrame], score : Score) -> List[Face]:
+	target_index = len(vision_frames) // 2
+	face_tracks = create_face_tracks(vision_frames, score)
+	temp_faces = []
+
+	for face_track in face_tracks:
+		track_indices = sorted(face_track)
+		track_index_first = get_first(track_indices)
+		track_index_last = get_last(track_indices)
+		track_range = range(track_index_first, track_index_last + 1)
+
+		if target_index in track_range:
+			fill_faces = []
+
+			for index in track_range:
+				fill_faces.append(face_track.get(index))
+
+			temp_faces.append(refill_faces(fill_faces)[target_index - track_index_first])
+
+	return temp_faces
+
+
+def create_face_tracks(vision_frames : List[VisionFrame], score : Score) -> List[FaceTrack]:
+	face_tracks : List[FaceTrack] = []
+
+	for frame_index, vision_frame in enumerate(vision_frames):
+		for face in get_static_faces([ vision_frame ]):
+			face_track = select_face_track(face_tracks, face, score)
+
+			if face_track:
+				face_track[frame_index] = face
+			else:
+				face_tracks.append(
+				{
+					frame_index : face
+				})
+
+	return face_tracks
+
+
+def select_face_track(face_tracks : List[FaceTrack], face : Face, score : Score) -> FaceTrack:
+	select_track : FaceTrack = {}
+	select_score = score
+
+	for face_track in face_tracks:
+		track_face = face_track.get(get_last(face_track))
+		track_score = calculate_bounding_box_overlap(face.bounding_box, track_face.bounding_box)
+
+		if track_score > select_score:
+			select_score = track_score
+			select_track = face_track
+
+	return select_track
@@ -132,6 +132,7 @@ def extract_frames(target_path : str, output_path : str, temp_video_resolution :
 		ffmpeg_builder.enforce_pixel_format('rgb24'),
 		ffmpeg_builder.select_frame_range(trim_frame_start, trim_frame_end, temp_video_fps),
 		ffmpeg_builder.prevent_frame_drop(),
+		ffmpeg_builder.set_start_number(trim_frame_start),
 		ffmpeg_builder.set_output(temp_frames_pattern)
 	)

@@ -207,6 +208,7 @@ def restore_audio(target_path : str, output_path : str, trim_frame_start : int,
 	temp_video_path = get_temp_file_path(state_manager.get_temp_path(), output_path)
 	temp_video_format = cast(VideoFormat, get_file_format(output_path))
 	temp_video_duration = detect_video_duration(temp_video_path)
+	output_video_format = cast(VideoFormat, get_file_format(output_path))

 	output_audio_encoder = fix_audio_encoder(temp_video_format, output_audio_encoder)
 	commands = ffmpeg_builder.chain(
@@ -220,6 +222,7 @@ def restore_audio(target_path : str, output_path : str, trim_frame_start : int,
 		ffmpeg_builder.select_media_stream('0:v:0'),
 		ffmpeg_builder.select_media_stream('1:a:0'),
 		ffmpeg_builder.set_video_duration(temp_video_duration),
+		ffmpeg_builder.set_faststart(output_video_format),
 		ffmpeg_builder.force_output(output_path)
 	)
 	return run_ffmpeg(commands).returncode == 0
@@ -232,6 +235,7 @@ def replace_audio(audio_path : str, output_path : str) -> bool:
 	temp_video_path = get_temp_file_path(state_manager.get_temp_path(), output_path)
 	temp_video_format = cast(VideoFormat, get_file_format(output_path))
 	temp_video_duration = detect_video_duration(temp_video_path)
+	output_video_format = cast(VideoFormat, get_file_format(output_path))

 	output_audio_encoder = fix_audio_encoder(temp_video_format, output_audio_encoder)
 	commands = ffmpeg_builder.chain(
@@ -242,6 +246,7 @@ def replace_audio(audio_path : str, output_path : str) -> bool:
 		ffmpeg_builder.set_audio_quality(output_audio_encoder, output_audio_quality),
 		ffmpeg_builder.set_audio_volume(output_audio_volume),
 		ffmpeg_builder.set_video_duration(temp_video_duration),
+		ffmpeg_builder.set_faststart(output_video_format),
 		ffmpeg_builder.force_output(output_path)
 	)
 	return run_ffmpeg(commands).returncode == 0
@@ -259,9 +264,11 @@ def merge_video(target_path : str, output_path : str, temp_video_fps : Fps, outp
 	output_video_encoder = fix_video_encoder(temp_video_format, output_video_encoder)
 	commands = ffmpeg_builder.chain(
 		ffmpeg_builder.set_input_fps(temp_video_fps),
+		ffmpeg_builder.set_start_number(trim_frame_start),
 		ffmpeg_builder.set_input(temp_frames_pattern),
 		ffmpeg_builder.set_media_resolution(pack_resolution(output_video_resolution)),
 		ffmpeg_builder.set_video_encoder(output_video_encoder),
+		ffmpeg_builder.set_video_tag(output_video_encoder, temp_video_format),
 		ffmpeg_builder.set_video_quality(output_video_encoder, output_video_quality),
 		ffmpeg_builder.set_video_preset(output_video_encoder, output_video_preset),
 		ffmpeg_builder.concat(
@@ -288,11 +295,13 @@ def concat_video(output_path : str, temp_output_paths : List[str]) -> bool:
 		concat_video_file.close()

 	output_path = os.path.abspath(output_path)
+	output_video_format = cast(VideoFormat, get_file_format(output_path))
 	commands = ffmpeg_builder.chain(
 		ffmpeg_builder.unsafe_concat(),
 		ffmpeg_builder.set_input(concat_video_file.name),
 		ffmpeg_builder.copy_video_encoder(),
 		ffmpeg_builder.copy_audio_encoder(),
+		ffmpeg_builder.set_faststart(output_video_format),
 		ffmpeg_builder.force_output(output_path)
 	)
 	process = run_ffmpeg(commands)
@@ -5,7 +5,7 @@ from typing import List, Optional
 import numpy

 from facefusion.filesystem import get_file_format
-from facefusion.types import AudioEncoder, Command, CommandSet, Duration, Fps, SampleRate, StreamMode, VideoEncoder, VideoPreset
+from facefusion.types import AudioEncoder, Command, CommandSet, Duration, Fps, SampleRate, StreamMode, VideoEncoder, VideoFormat, VideoPreset


 def run(commands : List[Command]) -> List[Command]:
@@ -51,6 +51,10 @@ def set_input_fps(input_fps : Fps) -> List[Command]:
 	return [ '-r', str(input_fps) ]


+def set_start_number(frame_number : int) -> List[Command]:
+	return [ '-start_number', str(frame_number) ]
+
+
 def set_output(output_path : str) -> List[Command]:
 	return [ output_path ]

@@ -207,6 +211,18 @@ def copy_video_encoder() -> List[Command]:
 	return set_video_encoder('copy')


+def set_faststart(video_format : VideoFormat) -> List[Command]:
+	if video_format in [ 'm4v', 'mov', 'mp4' ]:
+		return [ '-movflags', '+faststart' ]
+	return []
+
+
+def set_video_tag(video_encoder : VideoEncoder, video_format : VideoFormat) -> List[Command]:
+	if video_format in [ 'm4v', 'mov', 'mp4' ] and video_encoder in [ 'libx265', 'hevc_nvenc', 'hevc_amf', 'hevc_qsv', 'hevc_videotoolbox' ]:
+		return [ '-tag:v', 'hvc1' ]
+	return []
+
+
 def set_video_quality(video_encoder : VideoEncoder, video_quality : int) -> List[Command]:
 	if video_encoder in [ 'libx264', 'libx264rgb', 'libx265' ]:
 		video_compression = numpy.round(numpy.interp(video_quality, [ 0, 100 ], [ 51, 0 ])).astype(int).item()
@@ -129,6 +129,7 @@ LOCALES : Locales =\
 			'reference_face_position': 'specify the position used to create the reference face',
 			'reference_face_distance': 'specify the similarity between the reference face and target face',
 			'reference_frame_number': 'specify the frame used to create the reference face',
+			'face_tracker_score': 'specify the overlap score used to match the tracked faces',
 			'face_occluder_model': 'choose the model responsible for the occlusion mask',
 			'face_parser_model': 'choose the model responsible for the region mask',
 			'face_mask_types': 'mix and match different face mask types (choices: {choices})',
@@ -140,6 +141,7 @@ LOCALES : Locales =\
 			'trim_frame_start': 'specify the starting frame of the target video',
 			'trim_frame_end': 'specify the ending frame of the target video',
 			'temp_frame_format': 'specify the temporary resources format',
+			'target_frame_amount': 'specify the amount of target frames forwarded to the processor',
 			'output_image_quality': 'specify the image quality which translates to the image compression',
 			'output_image_scale': 'specify the image scale based on the target image',
 			'output_audio_encoder': 'specify the encoder used for the audio',
@@ -228,6 +230,7 @@ LOCALES : Locales =\
 			'face_selector_mode_dropdown': 'FACE SELECTOR MODE',
 			'face_selector_order_dropdown': 'FACE SELECTOR ORDER',
 			'face_selector_race_dropdown': 'FACE SELECTOR RACE',
+			'face_tracker_score_slider': 'FACE TRACKER SCORE',
 			'face_occluder_model_dropdown': 'FACE OCCLUDER MODEL',
 			'face_parser_model_dropdown': 'FACE PARSER MODEL',
 			'voice_extractor_model_dropdown': 'VOICE EXTRACTOR MODEL',
@@ -12,6 +12,7 @@ PROCESSORS_METHODS =\
 	'clear_inference_pool',
 	'register_args',
 	'apply_args',
+	'get_common_modules',
 	'pre_check',
 	'pre_process',
 	'post_process',
@@ -1,5 +1,7 @@
 from argparse import ArgumentParser
 from functools import lru_cache
+from types import ModuleType
+from typing import List

 import cv2
 import numpy
@@ -8,10 +10,9 @@ import facefusion.capability_store
 import facefusion.choices
 import facefusion.jobs.job_manager
 from facefusion import config, content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer, inference_manager, logger, state_manager, translator, video_manager
-from facefusion.common_helper import create_int_metavar, is_macos
+from facefusion.common_helper import create_int_metavar, get_middle
 from facefusion.download import conditional_download_hashes, conditional_download_sources, resolve_download_url
-from facefusion.execution import has_execution_provider
-from facefusion.face_analyser import scale_face
+from facefusion.face_creator import scale_face
 from facefusion.face_helper import merge_matrix, paste_back, scale_face_landmark_5, warp_face_by_face_landmark_5
 from facefusion.face_masker import create_box_mask, create_occlusion_mask
 from facefusion.face_selector import select_faces
@@ -22,7 +23,7 @@ from facefusion.processors.types import ApplyStateItem, ProcessorOutputs
 from facefusion.program_helper import find_argument_group
 from facefusion.thread_helper import thread_semaphore
 from facefusion.types import Args, DownloadScope, Face, InferencePool, ModelOptions, ModelSet, ProcessMode, VisionFrame
-from facefusion.vision import match_frame_color, read_static_image, read_static_video_frame
+from facefusion.vision import match_frame_color, read_static_image, read_static_video_chunk, read_static_video_frame


@lru_cache()
@@ -150,10 +151,18 @@ def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
 	apply_state_item('age_modifier_direction', args.get('age_modifier_direction'))


+def get_common_modules() -> List[ModuleType]:
+	return [ content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer ]
+
+
 def pre_check() -> bool:
 	model_hash_set = get_model_options().get('hashes')
 	model_source_set = get_model_options().get('sources')

+	for common_module in get_common_modules():
+		if not common_module.pre_check():
+			return False
+
 	return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)


@@ -171,16 +180,15 @@ def pre_process(mode : ProcessMode) -> bool:
 def post_process() -> None:
 	read_static_image.cache_clear()
 	read_static_video_frame.cache_clear()
+	read_static_video_chunk.cache_clear()
 	video_manager.clear_video_pool()
+
 	if state_manager.get_item('video_memory_strategy') in [ 'strict', 'moderate' ]:
 		clear_inference_pool()
+
 	if state_manager.get_item('video_memory_strategy') == 'strict':
-		content_analyser.clear_inference_pool()
-		face_classifier.clear_inference_pool()
-		face_detector.clear_inference_pool()
-		face_landmarker.clear_inference_pool()
-		face_masker.clear_inference_pool()
-		face_recognizer.clear_inference_pool()
+		for common_module in get_common_modules():
+			common_module.clear_inference_pool()


 def modify_age(target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
@@ -245,9 +253,6 @@ def forward(crop_vision_frame : VisionFrame, extend_vision_frame : VisionFrame,
 	age_modifier = get_inference_pool().get('age_modifier')
 	age_modifier_inputs = {}

-	if is_macos() and has_execution_provider('coreml'):
-		age_modifier.set_providers([ facefusion.choices.execution_provider_set.get('cpu') ])
-
 	for age_modifier_input in age_modifier.get_inputs():
 		if age_modifier_input.name == 'target':
 			age_modifier_inputs[age_modifier_input.name] = crop_vision_frame
@@ -294,10 +299,13 @@ def normalize_extend_frame(extend_vision_frame : VisionFrame) -> VisionFrame:

 def process_frame(inputs : AgeModifierInputs) -> ProcessorOutputs:
 	reference_vision_frame = inputs.get('reference_vision_frame')
-	target_vision_frame = inputs.get('target_vision_frame')
+	source_vision_frames = inputs.get('source_vision_frames')
+	target_vision_frames = inputs.get('target_vision_frames')
 	temp_vision_frame = inputs.get('temp_vision_frame')
 	temp_vision_mask = inputs.get('temp_vision_mask')
-	target_faces = select_faces(reference_vision_frame, target_vision_frame)
+
+	target_vision_frame = get_middle(target_vision_frames)
+	target_faces = select_faces(reference_vision_frame, source_vision_frames, target_vision_frames)

 	if target_faces:
 		for target_face in target_faces:
@@ -1,4 +1,4 @@
-from typing import Any, Literal, TypeAlias, TypedDict
+from typing import Any, List, Literal, TypeAlias, TypedDict

 from numpy.typing import NDArray

@@ -7,7 +7,8 @@ from facefusion.types import Mask, VisionFrame
 AgeModifierInputs = TypedDict('AgeModifierInputs',
 {
 	'reference_vision_frame' : VisionFrame,
-	'target_vision_frame' : VisionFrame,
+	'source_vision_frames' : List[VisionFrame],
+	'target_vision_frames' : List[VisionFrame],
 	'temp_vision_frame' : VisionFrame,
 	'temp_vision_mask' : Mask
 })
@@ -1,11 +1,13 @@
 from argparse import ArgumentParser
 from functools import lru_cache, partial
+from types import ModuleType
 from typing import List, Tuple

 import cv2
 import numpy

 import facefusion.capability_store
+import facefusion.choices
 import facefusion.jobs.job_manager
 from facefusion import config, content_analyser, inference_manager, logger, state_manager, translator, video_manager
 from facefusion.common_helper import is_macos, is_windows
@@ -19,8 +21,8 @@ from facefusion.processors.types import ApplyStateItem, ProcessorOutputs
 from facefusion.program_helper import find_argument_group
 from facefusion.sanitizer import sanitize_int_range
 from facefusion.thread_helper import thread_semaphore
-from facefusion.types import Args, DownloadScope, ExecutionProvider, InferencePool, Mask, ModelOptions, ModelSet, ProcessMode, VisionFrame
-from facefusion.vision import read_static_image, read_static_video_frame
+from facefusion.types import Args, DownloadScope, InferencePool, InferenceProvider, Mask, ModelOptions, ModelSet, ProcessMode, VisionFrame
+from facefusion.vision import read_static_image, read_static_video_chunk, read_static_video_frame


@lru_cache()
@@ -477,12 +479,13 @@ def clear_inference_pool() -> None:
 	inference_manager.clear_inference_pool(__name__, model_names)


-def resolve_execution_providers() -> List[ExecutionProvider]:
+def resolve_inference_providers() -> List[InferenceProvider]:
 	model_type = get_model_options().get('type')

 	if is_macos() and has_execution_provider('coreml') or is_windows() and has_execution_provider('directml') and model_type == 'corridor_key':
-		return [ 'cpu' ]
-	return state_manager.get_item('execution_providers')
+		return [ facefusion.choices.execution_provider_set.get('cpu') ]
+
+	return []


 def get_model_options() -> ModelOptions:
@@ -526,10 +529,18 @@ def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
 	apply_state_item('background_remover_despill_color', normalize_color(args.get('background_remover_despill_color')))


+def get_common_modules() -> List[ModuleType]:
+	return [ content_analyser ]
+
+
 def pre_check() -> bool:
 	model_hash_set = get_model_options().get('hashes')
 	model_source_set = get_model_options().get('sources')

+	for common_module in get_common_modules():
+		if not common_module.pre_check():
+			return False
+
 	return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)


@@ -547,11 +558,15 @@ def pre_process(mode : ProcessMode) -> bool:
 def post_process() -> None:
 	read_static_image.cache_clear()
 	read_static_video_frame.cache_clear()
+	read_static_video_chunk.cache_clear()
 	video_manager.clear_video_pool()
+
 	if state_manager.get_item('video_memory_strategy') in [ 'strict', 'moderate' ]:
 		clear_inference_pool()
+
 	if state_manager.get_item('video_memory_strategy') == 'strict':
-		content_analyser.clear_inference_pool()
+		for common_module in get_common_modules():
+			common_module.clear_inference_pool()


 def remove_background(temp_vision_frame : VisionFrame) -> Tuple[VisionFrame, Mask]:
@@ -1,10 +1,10 @@
-from typing import Literal, TypedDict
+from typing import List, Literal, TypedDict

 from facefusion.types import Mask, VisionFrame

 BackgroundRemoverInputs = TypedDict('BackgroundRemoverInputs',
 {
-	'target_vision_frame' : VisionFrame,
+	'target_vision_frames' : List[VisionFrame],
 	'temp_vision_frame' : VisionFrame,
 	'temp_vision_mask' : Mask
 })
@@ -1,6 +1,7 @@
 from argparse import ArgumentParser
 from functools import lru_cache
-from typing import Tuple
+from types import ModuleType
+from typing import List, Tuple

 import cv2
 import numpy
@@ -9,9 +10,9 @@ from cv2.typing import Size
 import facefusion.capability_store
 import facefusion.jobs.job_manager
 from facefusion import config, content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer, inference_manager, logger, state_manager, translator, video_manager
-from facefusion.common_helper import create_int_metavar
+from facefusion.common_helper import create_int_metavar, get_middle
 from facefusion.download import conditional_download_hashes, conditional_download_sources, resolve_download_url_by_provider
-from facefusion.face_analyser import scale_face
+from facefusion.face_creator import scale_face
 from facefusion.face_helper import paste_back, warp_face_by_face_landmark_5
 from facefusion.face_masker import create_area_mask, create_box_mask, create_occlusion_mask, create_region_mask
 from facefusion.face_selector import select_faces
@@ -22,7 +23,7 @@ from facefusion.processors.types import ApplyStateItem, ProcessorOutputs
 from facefusion.program_helper import find_argument_group
 from facefusion.thread_helper import thread_semaphore
 from facefusion.types import Args, DownloadScope, Face, InferencePool, Mask, ModelOptions, ModelSet, ProcessMode, VisionFrame
-from facefusion.vision import conditional_match_frame_color, read_static_image, read_static_video_frame
+from facefusion.vision import conditional_match_frame_color, read_static_image, read_static_video_chunk, read_static_video_frame


@lru_cache()
@@ -302,10 +303,18 @@ def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
 	apply_state_item('deep_swapper_morph', args.get('deep_swapper_morph'))


+def get_common_modules() -> List[ModuleType]:
+	return [ content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer ]
+
+
 def pre_check() -> bool:
 	model_hash_set = get_model_options().get('hashes')
 	model_source_set = get_model_options().get('sources')

+	for common_module in get_common_modules():
+		if not common_module.pre_check():
+			return False
+
 	if model_hash_set and model_source_set:
 		return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)
 	return True
@@ -325,16 +334,15 @@ def pre_process(mode : ProcessMode) -> bool:
 def post_process() -> None:
 	read_static_image.cache_clear()
 	read_static_video_frame.cache_clear()
+	read_static_video_chunk.cache_clear()
 	video_manager.clear_video_pool()
+
 	if state_manager.get_item('video_memory_strategy') in [ 'strict', 'moderate' ]:
 		clear_inference_pool()
+
 	if state_manager.get_item('video_memory_strategy') == 'strict':
-		content_analyser.clear_inference_pool()
-		face_classifier.clear_inference_pool()
-		face_detector.clear_inference_pool()
-		face_landmarker.clear_inference_pool()
-		face_masker.clear_inference_pool()
-		face_recognizer.clear_inference_pool()
+		for common_module in get_common_modules():
+			common_module.clear_inference_pool()


 def swap_face(target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
@@ -425,10 +433,13 @@ def prepare_crop_mask(crop_source_mask : Mask, crop_target_mask : Mask) -> Mask:

 def process_frame(inputs : DeepSwapperInputs) -> ProcessorOutputs:
 	reference_vision_frame = inputs.get('reference_vision_frame')
-	target_vision_frame = inputs.get('target_vision_frame')
+	source_vision_frames = inputs.get('source_vision_frames')
+	target_vision_frames = inputs.get('target_vision_frames')
 	temp_vision_frame = inputs.get('temp_vision_frame')
 	temp_vision_mask = inputs.get('temp_vision_mask')
-	target_faces = select_faces(reference_vision_frame, target_vision_frame)
+
+	target_vision_frame = get_middle(target_vision_frames)
+	target_faces = select_faces(reference_vision_frame, source_vision_frames, target_vision_frames)

 	if target_faces:
 		for target_face in target_faces:
@@ -1,4 +1,4 @@
-from typing import Any, TypeAlias, TypedDict
+from typing import Any, List, TypeAlias, TypedDict

 from numpy.typing import NDArray

@@ -7,7 +7,8 @@ from facefusion.types import Mask, VisionFrame
 DeepSwapperInputs = TypedDict('DeepSwapperInputs',
 {
 	'reference_vision_frame' : VisionFrame,
-	'target_vision_frame' : VisionFrame,
+	'source_vision_frames' : List[VisionFrame],
+	'target_vision_frames' : List[VisionFrame],
 	'temp_vision_frame' : VisionFrame,
 	'temp_vision_mask' : Mask
 })
@@ -1,6 +1,7 @@
 from argparse import ArgumentParser
 from functools import lru_cache
-from typing import Tuple
+from types import ModuleType
+from typing import List, Tuple

 import cv2
 import numpy
@@ -8,9 +9,9 @@ import numpy
 import facefusion.capability_store
 import facefusion.jobs.job_manager
 from facefusion import config, content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer, inference_manager, logger, state_manager, translator, video_manager
-from facefusion.common_helper import create_int_metavar
+from facefusion.common_helper import create_int_metavar, get_middle
 from facefusion.download import conditional_download_hashes, conditional_download_sources, resolve_download_url
-from facefusion.face_analyser import scale_face
+from facefusion.face_creator import scale_face
 from facefusion.face_helper import paste_back, warp_face_by_face_landmark_5
 from facefusion.face_masker import create_box_mask, create_occlusion_mask
 from facefusion.face_selector import select_faces
@@ -22,7 +23,7 @@ from facefusion.processors.types import ApplyStateItem, LivePortraitExpression,
 from facefusion.program_helper import find_argument_group
 from facefusion.thread_helper import conditional_thread_semaphore, thread_semaphore
 from facefusion.types import Args, DownloadScope, Face, InferencePool, ModelOptions, ModelSet, ProcessMode, VisionFrame
-from facefusion.vision import read_static_image, read_static_video_frame
+from facefusion.vision import read_static_image, read_static_video_chunk, read_static_video_frame


@lru_cache()
@@ -134,10 +135,18 @@ def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
 	apply_state_item('expression_restorer_areas', args.get('expression_restorer_areas'))


+def get_common_modules() -> List[ModuleType]:
+	return [ content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer ]
+
+
 def pre_check() -> bool:
 	model_hash_set = get_model_options().get('hashes')
 	model_source_set = get_model_options().get('sources')

+	for common_module in get_common_modules():
+		if not common_module.pre_check():
+			return False
+
 	return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)


@@ -158,16 +167,15 @@ def pre_process(mode : ProcessMode) -> bool:
 def post_process() -> None:
 	read_static_image.cache_clear()
 	read_static_video_frame.cache_clear()
+	read_static_video_chunk.cache_clear()
 	video_manager.clear_video_pool()
+
 	if state_manager.get_item('video_memory_strategy') in [ 'strict', 'moderate' ]:
 		clear_inference_pool()
+
 	if state_manager.get_item('video_memory_strategy') == 'strict':
-		content_analyser.clear_inference_pool()
-		face_classifier.clear_inference_pool()
-		face_detector.clear_inference_pool()
-		face_landmarker.clear_inference_pool()
-		face_masker.clear_inference_pool()
-		face_recognizer.clear_inference_pool()
+		for common_module in get_common_modules():
+			common_module.clear_inference_pool()


 def restore_expression(target_face : Face, target_vision_frame : VisionFrame, temp_vision_frame : VisionFrame) -> VisionFrame:
@@ -278,10 +286,13 @@ def normalize_crop_frame(crop_vision_frame : VisionFrame) -> VisionFrame:

 def process_frame(inputs : ExpressionRestorerInputs) -> ProcessorOutputs:
 	reference_vision_frame = inputs.get('reference_vision_frame')
-	target_vision_frame = inputs.get('target_vision_frame')
+	source_vision_frames = inputs.get('source_vision_frames')
+	target_vision_frames = inputs.get('target_vision_frames')
 	temp_vision_frame = inputs.get('temp_vision_frame')
 	temp_vision_mask = inputs.get('temp_vision_mask')
-	target_faces = select_faces(reference_vision_frame, target_vision_frame)
+
+	target_vision_frame = get_middle(target_vision_frames)
+	target_faces = select_faces(reference_vision_frame, source_vision_frames, target_vision_frames)

 	if target_faces:
 		for target_face in target_faces:
@@ -6,7 +6,7 @@ ExpressionRestorerInputs = TypedDict('ExpressionRestorerInputs',
 {
 	'reference_vision_frame' : VisionFrame,
 	'source_vision_frames' : List[VisionFrame],
-	'target_vision_frame' : VisionFrame,
+	'target_vision_frames' : List[VisionFrame],
 	'temp_vision_frame' : VisionFrame,
 	'temp_vision_mask' : Mask
 })
@@ -1,4 +1,6 @@
 from argparse import ArgumentParser
+from types import ModuleType
+from typing import List

 import cv2
 import numpy
@@ -6,7 +8,8 @@ import numpy
 import facefusion.capability_store
 import facefusion.jobs.job_manager
 from facefusion import config, content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer, logger, state_manager, translator, video_manager
-from facefusion.face_analyser import scale_face
+from facefusion.common_helper import get_middle
+from facefusion.face_creator import scale_face
 from facefusion.face_helper import warp_face_by_face_landmark_5
 from facefusion.face_masker import create_area_mask, create_box_mask, create_occlusion_mask, create_region_mask
 from facefusion.face_selector import select_faces
@@ -16,7 +19,7 @@ from facefusion.processors.modules.face_debugger.types import FaceDebuggerInputs
 from facefusion.processors.types import ApplyStateItem, ProcessorOutputs
 from facefusion.program_helper import find_argument_group
 from facefusion.types import Args, Face, InferencePool, ProcessMode, VisionFrame
-from facefusion.vision import read_static_image, read_static_video_frame
+from facefusion.vision import read_static_image, read_static_video_chunk, read_static_video_frame


 def get_inference_pool() -> InferencePool:
@@ -49,7 +52,14 @@ def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
 	apply_state_item('face_debugger_items', args.get('face_debugger_items'))


+def get_common_modules() -> List[ModuleType]:
+	return [ content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer ]
+
+
 def pre_check() -> bool:
+	for common_module in get_common_modules():
+		if not common_module.pre_check():
+			return False
 	return True


@@ -67,14 +77,12 @@ def pre_process(mode : ProcessMode) -> bool:
 def post_process() -> None:
 	read_static_image.cache_clear()
 	read_static_video_frame.cache_clear()
+	read_static_video_chunk.cache_clear()
 	video_manager.clear_video_pool()
+
 	if state_manager.get_item('video_memory_strategy') == 'strict':
-		content_analyser.clear_inference_pool()
-		face_classifier.clear_inference_pool()
-		face_detector.clear_inference_pool()
-		face_landmarker.clear_inference_pool()
-		face_masker.clear_inference_pool()
-		face_recognizer.clear_inference_pool()
+		for common_module in get_common_modules():
+			common_module.clear_inference_pool()


 def debug_face(target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
@@ -103,21 +111,22 @@ def debug_face(target_face : Face, temp_vision_frame : VisionFrame) -> VisionFra

 def draw_bounding_box(target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
 	temp_vision_frame = numpy.ascontiguousarray(temp_vision_frame)
-	box_color = 0, 0, 255
-	border_color = 100, 100, 255
 	bounding_box = target_face.bounding_box.astype(numpy.int32)
 	x1, y1, x2, y2 = bounding_box
+	box_color = 0, 0, 255
+	border_scale = calculate_scale(temp_vision_frame)
+	border_color = 100, 100, 255

-	cv2.rectangle(temp_vision_frame, (x1, y1), (x2, y2), box_color, 2)
+	cv2.rectangle(temp_vision_frame, (x1, y1), (x2, y2), box_color, border_scale)

 	if target_face.angle == 0:
-		cv2.line(temp_vision_frame, (x1, y1), (x2, y1), border_color, 3)
+		cv2.line(temp_vision_frame, (x1, y1), (x2, y1), border_color, border_scale + 1)
 	if target_face.angle == 180:
-		cv2.line(temp_vision_frame, (x1, y2), (x2, y2), border_color, 3)
+		cv2.line(temp_vision_frame, (x1, y2), (x2, y2), border_color, border_scale + 1)
 	if target_face.angle == 90:
-		cv2.line(temp_vision_frame, (x2, y1), (x2, y2), border_color, 3)
+		cv2.line(temp_vision_frame, (x2, y1), (x2, y2), border_color, border_scale + 1)
 	if target_face.angle == 270:
-		cv2.line(temp_vision_frame, (x1, y1), (x1, y2), border_color, 3)
+		cv2.line(temp_vision_frame, (x1, y1), (x1, y2), border_color, border_scale + 1)

 	return temp_vision_frame

@@ -131,11 +140,15 @@ def draw_face_mask(target_face : Face, temp_vision_frame : VisionFrame) -> Visio
 	crop_vision_frame, affine_matrix = warp_face_by_face_landmark_5(temp_vision_frame, face_landmark_5_68, 'arcface_128', (512, 512))
 	inverse_matrix = cv2.invertAffineTransform(affine_matrix)
 	temp_size = temp_vision_frame.shape[:2][::-1]
+	mask_scale = calculate_scale(temp_vision_frame)
 	mask_color = 0, 255, 0

 	if numpy.array_equal(face_landmark_5, face_landmark_5_68):
 		mask_color = 255, 255, 0

+	if target_face.origin == 'refill':
+		mask_color = 0, 165, 255
+
 	if 'box' in state_manager.get_item('face_mask_types'):
 		box_mask = create_box_mask(crop_vision_frame, 0, state_manager.get_item('face_mask_padding'))
 		crop_masks.append(box_mask)
@@ -158,7 +171,7 @@ def draw_face_mask(target_face : Face, temp_vision_frame : VisionFrame) -> Visio
 	inverse_vision_frame = cv2.warpAffine(crop_mask, inverse_matrix, temp_size)
 	inverse_vision_frame = cv2.threshold(inverse_vision_frame, 100, 255, cv2.THRESH_BINARY)[1]
 	inverse_contours, _ = cv2.findContours(inverse_vision_frame, cv2.RETR_LIST, cv2.CHAIN_APPROX_NONE)
-	cv2.drawContours(temp_vision_frame, inverse_contours, -1, mask_color, 2)
+	cv2.drawContours(temp_vision_frame, inverse_contours, -1, mask_color, mask_scale)

 	return temp_vision_frame

@@ -166,13 +179,17 @@ def draw_face_mask(target_face : Face, temp_vision_frame : VisionFrame) -> Visio
 def draw_face_landmark_5(target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
 	temp_vision_frame = numpy.ascontiguousarray(temp_vision_frame)
 	face_landmark_5 = target_face.landmark_set.get('5')
+	point_scale = calculate_scale(temp_vision_frame)
 	point_color = 0, 0, 255

+	if target_face.origin == 'refill':
+		point_color = 0, 165, 255
+
 	if numpy.any(face_landmark_5):
 		face_landmark_5 = face_landmark_5.astype(numpy.int32)

 		for point in face_landmark_5:
-			cv2.circle(temp_vision_frame, tuple(point), 3, point_color, -1)
+			cv2.circle(temp_vision_frame, tuple(point), point_scale, point_color, -1)

 	return temp_vision_frame

@@ -181,16 +198,20 @@ def draw_face_landmark_5_68(target_face : Face, temp_vision_frame : VisionFrame)
 	temp_vision_frame = numpy.ascontiguousarray(temp_vision_frame)
 	face_landmark_5 = target_face.landmark_set.get('5')
 	face_landmark_5_68 = target_face.landmark_set.get('5/68')
+	point_scale = calculate_scale(temp_vision_frame)
 	point_color = 0, 255, 0

 	if numpy.array_equal(face_landmark_5, face_landmark_5_68):
 		point_color = 255, 255, 0

+	if target_face.origin == 'refill':
+		point_color = 0, 165, 255
+
 	if numpy.any(face_landmark_5_68):
 		face_landmark_5_68 = face_landmark_5_68.astype(numpy.int32)

 		for point in face_landmark_5_68:
-			cv2.circle(temp_vision_frame, tuple(point), 3, point_color, -1)
+			cv2.circle(temp_vision_frame, tuple(point), point_scale, point_color, -1)

 	return temp_vision_frame

@@ -199,16 +220,20 @@ def draw_face_landmark_68(target_face : Face, temp_vision_frame : VisionFrame) -
 	temp_vision_frame = numpy.ascontiguousarray(temp_vision_frame)
 	face_landmark_68 = target_face.landmark_set.get('68')
 	face_landmark_68_5 = target_face.landmark_set.get('68/5')
+	point_scale = calculate_scale(temp_vision_frame)
 	point_color = 0, 255, 0

 	if numpy.array_equal(face_landmark_68, face_landmark_68_5):
 		point_color = 255, 255, 0

+	if target_face.origin == 'refill':
+		point_color = 0, 165, 255
+
 	if numpy.any(face_landmark_68):
 		face_landmark_68 = face_landmark_68.astype(numpy.int32)

 		for point in face_landmark_68:
-			cv2.circle(temp_vision_frame, tuple(point), 3, point_color, -1)
+			cv2.circle(temp_vision_frame, tuple(point), point_scale, point_color, -1)

 	return temp_vision_frame

@@ -216,23 +241,36 @@ def draw_face_landmark_68(target_face : Face, temp_vision_frame : VisionFrame) -
 def draw_face_landmark_68_5(target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
 	temp_vision_frame = numpy.ascontiguousarray(temp_vision_frame)
 	face_landmark_68_5 = target_face.landmark_set.get('68/5')
+	point_scale = calculate_scale(temp_vision_frame)
 	point_color = 255, 255, 0

+	if target_face.origin == 'refill':
+		point_color = 0, 165, 255
+
 	if numpy.any(face_landmark_68_5):
 		face_landmark_68_5 = face_landmark_68_5.astype(numpy.int32)

 		for point in face_landmark_68_5:
-			cv2.circle(temp_vision_frame, tuple(point), 3, point_color, -1)
+			cv2.circle(temp_vision_frame, tuple(point), point_scale, point_color, -1)

 	return temp_vision_frame


+def calculate_scale(temp_vision_frame : VisionFrame) -> int:
+	frame_height, _ = temp_vision_frame.shape[:2]
+	frame_scale = round(frame_height / 270)
+	return max(1, min(10, frame_scale))
+
+
 def process_frame(inputs : FaceDebuggerInputs) -> ProcessorOutputs:
 	reference_vision_frame = inputs.get('reference_vision_frame')
-	target_vision_frame = inputs.get('target_vision_frame')
+	source_vision_frames = inputs.get('source_vision_frames')
+	target_vision_frames = inputs.get('target_vision_frames')
 	temp_vision_frame = inputs.get('temp_vision_frame')
 	temp_vision_mask = inputs.get('temp_vision_mask')
-	target_faces = select_faces(reference_vision_frame, target_vision_frame)
+
+	target_vision_frame = get_middle(target_vision_frames)
+	target_faces = select_faces(reference_vision_frame, source_vision_frames, target_vision_frames)

 	if target_faces:
 		for target_face in target_faces:
@@ -240,5 +278,3 @@ def process_frame(inputs : FaceDebuggerInputs) -> ProcessorOutputs:
 			temp_vision_frame = debug_face(target_face, temp_vision_frame)

 	return temp_vision_frame, temp_vision_mask
-
-
@@ -1,11 +1,12 @@
-from typing import Literal, TypedDict
+from typing import List, Literal, TypedDict

 from facefusion.types import Mask, VisionFrame

 FaceDebuggerInputs = TypedDict('FaceDebuggerInputs',
 {
 	'reference_vision_frame' : VisionFrame,
-	'target_vision_frame' : VisionFrame,
+	'source_vision_frames' : List[VisionFrame],
+	'target_vision_frames' : List[VisionFrame],
 	'temp_vision_frame' : VisionFrame,
 	'temp_vision_mask' : Mask
 })
@@ -1,6 +1,7 @@
 from argparse import ArgumentParser
 from functools import lru_cache
-from typing import Tuple
+from types import ModuleType
+from typing import List, Tuple

 import cv2
 import numpy
@@ -8,9 +9,9 @@ import numpy
 import facefusion.capability_store
 import facefusion.jobs.job_manager
 from facefusion import config, content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer, inference_manager, logger, state_manager, translator, video_manager
-from facefusion.common_helper import create_float_metavar
+from facefusion.common_helper import create_float_metavar, get_middle
 from facefusion.download import conditional_download_hashes, conditional_download_sources, resolve_download_url
-from facefusion.face_analyser import scale_face
+from facefusion.face_creator import scale_face
 from facefusion.face_helper import paste_back, scale_face_landmark_5, warp_face_by_face_landmark_5
 from facefusion.face_masker import create_box_mask
 from facefusion.face_selector import select_faces
@@ -22,7 +23,7 @@ from facefusion.processors.types import ApplyStateItem, LivePortraitExpression,
 from facefusion.program_helper import find_argument_group
 from facefusion.thread_helper import conditional_thread_semaphore, thread_semaphore
 from facefusion.types import Args, DownloadScope, Face, FaceLandmark68, InferencePool, ModelOptions, ModelSet, ProcessMode, VisionFrame
-from facefusion.vision import read_static_image, read_static_video_frame
+from facefusion.vision import read_static_image, read_static_video_chunk, read_static_video_frame


@lru_cache()
@@ -272,10 +273,18 @@ def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
 	apply_state_item('face_editor_head_roll', args.get('face_editor_head_roll'))


+def get_common_modules() -> List[ModuleType]:
+	return [ content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer ]
+
+
 def pre_check() -> bool:
 	model_hash_set = get_model_options().get('hashes')
 	model_source_set = get_model_options().get('sources')

+	for common_module in get_common_modules():
+		if not common_module.pre_check():
+			return False
+
 	return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)


@@ -293,16 +302,15 @@ def pre_process(mode : ProcessMode) -> bool:
 def post_process() -> None:
 	read_static_image.cache_clear()
 	read_static_video_frame.cache_clear()
+	read_static_video_chunk.cache_clear()
 	video_manager.clear_video_pool()
+
 	if state_manager.get_item('video_memory_strategy') in [ 'strict', 'moderate' ]:
 		clear_inference_pool()
+
 	if state_manager.get_item('video_memory_strategy') == 'strict':
-		content_analyser.clear_inference_pool()
-		face_classifier.clear_inference_pool()
-		face_detector.clear_inference_pool()
-		face_landmarker.clear_inference_pool()
-		face_masker.clear_inference_pool()
-		face_recognizer.clear_inference_pool()
+		for common_module in get_common_modules():
+			common_module.clear_inference_pool()


 def edit_face(target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
@@ -591,10 +599,13 @@ def normalize_crop_frame(crop_vision_frame : VisionFrame) -> VisionFrame:

 def process_frame(inputs : FaceEditorInputs) -> ProcessorOutputs:
 	reference_vision_frame = inputs.get('reference_vision_frame')
-	target_vision_frame = inputs.get('target_vision_frame')
+	source_vision_frames = inputs.get('source_vision_frames')
+	target_vision_frames = inputs.get('target_vision_frames')
 	temp_vision_frame = inputs.get('temp_vision_frame')
 	temp_vision_mask = inputs.get('temp_vision_mask')
-	target_faces = select_faces(reference_vision_frame, target_vision_frame)
+
+	target_vision_frame = get_middle(target_vision_frames)
+	target_faces = select_faces(reference_vision_frame, source_vision_frames, target_vision_frames)

 	if target_faces:
 		for target_face in target_faces:
@@ -1,11 +1,12 @@
-from typing import Literal, TypedDict
+from typing import List, Literal, TypedDict

 from facefusion.types import Mask, VisionFrame

 FaceEditorInputs = TypedDict('FaceEditorInputs',
 {
 	'reference_vision_frame' : VisionFrame,
-	'target_vision_frame' : VisionFrame,
+	'source_vision_frames' : List[VisionFrame],
+	'target_vision_frames' : List[VisionFrame],
 	'temp_vision_frame' : VisionFrame,
 	'temp_vision_mask' : Mask
 })
@@ -1,14 +1,16 @@
 from argparse import ArgumentParser
 from functools import lru_cache
+from types import ModuleType
+from typing import List

 import numpy

 import facefusion.capability_store
 import facefusion.jobs.job_manager
 from facefusion import config, content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer, inference_manager, logger, state_manager, translator, video_manager
-from facefusion.common_helper import create_float_metavar, create_int_metavar
+from facefusion.common_helper import create_float_metavar, create_int_metavar, get_middle
 from facefusion.download import conditional_download_hashes, conditional_download_sources, resolve_download_url
-from facefusion.face_analyser import scale_face
+from facefusion.face_creator import scale_face
 from facefusion.face_helper import paste_back, warp_face_by_face_landmark_5
 from facefusion.face_masker import create_box_mask, create_occlusion_mask
 from facefusion.face_selector import select_faces
@@ -19,7 +21,7 @@ from facefusion.processors.types import ApplyStateItem, ProcessorOutputs
 from facefusion.program_helper import find_argument_group
 from facefusion.thread_helper import thread_semaphore
 from facefusion.types import Args, DownloadScope, Face, InferencePool, ModelOptions, ModelSet, ProcessMode, VisionFrame
-from facefusion.vision import blend_frame, read_static_image, read_static_video_frame
+from facefusion.vision import blend_frame, read_static_image, read_static_video_chunk, read_static_video_frame


@lru_cache()
@@ -327,10 +329,18 @@ def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
 	apply_state_item('face_enhancer_weight', args.get('face_enhancer_weight'))


+def get_common_modules() -> List[ModuleType]:
+	return [ content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer ]
+
+
 def pre_check() -> bool:
 	model_hash_set = get_model_options().get('hashes')
 	model_source_set = get_model_options().get('sources')

+	for common_module in get_common_modules():
+		if not common_module.pre_check():
+			return False
+
 	return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)


@@ -348,16 +358,15 @@ def pre_process(mode : ProcessMode) -> bool:
 def post_process() -> None:
 	read_static_image.cache_clear()
 	read_static_video_frame.cache_clear()
+	read_static_video_chunk.cache_clear()
 	video_manager.clear_video_pool()
+
 	if state_manager.get_item('video_memory_strategy') in [ 'strict', 'moderate' ]:
 		clear_inference_pool()
+
 	if state_manager.get_item('video_memory_strategy') == 'strict':
-		content_analyser.clear_inference_pool()
-		face_classifier.clear_inference_pool()
-		face_detector.clear_inference_pool()
-		face_landmarker.clear_inference_pool()
-		face_masker.clear_inference_pool()
-		face_recognizer.clear_inference_pool()
+		for common_module in get_common_modules():
+			common_module.clear_inference_pool()


 def enhance_face(target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
@@ -434,10 +443,13 @@ def blend_paste_frame(temp_vision_frame : VisionFrame, paste_vision_frame : Visi

 def process_frame(inputs : FaceEnhancerInputs) -> ProcessorOutputs:
 	reference_vision_frame = inputs.get('reference_vision_frame')
-	target_vision_frame = inputs.get('target_vision_frame')
+	source_vision_frames = inputs.get('source_vision_frames')
+	target_vision_frames = inputs.get('target_vision_frames')
 	temp_vision_frame = inputs.get('temp_vision_frame')
 	temp_vision_mask = inputs.get('temp_vision_mask')
-	target_faces = select_faces(reference_vision_frame, target_vision_frame)
+
+	target_vision_frame = get_middle(target_vision_frames)
+	target_faces = select_faces(reference_vision_frame, source_vision_frames, target_vision_frames)

 	if target_faces:
 		for target_face in target_faces:
@@ -1,4 +1,4 @@
-from typing import Any, Literal, TypeAlias, TypedDict
+from typing import Any, List, Literal, TypeAlias, TypedDict

 from numpy.typing import NDArray

@@ -7,7 +7,8 @@ from facefusion.types import Mask, VisionFrame
 FaceEnhancerInputs = TypedDict('FaceEnhancerInputs',
 {
 	'reference_vision_frame' : VisionFrame,
-	'target_vision_frame' : VisionFrame,
+	'source_vision_frames' : List[VisionFrame],
+	'target_vision_frames' : List[VisionFrame],
 	'temp_vision_frame' : VisionFrame,
 	'temp_vision_mask' : Mask
 })
@@ -1,5 +1,6 @@
 from argparse import ArgumentParser
 from functools import lru_cache
+from types import ModuleType
 from typing import List, Optional, Tuple

 import cv2
@@ -9,10 +10,10 @@ import facefusion.capability_store
 import facefusion.choices
 import facefusion.jobs.job_manager
 from facefusion import config, content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer, inference_manager, logger, state_manager, translator, video_manager
-from facefusion.common_helper import get_first, is_macos
+from facefusion.common_helper import get_first, get_middle, is_macos
 from facefusion.download import conditional_download_hashes, conditional_download_sources, resolve_download_url
 from facefusion.execution import has_execution_provider
-from facefusion.face_analyser import get_average_face, get_one_face, get_static_faces, scale_face
+from facefusion.face_creator import average_face_identity, get_one_face, get_static_faces, scale_face
 from facefusion.face_helper import paste_back, warp_face_by_face_landmark_5
 from facefusion.face_masker import create_area_mask, create_box_mask, create_occlusion_mask, create_region_mask
 from facefusion.face_selector import select_faces, sort_faces_by_order
@@ -24,8 +25,8 @@ from facefusion.processors.pixel_boost import explode_pixel_boost, implode_pixel
 from facefusion.processors.types import ApplyStateItem, ProcessorOutputs
 from facefusion.program_helper import find_argument_group
 from facefusion.thread_helper import conditional_thread_semaphore
-from facefusion.types import Args, DownloadScope, Embedding, Face, InferencePool, ModelOptions, ModelSet, ProcessMode, VisionFrame
-from facefusion.vision import read_static_image, read_static_images, read_static_video_frame, unpack_resolution
+from facefusion.types import Args, DownloadScope, Embedding, Face, InferencePool, InferenceProvider, ModelOptions, ModelSet, ProcessMode, VisionFrame
+from facefusion.vision import read_static_image, read_static_images, read_static_video_chunk, read_static_video_frame, unpack_resolution


@lru_cache()
@@ -246,6 +247,7 @@ def create_static_model_set(download_scope : DownloadScope) -> ModelSet:
 					'path': resolve_relative_path('../.assets/models/hyperswap_1a_256.onnx')
 				}
 			},
+			'precision': 'fp16',
 			'type': 'hyperswap',
 			'template': 'arcface_128',
 			'size': (256, 256),
@@ -276,6 +278,7 @@ def create_static_model_set(download_scope : DownloadScope) -> ModelSet:
 						'path': resolve_relative_path('../.assets/models/hyperswap_1b_256.onnx')
 					}
 				},
+			'precision': 'fp16',
 			'type': 'hyperswap',
 			'template': 'arcface_128',
 			'size': (256, 256),
@@ -306,6 +309,7 @@ def create_static_model_set(download_scope : DownloadScope) -> ModelSet:
 					'path': resolve_relative_path('../.assets/models/hyperswap_1c_256.onnx')
 				}
 			},
+			'precision': 'fp16',
 			'type': 'hyperswap',
 			'template': 'arcface_128',
 			'size': (256, 256),
@@ -366,6 +370,7 @@ def create_static_model_set(download_scope : DownloadScope) -> ModelSet:
 					'path': resolve_relative_path('../.assets/models/inswapper_128_fp16.onnx')
 				}
 			},
+			'precision': 'fp16',
 			'type': 'inswapper',
 			'template': 'arcface_128',
 			'size': (128, 128),
@@ -486,28 +491,38 @@ def create_static_model_set(download_scope : DownloadScope) -> ModelSet:


 def get_inference_pool() -> InferencePool:
-	model_names = [ get_model_name() ]
+	model_names = [ state_manager.get_item('face_swapper_model') ]
 	model_source_set = get_model_options().get('sources')

 	return inference_manager.get_inference_pool(__name__, model_names, model_source_set)


 def clear_inference_pool() -> None:
-	model_names = [ get_model_name() ]
+	model_names = [ state_manager.get_item('face_swapper_model') ]
 	inference_manager.clear_inference_pool(__name__, model_names)


+def resolve_inference_providers() -> List[InferenceProvider]:
+	model_precision = get_model_options().get('precision')
+	model_type = get_model_options().get('type')
+
+	if is_macos() and has_execution_provider('coreml'):
+		if model_type in [ 'ghost', 'uniface' ] or model_precision == 'fp16':
+			return\
+			[
+				(facefusion.choices.execution_provider_set.get('coreml'),
+				{
+					'ModelFormat': 'MLProgram',
+					'SpecializationStrategy': 'FastPrediction'
+				})
+			]
+
+	return []
+
+
 def get_model_options() -> ModelOptions:
-	model_name = get_model_name()
-	return create_static_model_set('full').get(model_name)
-
-
-def get_model_name() -> str:
 	model_name = state_manager.get_item('face_swapper_model')
-
-	if is_macos() and has_execution_provider('coreml') and model_name == 'inswapper_128_fp16':
-		return 'inswapper_128'
-	return model_name
+	return create_static_model_set('full').get(model_name)


 def register_args(program : ArgumentParser) -> None:
@@ -552,10 +567,18 @@ def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
 	apply_state_item('face_swapper_weight', args.get('face_swapper_weight'))


+def get_common_modules() -> List[ModuleType]:
+	return [ content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer ]
+
+
 def pre_check() -> bool:
 	model_hash_set = get_model_options().get('hashes')
 	model_source_set = get_model_options().get('sources')

+	for common_module in get_common_modules():
+		if not common_module.pre_check():
+			return False
+
 	return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)


@@ -587,20 +610,19 @@ def pre_process(mode : ProcessMode) -> bool:
 def post_process() -> None:
 	read_static_image.cache_clear()
 	read_static_video_frame.cache_clear()
+	read_static_video_chunk.cache_clear()
 	video_manager.clear_video_pool()
+
 	if state_manager.get_item('video_memory_strategy') in [ 'strict', 'moderate' ]:
 		get_static_model_initializer.cache_clear()
 		clear_inference_pool()
+
 	if state_manager.get_item('video_memory_strategy') == 'strict':
-		content_analyser.clear_inference_pool()
-		face_classifier.clear_inference_pool()
-		face_detector.clear_inference_pool()
-		face_landmarker.clear_inference_pool()
-		face_masker.clear_inference_pool()
-		face_recognizer.clear_inference_pool()
+		for common_module in get_common_modules():
+			common_module.clear_inference_pool()


-def swap_face(source_face : Face, target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
+def swap_face(source_face : Face, target_face : Face, source_vision_frame : VisionFrame, temp_vision_frame : VisionFrame) -> VisionFrame:
 	model_template = get_model_options().get('template')
 	model_size = get_model_options().get('size')
 	pixel_boost_size = unpack_resolution(state_manager.get_item('face_swapper_pixel_boost'))
@@ -620,7 +642,7 @@ def swap_face(source_face : Face, target_face : Face, temp_vision_frame : Vision
 	pixel_boost_vision_frames = implode_pixel_boost(crop_vision_frame, pixel_boost_total, model_size)
 	for pixel_boost_vision_frame in pixel_boost_vision_frames:
 		pixel_boost_vision_frame = prepare_crop_frame(pixel_boost_vision_frame)
-		pixel_boost_vision_frame = forward_swap_face(source_face, target_face, pixel_boost_vision_frame)
+		pixel_boost_vision_frame = forward_swap_face(source_face, target_face, source_vision_frame, pixel_boost_vision_frame)
 		pixel_boost_vision_frame = normalize_crop_frame(pixel_boost_vision_frame)
 		temp_vision_frames.append(pixel_boost_vision_frame)
 	crop_vision_frame = explode_pixel_boost(temp_vision_frames, pixel_boost_total, model_size, pixel_boost_size)
@@ -639,18 +661,15 @@ def swap_face(source_face : Face, target_face : Face, temp_vision_frame : Vision
 	return paste_vision_frame


-def forward_swap_face(source_face : Face, target_face : Face, crop_vision_frame : VisionFrame) -> VisionFrame:
+def forward_swap_face(source_face : Face, target_face : Face, source_vision_frame : VisionFrame, crop_vision_frame : VisionFrame) -> VisionFrame:
 	face_swapper = get_inference_pool().get('face_swapper')
 	model_type = get_model_options().get('type')
 	face_swapper_inputs = {}

-	if is_macos() and has_execution_provider('coreml') and model_type in [ 'ghost', 'uniface' ]:
-		face_swapper.set_providers([ facefusion.choices.execution_provider_set.get('cpu') ])
-
 	for face_swapper_input in face_swapper.get_inputs():
 		if face_swapper_input.name == 'source':
 			if model_type in [ 'blendswap', 'uniface' ]:
-				face_swapper_inputs[face_swapper_input.name] = prepare_source_frame(source_face)
+				face_swapper_inputs[face_swapper_input.name] = prepare_source_frame(source_face, source_vision_frame)
 			else:
 				source_embedding = prepare_source_embedding(source_face)
 				source_embedding = balance_source_embedding(source_embedding, target_face.embedding)
@@ -676,9 +695,8 @@ def forward_convert_embedding(face_embedding : Embedding) -> Embedding:
 	return face_embedding


-def prepare_source_frame(source_face : Face) -> VisionFrame:
+def prepare_source_frame(source_face : Face, source_vision_frame : VisionFrame) -> VisionFrame:
 	model_type = get_model_options().get('type')
-	source_vision_frame = read_static_image(get_first(state_manager.get_item('source_paths')))

 	if model_type == 'blendswap':
 		source_vision_frame, _ = warp_face_by_face_landmark_5(source_vision_frame, source_face.landmark_set.get('5/68'), 'arcface_112_v2', (112, 112))
@@ -776,21 +794,25 @@ def extract_source_face(source_vision_frames : List[VisionFrame]) -> Optional[Fa
 			if temp_faces:
 				source_faces.append(get_first(temp_faces))

-	return get_average_face(source_faces)
+	return average_face_identity(source_faces)


 def process_frame(inputs : FaceSwapperInputs) -> ProcessorOutputs:
 	reference_vision_frame = inputs.get('reference_vision_frame')
 	source_vision_frames = inputs.get('source_vision_frames')
-	target_vision_frame = inputs.get('target_vision_frame')
+	target_vision_frames = inputs.get('target_vision_frames')
 	temp_vision_frame = inputs.get('temp_vision_frame')
 	temp_vision_mask = inputs.get('temp_vision_mask')
+
+	target_vision_frame = get_middle(target_vision_frames)
 	source_face = extract_source_face(source_vision_frames)
-	target_faces = select_faces(reference_vision_frame, target_vision_frame)
+	target_faces = select_faces(reference_vision_frame, source_vision_frames, target_vision_frames)

 	if source_face and target_faces:
+		source_vision_frame = get_first(source_vision_frames)
+
 		for target_face in target_faces:
 			target_face = scale_face(target_face, target_vision_frame, temp_vision_frame)
-			temp_vision_frame = swap_face(source_face, target_face, temp_vision_frame)
+			temp_vision_frame = swap_face(source_face, target_face, source_vision_frame, temp_vision_frame)

 	return temp_vision_frame, temp_vision_mask
@@ -6,7 +6,7 @@ FaceSwapperInputs = TypedDict('FaceSwapperInputs',
 {
 	'reference_vision_frame' : VisionFrame,
 	'source_vision_frames' : List[VisionFrame],
-	'target_vision_frame' : VisionFrame,
+	'target_vision_frames' : List[VisionFrame],
 	'temp_vision_frame' : VisionFrame,
 	'temp_vision_mask' : Mask
 })
@@ -1,11 +1,13 @@
 from argparse import ArgumentParser
 from functools import lru_cache
+from types import ModuleType
 from typing import List

 import cv2
 import numpy

 import facefusion.capability_store
+import facefusion.choices
 import facefusion.jobs.job_manager
 from facefusion import config, content_analyser, inference_manager, logger, state_manager, translator, video_manager
 from facefusion.common_helper import create_int_metavar, is_macos
@@ -17,8 +19,8 @@ from facefusion.processors.modules.frame_colorizer.types import FrameColorizerIn
 from facefusion.processors.types import ApplyStateItem, ProcessorOutputs
 from facefusion.program_helper import find_argument_group
 from facefusion.thread_helper import thread_semaphore
-from facefusion.types import Args, DownloadScope, ExecutionProvider, InferencePool, ModelOptions, ModelSet, ProcessMode, VisionFrame
-from facefusion.vision import blend_frame, read_static_image, read_static_video_frame, unpack_resolution
+from facefusion.types import Args, DownloadScope, InferencePool, InferenceProvider, ModelOptions, ModelSet, ProcessMode, VisionFrame
+from facefusion.vision import blend_frame, read_static_image, read_static_video_chunk, read_static_video_frame, unpack_resolution


@lru_cache()
@@ -170,10 +172,11 @@ def clear_inference_pool() -> None:
 	inference_manager.clear_inference_pool(__name__, model_names)


-def resolve_execution_providers() -> List[ExecutionProvider]:
+def resolve_inference_providers() -> List[InferenceProvider]:
 	if is_macos() and has_execution_provider('coreml'):
-		return [ 'cpu' ]
-	return state_manager.get_item('execution_providers')
+		return [ facefusion.choices.execution_provider_set.get('cpu') ]
+
+	return []


 def get_model_options() -> ModelOptions:
@@ -218,10 +221,18 @@ def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
 	apply_state_item('frame_colorizer_size', args.get('frame_colorizer_size'))


+def get_common_modules() -> List[ModuleType]:
+	return [ content_analyser ]
+
+
 def pre_check() -> bool:
 	model_hash_set = get_model_options().get('hashes')
 	model_source_set = get_model_options().get('sources')

+	for common_module in get_common_modules():
+		if not common_module.pre_check():
+			return False
+
 	return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)


@@ -239,11 +250,15 @@ def pre_process(mode : ProcessMode) -> bool:
 def post_process() -> None:
 	read_static_image.cache_clear()
 	read_static_video_frame.cache_clear()
+	read_static_video_chunk.cache_clear()
 	video_manager.clear_video_pool()
+
 	if state_manager.get_item('video_memory_strategy') in [ 'strict', 'moderate' ]:
 		clear_inference_pool()
+
 	if state_manager.get_item('video_memory_strategy') == 'strict':
-		content_analyser.clear_inference_pool()
+		for common_module in get_common_modules():
+			common_module.clear_inference_pool()


 def colorize_frame(temp_vision_frame : VisionFrame) -> VisionFrame:
@@ -1,10 +1,10 @@
-from typing import Literal, TypedDict
+from typing import List, Literal, TypedDict

 from facefusion.types import Mask, VisionFrame

 FrameColorizerInputs = TypedDict('FrameColorizerInputs',
 {
-	'target_vision_frame' : VisionFrame,
+	'target_vision_frames' : List[VisionFrame],
 	'temp_vision_frame' : VisionFrame,
 	'temp_vision_mask' : Mask
 })
@@ -1,10 +1,13 @@
 from argparse import ArgumentParser
 from functools import lru_cache
+from types import ModuleType
+from typing import List

 import cv2
 import numpy

 import facefusion.capability_store
+import facefusion.choices
 import facefusion.jobs.job_manager
 from facefusion import config, content_analyser, inference_manager, logger, state_manager, translator, video_manager
 from facefusion.common_helper import create_int_metavar, is_macos
@@ -16,8 +19,8 @@ from facefusion.processors.modules.frame_enhancer.types import FrameEnhancerInpu
 from facefusion.processors.types import ApplyStateItem, ProcessorOutputs
 from facefusion.program_helper import find_argument_group
 from facefusion.thread_helper import conditional_thread_semaphore
-from facefusion.types import Args, DownloadScope, InferencePool, ModelOptions, ModelSet, ProcessMode, VisionFrame
-from facefusion.vision import blend_frame, create_tile_frames, merge_tile_frames, read_static_image, read_static_video_frame
+from facefusion.types import Args, DownloadScope, InferencePool, InferenceProvider, ModelOptions, ModelSet, ProcessMode, VisionFrame
+from facefusion.vision import blend_frame, create_tile_frames, merge_tile_frames, read_static_image, read_static_video_chunk, read_static_video_frame


@lru_cache()
@@ -156,6 +159,7 @@ def create_static_model_set(download_scope : DownloadScope) -> ModelSet:
 					'path': resolve_relative_path('../.assets/models/real_esrgan_x2_fp16.onnx')
 				}
 			},
+			'precision': 'fp16',
 			'size': (256, 16, 8),
 			'scale': 2
 		},
@@ -210,6 +214,7 @@ def create_static_model_set(download_scope : DownloadScope) -> ModelSet:
 					'path': resolve_relative_path('../.assets/models/real_esrgan_x4_fp16.onnx')
 				}
 			},
+			'precision': 'fp16',
 			'size': (256, 16, 8),
 			'scale': 4
 		},
@@ -264,6 +269,7 @@ def create_static_model_set(download_scope : DownloadScope) -> ModelSet:
 					'path': resolve_relative_path('../.assets/models/real_esrgan_x8_fp16.onnx')
 				}
 			},
+			'precision': 'fp16',
 			'size': (256, 16, 8),
 			'scale': 8
 		},
@@ -541,35 +547,38 @@ def create_static_model_set(download_scope : DownloadScope) -> ModelSet:


 def get_inference_pool() -> InferencePool:
-	model_names = [ get_frame_enhancer_model() ]
+	model_names = [ state_manager.get_item('frame_enhancer_model') ]
 	model_source_set = get_model_options().get('sources')

 	return inference_manager.get_inference_pool(__name__, model_names, model_source_set)


 def clear_inference_pool() -> None:
-	model_names = [ get_frame_enhancer_model() ]
+	model_names = [ state_manager.get_item('frame_enhancer_model') ]
 	inference_manager.clear_inference_pool(__name__, model_names)


+def resolve_inference_providers() -> List[InferenceProvider]:
+	model_precision = get_model_options().get('precision')
+
+	if is_macos() and has_execution_provider('coreml') and model_precision == 'fp16':
+		return\
+		[
+			(facefusion.choices.execution_provider_set.get('coreml'),
+			{
+				'ModelFormat': 'MLProgram',
+				'SpecializationStrategy': 'FastPrediction'
+			})
+		]
+
+	return []
+
+
 def get_model_options() -> ModelOptions:
-	model_name = get_frame_enhancer_model()
+	model_name = state_manager.get_item('frame_enhancer_model')
 	return create_static_model_set('full').get(model_name)


-def get_frame_enhancer_model() -> str:
-	frame_enhancer_model = state_manager.get_item('frame_enhancer_model')
-
-	if is_macos() and has_execution_provider('coreml'):
-		if frame_enhancer_model == 'real_esrgan_x2_fp16':
-			return 'real_esrgan_x2'
-		if frame_enhancer_model == 'real_esrgan_x4_fp16':
-			return 'real_esrgan_x4'
-		if frame_enhancer_model == 'real_esrgan_x8_fp16':
-			return 'real_esrgan_x8'
-	return frame_enhancer_model
-
-
 def register_args(program : ArgumentParser) -> None:
 	group_processors = find_argument_group(program, 'processors')
 	if group_processors:
@@ -599,10 +608,18 @@ def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
 	apply_state_item('frame_enhancer_blend', args.get('frame_enhancer_blend'))


+def get_common_modules() -> List[ModuleType]:
+	return [ content_analyser ]
+
+
 def pre_check() -> bool:
 	model_hash_set = get_model_options().get('hashes')
 	model_source_set = get_model_options().get('sources')

+	for common_module in get_common_modules():
+		if not common_module.pre_check():
+			return False
+
 	return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)


@@ -620,11 +637,15 @@ def pre_process(mode : ProcessMode) -> bool:
 def post_process() -> None:
 	read_static_image.cache_clear()
 	read_static_video_frame.cache_clear()
+	read_static_video_chunk.cache_clear()
 	video_manager.clear_video_pool()
+
 	if state_manager.get_item('video_memory_strategy') in [ 'strict', 'moderate' ]:
 		clear_inference_pool()
+
 	if state_manager.get_item('video_memory_strategy') == 'strict':
-		content_analyser.clear_inference_pool()
+		for common_module in get_common_modules():
+			common_module.clear_inference_pool()


 def enhance_frame(temp_vision_frame : VisionFrame) -> VisionFrame:
@@ -1,10 +1,10 @@
-from typing import Literal, TypedDict
+from typing import List, Literal, TypedDict

 from facefusion.types import Mask, VisionFrame

 FrameEnhancerInputs = TypedDict('FrameEnhancerInputs',
 {
-	'target_vision_frame' : VisionFrame,
+	'target_vision_frames' : List[VisionFrame],
 	'temp_vision_frame' : VisionFrame,
 	'temp_vision_mask' : Mask
 })
@@ -1,5 +1,7 @@
 from argparse import ArgumentParser
 from functools import lru_cache
+from types import ModuleType
+from typing import List

 import cv2
 import numpy
@@ -8,9 +10,9 @@ import facefusion.capability_store
 import facefusion.jobs.job_manager
 from facefusion import config, content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer, inference_manager, logger, state_manager, translator, video_manager, voice_extractor
 from facefusion.audio import read_static_voice
-from facefusion.common_helper import create_float_metavar
+from facefusion.common_helper import create_float_metavar, get_middle
 from facefusion.download import conditional_download_hashes, conditional_download_sources, resolve_download_url
-from facefusion.face_analyser import scale_face
+from facefusion.face_creator import scale_face
 from facefusion.face_helper import create_bounding_box, paste_back, warp_face_by_bounding_box, warp_face_by_face_landmark_5
 from facefusion.face_masker import create_area_mask, create_box_mask, create_occlusion_mask
 from facefusion.face_selector import select_faces
@@ -21,7 +23,7 @@ from facefusion.processors.types import ApplyStateItem, ProcessorOutputs
 from facefusion.program_helper import find_argument_group
 from facefusion.thread_helper import conditional_thread_semaphore
 from facefusion.types import Args, AudioFrame, DownloadScope, Face, InferencePool, ModelOptions, ModelSet, ProcessMode, VisionFrame
-from facefusion.vision import read_static_image, read_static_video_frame
+from facefusion.vision import read_static_image, read_static_video_chunk, read_static_video_frame


@lru_cache()
@@ -158,10 +160,18 @@ def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
 	apply_state_item('lip_syncer_weight', args.get('lip_syncer_weight'))


+def get_common_modules() -> List[ModuleType]:
+	return [ content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer, voice_extractor ]
+
+
 def pre_check() -> bool:
 	model_hash_set = get_model_options().get('hashes')
 	model_source_set = get_model_options().get('sources')

+	for common_module in get_common_modules():
+		if not common_module.pre_check():
+			return False
+
 	return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)


@@ -175,18 +185,16 @@ def pre_process(mode : ProcessMode) -> bool:
 def post_process() -> None:
 	read_static_image.cache_clear()
 	read_static_video_frame.cache_clear()
+	read_static_video_chunk.cache_clear()
 	read_static_voice.cache_clear()
 	video_manager.clear_video_pool()
+
 	if state_manager.get_item('video_memory_strategy') in [ 'strict', 'moderate' ]:
 		clear_inference_pool()
+
 	if state_manager.get_item('video_memory_strategy') == 'strict':
-		content_analyser.clear_inference_pool()
-		face_classifier.clear_inference_pool()
-		face_detector.clear_inference_pool()
-		face_landmarker.clear_inference_pool()
-		face_masker.clear_inference_pool()
-		face_recognizer.clear_inference_pool()
-		voice_extractor.clear_inference_pool()
+		for common_module in get_common_modules():
+			common_module.clear_inference_pool()


 def sync_lip(target_face : Face, source_voice_frame : AudioFrame, temp_vision_frame : VisionFrame) -> VisionFrame:
@@ -298,11 +306,14 @@ def normalize_crop_frame(crop_vision_frame : VisionFrame) -> VisionFrame:

 def process_frame(inputs : LipSyncerInputs) -> ProcessorOutputs:
 	reference_vision_frame = inputs.get('reference_vision_frame')
+	source_vision_frames = inputs.get('source_vision_frames')
 	source_voice_frame = inputs.get('source_voice_frame')
-	target_vision_frame = inputs.get('target_vision_frame')
+	target_vision_frames = inputs.get('target_vision_frames')
 	temp_vision_frame = inputs.get('temp_vision_frame')
 	temp_vision_mask = inputs.get('temp_vision_mask')
-	target_faces = select_faces(reference_vision_frame, target_vision_frame)
+
+	target_vision_frame = get_middle(target_vision_frames)
+	target_faces = select_faces(reference_vision_frame, source_vision_frames, target_vision_frames)

 	if target_faces:
 		for target_face in target_faces:
@@ -1,4 +1,4 @@
-from typing import Any, Literal, TypeAlias, TypedDict
+from typing import Any, List, Literal, TypeAlias, TypedDict

 from numpy.typing import NDArray

@@ -7,8 +7,9 @@ from facefusion.types import AudioFrame, Mask, VisionFrame
 LipSyncerInputs = TypedDict('LipSyncerInputs',
 {
 	'reference_vision_frame' : VisionFrame,
+	'source_vision_frames' : List[VisionFrame],
 	'source_voice_frame' : AudioFrame,
-	'target_vision_frame' : VisionFrame,
+	'target_vision_frames' : List[VisionFrame],
 	'temp_vision_frame' : VisionFrame,
 	'temp_vision_mask' : Mask
 })
@@ -421,6 +421,27 @@ def create_face_selector_program() -> ArgumentParser:
 	return program


+def create_face_tracker_program() -> ArgumentParser:
+	program = ArgumentParser(add_help = False)
+	group_face_tracker = program.add_argument_group('face tracker')
+
+	capability_store.register_capability_set(
+		[
+			group_face_tracker.add_argument(
+				'--face-tracker-score',
+				help = translator.get('help.face_tracker_score'),
+				type = float,
+				default = config.get_float_value('face_tracker', 'face_tracker_score', '0.0'),
+				choices = facefusion.choices.face_tracker_score_range,
+				metavar = create_float_metavar(facefusion.choices.face_tracker_score_range)
+			)
+		],
+		scopes = [ 'api', 'cli' ]
+	)
+
+	return program
+
+
 def create_face_masker_program() -> ArgumentParser:
 	program = ArgumentParser(add_help = False)
 	group_face_masker = program.add_argument_group('face masker')
@@ -575,6 +596,27 @@ def create_frame_extraction_program() -> ArgumentParser:
 	return program


+def create_frame_distribution_program() -> ArgumentParser:
+	program = ArgumentParser(add_help = False)
+	group_frame_distribution = program.add_argument_group('frame distribution')
+
+	capability_store.register_capability_set(
+		[
+			group_frame_distribution.add_argument(
+				'--target-frame-amount',
+				help = translator.get('help.target_frame_amount'),
+				type = int,
+				default = config.get_int_value('frame_distribution', 'target_frame_amount', '5'),
+				choices = facefusion.choices.target_frame_amount_range,
+				metavar = create_int_metavar(facefusion.choices.target_frame_amount_range)
+			)
+		],
+		scopes = [ 'api', 'cli' ]
+	)
+
+	return program
+
+
 def create_output_creation_program() -> ArgumentParser:
 	program = ArgumentParser(add_help = False)
 	available_encoder_set = get_available_encoder_set()
@@ -985,9 +1027,11 @@ def collect_step_program() -> ArgumentParser:
 			create_face_detector_program(),
 			create_face_landmarker_program(),
 			create_face_selector_program(),
+			create_face_tracker_program(),
 			create_face_masker_program(),
 			create_voice_extractor_program(),
 			create_frame_extraction_program(),
+			create_frame_distribution_program(),
 			create_output_creation_program(),
 			create_processors_program()
 		],
@@ -60,7 +60,7 @@ def process_frame(stream_audio_frame : AudioFrame, stream_vision_frame : VisionF
 				'source_vision_frames': source_vision_frames,
 				'source_audio_frame': stream_audio_frame,
 				'source_voice_frame': source_voice_frame,
-				'target_vision_frame': stream_vision_frame,
+				'target_vision_frames': [ stream_vision_frame ],
 				'temp_vision_frame': temp_vision_frame,
 				'temp_vision_mask': temp_vision_mask
 			})
@@ -1,6 +1,7 @@
 import ctypes
 from collections import namedtuple
 from datetime import datetime
+from threading import Lock
 from typing import Any, Callable, Dict, List, Literal, NotRequired, Optional, Tuple, TypeAlias, TypedDict, Union

 import cv2
@@ -31,22 +32,34 @@ FaceScoreSet = TypedDict('FaceScoreSet',
 	'landmarker' : Score
 })
 Embedding : TypeAlias = NDArray[numpy.float64]
-Gender = Literal['female', 'male']
+
 Age : TypeAlias = range
+Gender = Literal['female', 'male']
 Race = Literal['white', 'black', 'latino', 'asian', 'indian', 'arabic']
+
+FaceSelectorGender = Literal['auto', 'female', 'male']
+FaceSelectorRace = Literal['auto', 'white', 'black', 'latino', 'asian', 'indian', 'arabic']
+
 Face = namedtuple('Face',
 [
+	'origin',
 	'bounding_box',
 	'score_set',
 	'landmark_set',
 	'angle',
 	'embedding',
 	'embedding_norm',
-	'gender',
 	'age',
+	'gender',
 	'race'
 ])
-FaceStore : TypeAlias = Dict[str, List[Face]]
+FaceSet = TypedDict('FaceSet',
+{
+	'lock': Lock,
+	'faces': NotRequired[List[Face]]
+})
+FaceStore : TypeAlias = Dict[str, FaceSet]
+FaceTrack : TypeAlias = Dict[int, Face]

 Language = Literal['en']
 Locales : TypeAlias = Dict[Language, Dict[str, Any]]
@@ -185,6 +198,8 @@ ImageFormat = Literal['bmp', 'jpeg', 'png', 'tiff', 'webp']
 VideoFormat = Literal['avi', 'm4v', 'mkv', 'mov', 'mp4', 'mpeg', 'mxf', 'webm', 'wmv']
 TempFrameFormat = Literal['bmp', 'jpeg', 'png', 'tiff']

+FrameSet : TypeAlias = Dict[int, str]
+
 AudioEncoder = Literal['flac', 'aac', 'libmp3lame', 'libopus', 'libvorbis', 'pcm_s16le', 'pcm_s32le']
 ImageEncoder = Literal['bmp', 'mjpeg', 'png', 'tiff', 'libwebp']
 VideoEncoder = Literal['libx264', 'libx264rgb', 'libx265', 'libvpx-vp9', 'h264_nvenc', 'hevc_nvenc', 'h264_amf', 'hevc_amf', 'h264_qsv', 'hevc_qsv', 'h264_videotoolbox', 'hevc_videotoolbox', 'rawvideo']
@@ -491,6 +506,7 @@ StateKey = Literal\
 	'reference_face_position',
 	'reference_face_distance',
 	'reference_frame_number',
+	'face_tracker_score',
 	'face_occluder_model',
 	'face_parser_model',
 	'face_mask_types',
@@ -503,6 +519,7 @@ StateKey = Literal\
 	'trim_frame_end',
 	'temp_frame_format',
 	'keep_temp',
+	'target_frame_amount',
 	'output_image_quality',
 	'output_image_scale',
 	'output_audio_encoder',
@@ -556,13 +573,14 @@ State = TypedDict('State',
 	'face_landmarker_score' : Score,
 	'face_selector_mode' : FaceSelectorMode,
 	'face_selector_order' : FaceSelectorOrder,
-	'face_selector_race' : Race,
-	'face_selector_gender' : Gender,
+	'face_selector_race' : FaceSelectorRace,
+	'face_selector_gender' : FaceSelectorGender,
 	'face_selector_age_start' : int,
 	'face_selector_age_end' : int,
 	'reference_face_position' : int,
 	'reference_face_distance' : float,
 	'reference_frame_number' : int,
+	'face_tracker_score' : Score,
 	'face_occluder_model' : FaceOccluderModel,
 	'face_parser_model' : FaceParserModel,
 	'face_mask_types' : List[FaceMaskType],
@@ -575,6 +593,7 @@ State = TypedDict('State',
 	'trim_frame_end' : int,
 	'temp_frame_format' : TempFrameFormat,
 	'keep_temp' : bool,
+	'target_frame_amount' : int,
 	'output_image_quality' : int,
 	'output_image_scale' : Scale,
 	'output_audio_encoder' : AudioEncoder,
@@ -1,6 +1,6 @@
 import math
 from functools import lru_cache
-from typing import List, Optional, Tuple
+from typing import Dict, List, Optional, Tuple

 import cv2
 import numpy
@@ -9,7 +9,7 @@ from cv2.typing import Size
 from facefusion.common_helper import is_windows
 from facefusion.filesystem import get_file_extension, is_image, is_video
 from facefusion.media_helper import restrict_trim_frame
-from facefusion.thread_helper import thread_semaphore
+from facefusion.thread_helper import thread_lock, thread_semaphore
 from facefusion.types import ColorMode, Duration, Fps, Mask, Orientation, Resolution, Scale, VisionFrame
 from facefusion.video_manager import get_video_capture

@@ -82,9 +82,10 @@ def read_video_frame(video_path : str, frame_number : int = 0) -> Optional[Visio

 		if video_capture and video_capture.isOpened():
 			frame_total = video_capture.get(cv2.CAP_PROP_FRAME_COUNT)
+			frame_position = min(frame_total, frame_number)

 			with thread_semaphore():
-				video_capture.set(cv2.CAP_PROP_POS_FRAMES, min(frame_total, frame_number - 1))
+				video_capture.set(cv2.CAP_PROP_POS_FRAMES, frame_position)
 				has_vision_frame, vision_frame = video_capture.read()

 			if has_vision_frame:
@@ -93,6 +94,51 @@ def read_video_frame(video_path : str, frame_number : int = 0) -> Optional[Visio
 	return None


+@lru_cache(maxsize = 2)
+def read_static_video_chunk(video_path : str, chunk_number : int, chunk_size : int) -> Dict[int, VisionFrame]:
+	return read_video_chunk(video_path, chunk_number, chunk_size)
+
+
+def read_video_chunk(video_path : str, chunk_number : int, chunk_size : int) -> Dict[int, VisionFrame]:
+	video_frame_chunk = {}
+
+	if is_video(video_path) and chunk_number > -1:
+		video_capture = get_video_capture(video_path)
+
+		if video_capture and video_capture.isOpened():
+			frame_total = int(video_capture.get(cv2.CAP_PROP_FRAME_COUNT))
+			frame_position = chunk_number * chunk_size
+
+			with thread_semaphore():
+				video_capture.set(cv2.CAP_PROP_POS_FRAMES, frame_position)
+
+				for frame_number in range(frame_position, min(frame_position + chunk_size, frame_total)):
+					has_vision_frame, vision_frame = video_capture.read()
+
+					if has_vision_frame:
+						video_frame_chunk[frame_number] = vision_frame
+
+	return video_frame_chunk
+
+
+def select_video_frames(video_path : str, frame_number : int = 0, frame_offset : int = 5) -> List[VisionFrame]:
+	vision_frames = []
+	chunk_size = (frame_offset * 2 + 1) * 4
+
+	if is_video(video_path):
+		with thread_lock():
+			for current_number in range(frame_number - frame_offset, frame_number + frame_offset + 1):
+				video_frame_chunk = read_static_video_chunk(video_path, current_number // chunk_size, chunk_size)
+				vision_frame = create_empty_vision_frame()
+
+				if current_number in video_frame_chunk:
+					vision_frame = video_frame_chunk.get(current_number)
+
+				vision_frames.append(vision_frame)
+
+	return vision_frames
+
+
 def count_video_frame_total(video_path : str) -> int:
 	if is_video(video_path):
 		video_capture = get_video_capture(video_path)
@@ -290,6 +336,10 @@ def blend_vision_frames(source_vision_frame : VisionFrame, target_vision_frame :
 	return blend_vision_frame


+def create_empty_vision_frame() -> VisionFrame:
+	return numpy.zeros((1, 1, 3)).astype(numpy.uint8)
+
+
 def create_tile_frames(vision_frame : VisionFrame, size : Size) -> Tuple[List[VisionFrame], int, int]:
 	tile_width = size[0] - 2 * size[2]
 	pad_size_top = size[1] + size[2]
@@ -1,4 +1,5 @@
 from concurrent.futures import ThreadPoolExecutor, as_completed
+from typing import List

 import numpy
 from tqdm import tqdm
@@ -10,7 +11,7 @@ from facefusion.filesystem import filter_audio_paths
 from facefusion.processors.core import get_processors_modules
 from facefusion.temp_helper import clear_temp_directory, create_temp_directory, resolve_temp_frame_paths
 from facefusion.types import AudioFrame, ErrorCode, VisionFrame
-from facefusion.vision import conditional_merge_vision_mask, extract_vision_mask, read_static_image, read_static_images, read_static_video_frame, restrict_video_fps, write_image
+from facefusion.vision import conditional_merge_vision_mask, extract_vision_mask, read_static_image, read_static_images, read_static_video_frame, restrict_video_fps, select_video_frames, write_image


 def is_process_stopping() -> bool:
@@ -76,13 +77,19 @@ def conditional_get_reference_vision_frame() -> VisionFrame:
 	return read_static_image(state_manager.get_item('target_path'))


+def conditional_get_target_vision_frames(temp_frame_path : str, frame_number : int) -> List[VisionFrame]:
+	if state_manager.get_item('workflow') in [ 'image-to-video', 'image-to-video:frames' ]:
+		return select_video_frames(state_manager.get_item('target_path'), frame_number, state_manager.get_item('target_frame_amount'))
+	return [ read_static_image(temp_frame_path) ]
+
+
 def process_temp_frame(temp_frame_path : str, frame_number : int) -> bool:
 	reference_vision_frame = conditional_get_reference_vision_frame()
 	source_vision_frames = read_static_images(state_manager.get_item('source_paths'))
-	target_vision_frame = read_static_image(temp_frame_path, 'rgba')
+	target_vision_frames = conditional_get_target_vision_frames(temp_frame_path, frame_number)
 	source_audio_frame = conditional_get_source_audio_frame(frame_number)
 	source_voice_frame = conditional_get_source_voice_frame(frame_number)
-	temp_vision_frame = target_vision_frame.copy()
+	temp_vision_frame = read_static_image(temp_frame_path, 'rgba').copy()
 	temp_vision_mask = extract_vision_mask(temp_vision_frame)

 	for processor_module in get_processors_modules(state_manager.get_item('processors')):
@@ -92,7 +99,7 @@ def process_temp_frame(temp_frame_path : str, frame_number : int) -> bool:
 			'source_vision_frames': source_vision_frames,
 			'source_audio_frame': source_audio_frame,
 			'source_voice_frame': source_voice_frame,
-			'target_vision_frame': target_vision_frame[:, :, :3],
+			'target_vision_frames': target_vision_frames,
 			'temp_vision_frame': temp_vision_frame[:, :, :3],
 			'temp_vision_mask': temp_vision_mask
 		})
@@ -1,4 +1,4 @@
-from facefusion.common_helper import calculate_float_step, calculate_int_step, create_float_metavar, create_float_range, create_int_metavar, create_int_range
+from facefusion.common_helper import calculate_float_step, calculate_int_step, create_float_metavar, create_float_range, create_int_metavar, create_int_range, get_middle


 def test_create_int_metavar() -> None:
@@ -25,3 +25,8 @@ def test_calc_int_step() -> None:

 def test_calc_float_step() -> None:
 	assert calculate_float_step([ 0.1, 0.2 ]) == 0.1
+
+
+def test_get_middle() -> None:
+	assert get_middle([ 1, 2, 3, 4, 5 ]) == 3
+	assert get_middle([ 1 ]) == 1
@@ -1,80 +0,0 @@
-import subprocess
-
-import pytest
-
-from facefusion import face_classifier, face_detector, face_landmarker, face_recognizer, state_manager
-from facefusion.download import conditional_download
-from facefusion.face_analyser import get_many_faces
-from facefusion.face_store import clear_faces
-from facefusion.vision import read_static_image
-from .assert_helper import get_test_example_file, get_test_examples_directory
-
-
-@pytest.fixture(scope = 'module', autouse = True)
-def before_all() -> None:
-	state_manager.init_item('execution_device_ids', [ 0 ])
-	state_manager.init_item('execution_providers', [ 'cpu' ])
-	state_manager.init_item('download_providers', [ 'github' ])
-	state_manager.init_item('face_detector_angles', [ 0 ])
-	state_manager.init_item('face_detector_model', 'many')
-	state_manager.init_item('face_detector_score', 0.5)
-	state_manager.init_item('face_landmarker_model', 'many')
-	state_manager.init_item('face_landmarker_score', 0.5)
-
-	face_classifier.pre_check()
-	face_landmarker.pre_check()
-	face_recognizer.pre_check()
-
-	conditional_download(get_test_examples_directory(),
-	[
-		'https://github.com/facefusion/facefusion-assets/releases/download/examples-3.0.0/source.jpg'
-	])
-
-	subprocess.run([ 'ffmpeg', '-i', get_test_example_file('source.jpg'), '-vf', 'crop=iw*0.8:ih*0.8', get_test_example_file('source-80crop.jpg') ])
-	subprocess.run([ 'ffmpeg', '-i', get_test_example_file('source.jpg'), '-vf', 'crop=iw*0.7:ih*0.7', get_test_example_file('source-70crop.jpg') ])
-	subprocess.run([ 'ffmpeg', '-i', get_test_example_file('source.jpg'), '-vf', 'crop=iw*0.6:ih*0.6', get_test_example_file('source-60crop.jpg') ])
-
-
-@pytest.fixture(autouse = True)
-def before_each() -> None:
-	face_classifier.clear_inference_pool()
-	face_detector.clear_inference_pool()
-	face_landmarker.clear_inference_pool()
-	face_recognizer.clear_inference_pool()
-	clear_faces()
-
-
-@pytest.mark.parametrize('face_detector_model, face_detector_size',
-[
-	('retinaface', '320x320'),
-	('scrfd', '320x320'),
-	('yolo_face', '640x640'),
-	('yunet', '640x640')
-])
-def test_get_one_face(face_detector_model : str, face_detector_size : str) -> None:
-	state_manager.init_item('face_detector_model', face_detector_model)
-	state_manager.init_item('face_detector_size', face_detector_size)
-	state_manager.init_item('face_detector_margin', (0, 0, 0, 0))
-	face_detector.pre_check()
-
-	source_paths =\
-	[
-		get_test_example_file('source.jpg'),
-		get_test_example_file('source-80crop.jpg'),
-		get_test_example_file('source-70crop.jpg'),
-		get_test_example_file('source-60crop.jpg')
-	]
-
-	for source_path in source_paths:
-		source_frame = read_static_image(source_path)
-		many_faces = get_many_faces([ source_frame ])
-
-		assert len(many_faces) == 1
-
-
-def test_get_many_faces() -> None:
-	source_path = get_test_example_file('source.jpg')
-	source_frame = read_static_image(source_path)
-	many_faces = get_many_faces([ source_frame, source_frame, source_frame ])
-
-	assert len(many_faces) == 3
@@ -0,0 +1,105 @@
+import subprocess
+
+import numpy
+import pytest
+
+from facefusion import face_classifier, face_detector, face_landmarker, face_recognizer, state_manager
+from facefusion.download import conditional_download
+from facefusion.face_creator import average_face_geometry, get_many_faces, get_one_face, refill_faces
+from facefusion.face_store import clear_faces
+from facefusion.vision import read_static_image
+from .assert_helper import get_test_example_file, get_test_examples_directory
+
+
+@pytest.fixture(scope = 'module', autouse = True)
+def before_all() -> None:
+	conditional_download(get_test_examples_directory(),
+	[
+		'https://github.com/facefusion/facefusion-assets/releases/download/examples-3.0.0/source.jpg'
+	])
+	subprocess.run([ 'ffmpeg', '-i', get_test_example_file('source.jpg'), '-vf', 'crop=iw*0.8:ih*0.8', get_test_example_file('source-80crop.jpg') ])
+	subprocess.run([ 'ffmpeg', '-i', get_test_example_file('source.jpg'), '-vf', 'crop=iw*0.7:ih*0.7', get_test_example_file('source-70crop.jpg') ])
+	subprocess.run([ 'ffmpeg', '-i', get_test_example_file('source.jpg'), '-vf', 'crop=iw*0.6:ih*0.6', get_test_example_file('source-60crop.jpg') ])
+
+	state_manager.init_item('execution_device_ids', [ 0 ])
+	state_manager.init_item('execution_providers', [ 'cpu' ])
+	state_manager.init_item('download_providers', [ 'github' ])
+	state_manager.init_item('face_detector_angles', [ 0 ])
+	state_manager.init_item('face_detector_model', 'many')
+	state_manager.init_item('face_detector_size', '640x640')
+	state_manager.init_item('face_detector_margin', (0, 0, 0, 0))
+	state_manager.init_item('face_detector_score', 0.5)
+	state_manager.init_item('face_landmarker_model', 'many')
+	state_manager.init_item('face_landmarker_score', 0.5)
+
+	face_classifier.pre_check()
+	face_detector.pre_check()
+	face_landmarker.pre_check()
+	face_recognizer.pre_check()
+
+
+@pytest.fixture(autouse = True)
+def before_each() -> None:
+	face_classifier.clear_inference_pool()
+	face_detector.clear_inference_pool()
+	face_landmarker.clear_inference_pool()
+	face_recognizer.clear_inference_pool()
+	clear_faces()
+
+
+def test_get_one_face() -> None:
+	source_vision_frame = read_static_image(get_test_example_file('source.jpg'))
+	face = get_one_face(get_many_faces([ source_vision_frame ]))
+
+	assert face.bounding_box.size == 4
+
+
+def test_get_many_faces() -> None:
+	source_path = get_test_example_file('source.jpg')
+	source_vision_frame = read_static_image(source_path)
+	many_faces = get_many_faces([ source_vision_frame, source_vision_frame, source_vision_frame ])
+
+	assert len(many_faces) == 3
+
+
+def test_refill_faces() -> None:
+	source_vision_frame = read_static_image(get_test_example_file('source.jpg'))
+	face = get_one_face(get_many_faces([ source_vision_frame ]))
+	face_first = face._replace(bounding_box = numpy.array([ 0, 0, 10, 10 ]))
+	face_middle = face._replace(bounding_box = numpy.array([ 40, 40, 50, 50 ]))
+	face_last = face._replace(bounding_box = numpy.array([ 80, 80, 90, 90 ]))
+
+	fill_faces = refill_faces([ face_first, None, face_last ])
+
+	assert fill_faces[0].bounding_box.tolist() == [ 0.0, 0.0, 10.0, 10.0 ]
+	assert fill_faces[1].bounding_box.tolist() == [ 40.0, 40.0, 50.0, 50.0 ]
+	assert fill_faces[2].bounding_box.tolist() == [ 80.0, 80.0, 90.0, 90.0 ]
+
+	fill_faces = refill_faces([ face_first, None, None, None, face_last ])
+
+	assert fill_faces[0].bounding_box.tolist() == [ 0.0, 0.0, 10.0, 10.0 ]
+	assert fill_faces[1].bounding_box.tolist() == [ 20.0, 20.0, 30.0, 30.0 ]
+	assert fill_faces[2].bounding_box.tolist() == [ 40.0, 40.0, 50.0, 50.0 ]
+	assert fill_faces[3].bounding_box.tolist() == [ 60.0, 60.0, 70.0, 70.0 ]
+	assert fill_faces[4].bounding_box.tolist() == [ 80.0, 80.0, 90.0, 90.0 ]
+
+	fill_faces = refill_faces([ face_first, None, face_middle, None, face_last ])
+
+	assert fill_faces[0].bounding_box.tolist() == [ 0.0, 0.0, 10.0, 10.0 ]
+	assert fill_faces[1].bounding_box.tolist() == [ 20.0, 20.0, 30.0, 30.0 ]
+	assert fill_faces[2].bounding_box.tolist() == [ 40.0, 40.0, 50.0, 50.0 ]
+	assert fill_faces[3].bounding_box.tolist() == [ 60.0, 60.0, 70.0, 70.0 ]
+	assert fill_faces[4].bounding_box.tolist() == [ 80.0, 80.0, 90.0, 90.0 ]
+
+
+def test_average_face_geometry() -> None:
+	source_vision_frame = read_static_image(get_test_example_file('source.jpg'))
+	face_previous = get_one_face(get_many_faces([ source_vision_frame ]))
+	face_next = get_one_face(get_many_faces([ source_vision_frame ]))
+	face_previous = face_previous._replace(bounding_box = numpy.array([ 0, 0, 10, 10 ]))
+	face_next = face_next._replace(bounding_box = numpy.array([ 80, 80, 90, 90 ]))
+
+	assert average_face_geometry([face_previous, face_next], 0.5).bounding_box.tolist() == [40.0, 40.0, 50.0, 50.0]
+	assert average_face_geometry([face_previous, face_next], 0.5).angle == face_next.angle
+	assert average_face_geometry([face_previous, face_next], 0.5).embedding is face_next.embedding
+	assert average_face_geometry([face_previous, face_next], 0.25).embedding is face_previous.embedding
@@ -0,0 +1,112 @@
+import subprocess
+
+import pytest
+
+from facefusion import face_detector, state_manager
+from facefusion.download import conditional_download
+from facefusion.face_detector import detect_with_retinaface, detect_with_scrfd, detect_with_yolo_face, detect_with_yunet
+from facefusion.face_helper import apply_nms, get_nms_threshold
+from facefusion.vision import read_static_image
+from .assert_helper import get_test_example_file, get_test_examples_directory
+
+
+@pytest.fixture(scope = 'module', autouse = True)
+def before_all() -> None:
+	conditional_download(get_test_examples_directory(),
+	[
+		'https://github.com/facefusion/facefusion-assets/releases/download/examples-3.0.0/source.jpg'
+	])
+	subprocess.run([ 'ffmpeg', '-i', get_test_example_file('source.jpg'), '-vf', 'crop=iw*0.8:ih*0.8', get_test_example_file('source-80crop.jpg') ])
+	subprocess.run([ 'ffmpeg', '-i', get_test_example_file('source.jpg'), '-vf', 'crop=iw*0.7:ih*0.7', get_test_example_file('source-70crop.jpg') ])
+	subprocess.run([ 'ffmpeg', '-i', get_test_example_file('source.jpg'), '-vf', 'crop=iw*0.6:ih*0.6', get_test_example_file('source-60crop.jpg') ])
+
+	state_manager.init_item('execution_device_ids', [ 0 ])
+	state_manager.init_item('execution_providers', [ 'cpu' ])
+	state_manager.init_item('download_providers', [ 'github' ])
+	state_manager.init_item('face_detector_angles', [ 0 ])
+	state_manager.init_item('face_detector_model', 'many')
+	state_manager.init_item('face_detector_score', 0.5)
+
+	face_detector.pre_check()
+
+	conditional_download(get_test_examples_directory(),
+	[
+		'https://github.com/facefusion/facefusion-assets/releases/download/examples-3.0.0/source.jpg'
+	])
+
+	subprocess.run([ 'ffmpeg', '-i', get_test_example_file('source.jpg'), '-vf', 'crop=iw*0.8:ih*0.8', get_test_example_file('source-80crop.jpg') ])
+	subprocess.run([ 'ffmpeg', '-i', get_test_example_file('source.jpg'), '-vf', 'crop=iw*0.7:ih*0.7', get_test_example_file('source-70crop.jpg') ])
+	subprocess.run([ 'ffmpeg', '-i', get_test_example_file('source.jpg'), '-vf', 'crop=iw*0.6:ih*0.6', get_test_example_file('source-60crop.jpg') ])
+
+
+@pytest.fixture(autouse = True)
+def before_each() -> None:
+	face_detector.clear_inference_pool()
+
+
+def test_detect_with_retinaface() -> None:
+	source_paths =\
+	[
+		get_test_example_file('source.jpg'),
+		get_test_example_file('source-80crop.jpg'),
+		get_test_example_file('source-70crop.jpg'),
+		get_test_example_file('source-60crop.jpg')
+	]
+
+	for source_path in source_paths:
+		source_frame = read_static_image(source_path)
+		bounding_boxes, face_scores, face_landmarks_5 = detect_with_retinaface(source_frame, '320x320')
+		keep_indices = apply_nms(bounding_boxes, face_scores, 0.5, get_nms_threshold('retinaface', [ 0 ]))
+
+		assert len(keep_indices) == 1
+
+
+def test_detect_with_scrfd() -> None:
+	source_paths =\
+	[
+		get_test_example_file('source.jpg'),
+		get_test_example_file('source-80crop.jpg'),
+		get_test_example_file('source-70crop.jpg'),
+		get_test_example_file('source-60crop.jpg')
+	]
+
+	for source_path in source_paths:
+		source_frame = read_static_image(source_path)
+		bounding_boxes, face_scores, face_landmarks_5 = detect_with_scrfd(source_frame, '320x320')
+		keep_indices = apply_nms(bounding_boxes, face_scores, 0.5, get_nms_threshold('scrfd', [ 0 ]))
+
+		assert len(keep_indices) == 1
+
+
+def test_detect_with_yolo_face() -> None:
+	source_paths =\
+	[
+		get_test_example_file('source.jpg'),
+		get_test_example_file('source-80crop.jpg'),
+		get_test_example_file('source-70crop.jpg'),
+		get_test_example_file('source-60crop.jpg')
+	]
+
+	for source_path in source_paths:
+		source_frame = read_static_image(source_path)
+		bounding_boxes, face_scores, face_landmarks_5 = detect_with_yolo_face(source_frame, '640x640')
+		keep_indices = apply_nms(bounding_boxes, face_scores, 0.5, get_nms_threshold('yolo_face', [ 0 ]))
+
+		assert len(keep_indices) == 1
+
+
+def test_detect_with_yunet() -> None:
+	source_paths =\
+	[
+		get_test_example_file('source.jpg'),
+		get_test_example_file('source-80crop.jpg'),
+		get_test_example_file('source-70crop.jpg'),
+		get_test_example_file('source-60crop.jpg')
+	]
+
+	for source_path in source_paths:
+		source_frame = read_static_image(source_path)
+		bounding_boxes, face_scores, face_landmarks_5 = detect_with_yunet(source_frame, '640x640')
+		keep_indices = apply_nms(bounding_boxes, face_scores, 0.5, get_nms_threshold('yunet', [ 0 ]))
+
+		assert len(keep_indices) == 1
@@ -0,0 +1,102 @@
+import numpy
+import pytest
+
+from facefusion import face_classifier, face_detector, face_landmarker, face_recognizer, state_manager
+from facefusion.common_helper import get_first, get_last
+from facefusion.download import conditional_download
+from facefusion.face_creator import get_many_faces, get_one_face
+from facefusion.face_store import clear_faces
+from facefusion.face_tracker import create_face_tracks, select_face_track, track_faces
+from facefusion.vision import read_static_video_chunk, read_static_video_frame
+from .assert_helper import get_test_example_file, get_test_examples_directory
+
+
+@pytest.fixture(scope = 'module', autouse = True)
+def before_all() -> None:
+	conditional_download(get_test_examples_directory(),
+	[
+		'https://github.com/facefusion/facefusion-assets/releases/download/examples-3.0.0/target-240p.mp4'
+	])
+
+	state_manager.init_item('execution_device_ids', [ 0 ])
+	state_manager.init_item('execution_providers', [ 'cpu' ])
+	state_manager.init_item('download_providers', [ 'github' ])
+	state_manager.init_item('face_detector_angles', [ 0 ])
+	state_manager.init_item('face_detector_model', 'yolo_face')
+	state_manager.init_item('face_detector_size', '640x640')
+	state_manager.init_item('face_detector_margin', (0, 0, 0, 0))
+	state_manager.init_item('face_detector_score', 0.5)
+	state_manager.init_item('face_landmarker_model', 'many')
+	state_manager.init_item('face_landmarker_score', 0.5)
+	state_manager.init_item('face_tracker_score', 0.3)
+
+	face_classifier.pre_check()
+	face_detector.pre_check()
+	face_landmarker.pre_check()
+	face_recognizer.pre_check()
+
+
+@pytest.fixture(autouse = True)
+def before_each() -> None:
+	face_classifier.clear_inference_pool()
+	face_detector.clear_inference_pool()
+	face_landmarker.clear_inference_pool()
+	face_recognizer.clear_inference_pool()
+	clear_faces()
+
+
+def test_track_faces() -> None:
+	target_path = get_test_example_file('target-240p.mp4')
+	video_frame_chunk = read_static_video_chunk(target_path, 0, 7)
+	target_vision_frames = [ video_frame_chunk.get(frame_number) for frame_number in sorted(video_frame_chunk) ]
+	empty_vision_frame = numpy.zeros_like(get_first(target_vision_frames))
+
+	target_vision_frames[2] = empty_vision_frame
+	target_vision_frames[3] = empty_vision_frame
+	target_vision_frames[4] = empty_vision_frame
+	target_vision_frames[5] = empty_vision_frame
+
+	assert len(track_faces(target_vision_frames, 0.3)) == 1
+
+	target_vision_frames = [ video_frame_chunk.get(frame_number) for frame_number in sorted(video_frame_chunk)[:5] ]
+	target_vision_frames[0] = empty_vision_frame
+	target_vision_frames[1] = empty_vision_frame
+	target_vision_frames[2] = empty_vision_frame
+
+	assert len(track_faces(target_vision_frames, 0.3)) == 0
+
+
+def test_create_face_tracks() -> None:
+	target_vision_frame = read_static_video_frame(get_test_example_file('target-240p.mp4'), 0)
+	multi_face_vision_frame = numpy.hstack([ target_vision_frame, target_vision_frame ])
+
+	face_tracks = create_face_tracks([ target_vision_frame, target_vision_frame ], 0.3)
+
+	assert len(face_tracks) == 1
+	assert sorted(get_first(face_tracks)) == [ 0, 1 ]
+
+	face_tracks = create_face_tracks([ multi_face_vision_frame, multi_face_vision_frame ], 0.3)
+
+	assert len(face_tracks) == 2
+	assert sorted(get_first(face_tracks)) == [ 0, 1 ]
+	assert sorted(get_last(face_tracks)) == [ 0, 1 ]
+
+	assert len(create_face_tracks([ target_vision_frame, target_vision_frame ], 1.0)) == 2
+
+
+def test_select_face_track() -> None:
+	target_vision_frame = read_static_video_frame(get_test_example_file('target-240p.mp4'), 0)
+	face = get_one_face(get_many_faces([ target_vision_frame ]))
+	face_overlap = face._replace(bounding_box = numpy.array([ 12, 12, 52, 52 ]))
+	face_distant = face._replace(bounding_box = numpy.array([ 200, 200, 240, 240 ]))
+	face_track_overlap =\
+	{
+		0 : face._replace(bounding_box = numpy.array([ 10, 10, 50, 50 ]))
+	}
+	face_track_distant =\
+	{
+		0 : face._replace(bounding_box = numpy.array([ 100, 100, 140, 140 ]))
+	}
+
+	assert select_face_track([ face_track_overlap, face_track_distant ], face_overlap, 0.3) is face_track_overlap
+	assert select_face_track([ face_track_overlap, face_track_distant ], face_distant, 0.3) == {}
@@ -1,7 +1,7 @@
 from shutil import which

 from facefusion import ffmpeg_builder
-from facefusion.ffmpeg_builder import capture_video, chain, concat, enforce_pixel_format, keep_video_alpha, run, select_frame_range, set_audio_quality, set_audio_sample_size, set_stream_mode, set_stream_quality, set_video_encoder, set_video_fps, set_video_quality
+from facefusion.ffmpeg_builder import capture_video, chain, concat, enforce_pixel_format, keep_video_alpha, run, select_frame_range, set_audio_quality, set_audio_sample_size, set_faststart, set_stream_mode, set_stream_quality, set_video_encoder, set_video_fps, set_video_quality, set_video_tag


 def test_run() -> None:
@@ -71,6 +71,24 @@ def test_set_audio_quality() -> None:
 	assert set_audio_quality('flac', 100) == []


+def test_set_faststart() -> None:
+	assert set_faststart('m4v') == [ '-movflags', '+faststart' ]
+	assert set_faststart('mov') == [ '-movflags', '+faststart' ]
+	assert set_faststart('mp4') == [ '-movflags', '+faststart' ]
+	assert set_faststart('mkv') == []
+	assert set_faststart('webm') == []
+
+
+def test_set_video_tag() -> None:
+	assert set_video_tag('libx265', 'm4v') == [ '-tag:v', 'hvc1' ]
+	assert set_video_tag('hevc_nvenc', 'mov') == [ '-tag:v', 'hvc1' ]
+	assert set_video_tag('hevc_videotoolbox', 'mp4') == [ '-tag:v', 'hvc1' ]
+	assert set_video_tag('libx265', 'mkv') == []
+	assert set_video_tag('libx265', 'webm') == []
+	assert set_video_tag('libx264', 'mp4') == []
+	assert set_video_tag('h264_nvenc', 'mp4') == []
+
+
 def test_set_video_quality() -> None:
 	assert set_video_quality('libx264', 0) == [ '-crf', '51' ]
 	assert set_video_quality('libx264', 50) == [ '-crf', '26' ]
@@ -1,9 +1,10 @@
 import subprocess

+import numpy
 import pytest

 from facefusion.download import conditional_download
-from facefusion.vision import calculate_histogram_difference, count_video_frame_total, detect_image_resolution, detect_video_duration, detect_video_fps, detect_video_resolution, match_frame_color, normalize_resolution, pack_resolution, predict_video_frame_total, read_image, read_video_frame, restrict_image_resolution, restrict_trim_video_frame, restrict_video_fps, restrict_video_resolution, scale_resolution, unpack_resolution, write_image
+from facefusion.vision import calculate_histogram_difference, count_video_frame_total, detect_image_resolution, detect_video_duration, detect_video_fps, detect_video_resolution, match_frame_color, normalize_resolution, pack_resolution, predict_video_frame_total, read_image, read_video_chunk, read_video_frame, restrict_image_resolution, restrict_trim_video_frame, restrict_video_fps, restrict_video_resolution, scale_resolution, select_video_frames, unpack_resolution, write_image
 from .assert_helper import get_test_example_file, get_test_examples_directory, get_test_output_path, prepare_test_output_directory


@@ -62,10 +63,27 @@ def test_restrict_image_resolution() -> None:


 def test_read_video_frame() -> None:
-	assert hasattr(read_video_frame(get_test_example_file('target-240p-25fps.mp4')), '__array_interface__')
+	target_path = get_test_example_file('target-240p-25fps.mp4')
+
+	assert read_video_frame(target_path).shape == (226, 426, 3)
+	assert numpy.array_equal(read_video_frame(target_path, 49), select_video_frames(target_path, 49, 5)[5])
+	assert numpy.array_equal(read_video_frame(target_path, 50), select_video_frames(target_path, 50, 5)[5])
+	assert numpy.array_equal(read_video_frame(target_path, 51), select_video_frames(target_path, 51, 5)[5])
 	assert read_video_frame('invalid') is None


+def test_read_video_chunk() -> None:
+	assert len(read_video_chunk(get_test_example_file('target-240p-25fps.mp4'), 1, 40)) == 40
+	assert read_video_chunk('invalid', 1, 40) == {}
+
+
+def test_select_video_frames() -> None:
+	assert len(select_video_frames(get_test_example_file('target-240p-25fps.mp4'), 50, 5)) == 11
+	assert len(select_video_frames(get_test_example_file('target-240p-25fps.mp4'), 1, 5)) == 11
+	assert len(select_video_frames(get_test_example_file('target-240p-25fps.mp4'), 269, 5)) == 11
+	assert select_video_frames('invalid', 50, 5) == []
+
+
 def test_count_video_frame_total() -> None:
 	assert count_video_frame_total(get_test_example_file('target-240p-25fps.mp4')) == 270
 	assert count_video_frame_total(get_test_example_file('target-240p-30fps.mp4')) == 324