Compare commits

..
Author SHA1 Message Date
henryruhs a457214575 fix potential memory leak 2026-07-31 14:10:20 +02:00
b60ea40d26 3.8.0 (#1212)
* mark as next

* unify the dependency checks in pre_check and add ffprobe (#1181)

* drop keep_temp and the common options component (#1180)

Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

* introduce ffprobe and ffprobe_builder (#1182)

* introduce ffprobe and ffprobe_builder

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

* introduce ffprobe and ffprobe_builder

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

---------

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

* probe video metadata via ffprobe in vision (#1184)

* probe video metadata via ffprobe in vision

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

* probe video metadata via ffprobe in vision

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

---------

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

* adopt the workflow task vocabulary from next major (#1185)

Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

* introduce workflow-mode and workflow-strategy like next major (#1187)

Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

* restrict hdr color transfer and tag the merge output as bt709 (#1188)

* restrict hdr color transfer and tag the merge output as bt709

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

* restrict hdr color transfer and tag the merge output as bt709

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

* full video migration

* compose the hdr fixture via the builder chain

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

* compose the test fixtures via the builder and run_ffmpeg

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

---------

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

* compose every test fixture via the builder and run_ffmpeg (#1189)

* compose every test fixture via the builder and run_ffmpeg

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

* use loops in tests for ffmpeg stuff

---------

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

* New Video Manager (#1191)

* tiny adjustment for tests

* address the review on the video manager

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

* introduce the stream strategy for the video workflow (#1192)

* introduce the stream strategy for the video workflow

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

* address the review on the stream strategy

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

---------

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

* annotate the changes for review

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

* annotate the new tests for review

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

* match the temp pixel format help to the locale style

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

* question the set_input_seek naming

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

* question the reader and writer keys

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

* drop the review annotations from the encoder mapping tests

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

* drop the review annotations from the thread count tests

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

* drop the review annotations from the ui files

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

* capture the open review questions as annotations

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

* drop the settled annotations from the types

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbcd6VWCiU4BQP1gywPr2a

---------

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

* switch to ffmpeg.style for audio.py

* remove todos that were never needed

* fix for ffmpeg7

* Add frame_store module (#1194)

* add frame_store module

* rename and change tests

* rename and update tests

* route window read through frame_store (#1196)

* route window read through frame_store

* update proper id

* restore todos

* restore todos

* go v4 style for workflow (#1197)

* go v4 style for workflow

* remove some todos

* route chunk read through frame_store (#1198)

* vision integration

* Deleted read_video_chunk + read_static_video_chunk

* margin decouple (#1199)

* fix windows CI fail (#1200)

* Cleanup Part1 (#1201)

* remove some todos, improve video manager, simplify ffmpeg commands and more

* do more

* remove thread count for filters

* Cleanup Part 2 (#1202)

* tons of renaming

* tons of renaming

* multi reader approach

* bring tests to an okay-ish state

* bring drain back

* improve read_video_frame speed

* rename method

* move variables

* seek video reader only when trim frame start is larger 0

* make stream the default

* Cleanup/part 3 (#1203)

* remove todo

* sort out workflow, to match upcoming v4

* remove look ahead

* remove core namespace again

* Revamp execution provider overrides/adjustments (#1206)

* Split provider hooks into override/adjust with cached CoreML base

Replace the single resolve_inference_providers processor hook with two:
override_inference_providers (full replacement) and adjust_inference_providers
(merge options onto the base providers built by create_inference_providers).
This lets CoreML processors inherit ModelCacheDirectory + SpecializationStrategy
from the base while layering ModelFormat/MLComputeUnits on top.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HTQCZiYjJyUX11bDpbRSiB

* fix caching for execution provider by having override and adjust ways

* fix caching for execution provider by having override and adjust ways

* fix lint

* use proper pytest fixtures

---------

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix update preview bug (#1205)

* fix update preview bug

* fix update preview bug

* remove guard

* add is_vision_frame

* Restrict the preview frame slider and the reader seek to the last frame index (#1207)

* fix index bug

* fix rounding bug

* avoid tobytes copy (#1208)

* beautify tests

* hide ffmpeg warnings

* simplify process_stream_frame

* Use is vision frame everywhere (#1210)

* use is_vision_frame everywhere

* fix hash

* fix lint

* fix hash creation in face store

* that model does not exist

* update workflow ffmpeg

* guard workflow (#1211)

* bump version and dependencies

* Update preview

* switch workflow strategy to disk|memory

* update preview

* update preview

* fix wording

* last minute change workflow position

* adjust wording

---------

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: Harisreedhar <46858047+harisreedhar@users.noreply.github.com>
Co-authored-by: harisreedhar <h4harisreedhar.s.s@gmail.com>
2026-07-30 22:25:09 +02:00
22 changed files with 85 additions and 114 deletions
Binary file not shown.

Before

Width:  |  Height:  |  Size: 1.1 MiB

After

Width:  |  Height:  |  Size: 1.1 MiB

+1 -1
View File
@@ -43,7 +43,7 @@ jobs:
- name: Checkout
uses: actions/checkout@v4
- name: Set up FFmpeg
uses: FedericoCarboni/setup-ffmpeg@v3
uses: AnimMouse/setup-ffmpeg@v1
- name: Set up Python 3.12
uses: actions/setup-python@v5
with:
+4 -4
View File
@@ -1,7 +1,3 @@
[workflow]
workflow_mode =
workflow_strategy =
[paths]
temp_path =
jobs_path =
@@ -72,6 +68,10 @@ output_video_quality =
output_video_scale =
output_video_fps =
[workflow]
workflow_mode =
workflow_strategy =
[processors]
processors =
age_modifier_model =
+2 -2
View File
@@ -9,8 +9,6 @@ from facefusion.vision import detect_video_fps
def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
apply_state_item('command', args.get('command'))
apply_state_item('workflow_mode', args.get('workflow_mode'))
apply_state_item('workflow_strategy', args.get('workflow_strategy'))
apply_state_item('temp_path', args.get('temp_path'))
apply_state_item('jobs_path', args.get('jobs_path'))
apply_state_item('source_paths', args.get('source_paths'))
@@ -63,6 +61,8 @@ def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
output_video_fps = normalize_fps(args.get('output_video_fps')) or detect_video_fps(args.get('target_path'))
apply_state_item('output_video_fps', output_video_fps)
apply_state_item('workflow_mode', args.get('workflow_mode'))
apply_state_item('workflow_strategy', args.get('workflow_strategy'))
available_processors = [ get_file_name(file_path) for file_path in resolve_file_paths('facefusion/processors/modules') ]
apply_state_item('processors', args.get('processors'))
+2 -2
View File
@@ -9,7 +9,7 @@ from facefusion.download import conditional_download_hashes, conditional_downloa
from facefusion.filesystem import resolve_relative_path
from facefusion.thread_helper import conditional_thread_semaphore
from facefusion.types import Detection, DownloadScope, DownloadSet, Fps, InferencePool, ModelSet, VisionFrame
from facefusion.vision import detect_video_fps, fit_contain_frame, read_image
from facefusion.vision import detect_video_fps, fit_contain_frame, is_vision_frame, read_image
STREAM_COUNTER = 0
@@ -172,7 +172,7 @@ def analyse_video(video_path : str, trim_frame_start : int, trim_frame_end : int
vision_frame = video_manager.read_video_frame(video_reader)
if frame_number % int(video_fps) == 0:
if numpy.any(vision_frame):
if is_vision_frame(vision_frame):
total += 1
if analyse_frame(vision_frame):
+12 -9
View File
@@ -100,7 +100,7 @@ def pre_check() -> bool:
def common_pre_check() -> bool:
content_analyser_content = inspect.getsource(content_analyser).encode()
return hash_helper.create_hash(content_analyser_content) == '0922c180'
return hash_helper.create_hash(content_analyser_content) == '3c6ce25e'
def processors_pre_check() -> bool:
@@ -315,16 +315,19 @@ def conditional_process() -> ErrorCode:
if state_manager.get_item('workflow_mode') == 'auto':
state_manager.set_item('workflow_mode', detect_workflow_mode())
for processor_module in get_processors_modules(state_manager.get_item('processors')):
if not processor_module.pre_process('output'):
return 2
if state_manager.get_item('workflow_mode') == detect_workflow_mode():
for processor_module in get_processors_modules(state_manager.get_item('processors')):
if not processor_module.pre_process('output'):
return 2
if state_manager.get_item('workflow_mode') == 'image-to-image':
return image_to_image.process(start_time)
if state_manager.get_item('workflow_mode') == 'image-to-video':
return image_to_video.process(start_time)
if state_manager.get_item('workflow_mode') == 'image-to-image':
return image_to_image.process(start_time)
if state_manager.get_item('workflow_mode') == 'image-to-video':
return image_to_video.process(start_time)
return 0
return 0
return 2
def detect_workflow_mode() -> WorkflowMode:
+2 -1
View File
@@ -10,6 +10,7 @@ from facefusion.face_helper import apply_nms, average_points, convert_to_face_la
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
from facefusion.vision import is_vision_frame
def create_faces(vision_frame : VisionFrame, bounding_boxes : List[BoundingBox], face_scores : List[Score], face_landmarks_5 : List[FaceLandmark5]) -> List[Face]:
@@ -73,7 +74,7 @@ def get_many_faces(vision_frames : List[VisionFrame]) -> List[Face]:
many_faces : List[Face] = []
for vision_frame in vision_frames:
if numpy.any(vision_frame):
if is_vision_frame(vision_frame):
all_bounding_boxes = []
all_face_scores = []
all_face_landmarks_5 = []
+7 -8
View File
@@ -1,17 +1,16 @@
import threading
from typing import List, Optional
import numpy
from facefusion.hash_helper import create_hash
from facefusion.types import Face, FaceStore, VisionFrame
from facefusion.vision import is_vision_frame
FACE_STORE : FaceStore = {}
def get_faces(vision_frame : VisionFrame) -> Optional[List[Face]]:
if numpy.any(vision_frame):
vision_hash = create_hash(vision_frame.data)
if is_vision_frame(vision_frame):
vision_hash = create_hash(vision_frame.tobytes())
if FACE_STORE.get(vision_hash):
return FACE_STORE.get(vision_hash).get('faces')
@@ -20,8 +19,8 @@ def get_faces(vision_frame : VisionFrame) -> Optional[List[Face]]:
def set_faces(vision_frame : VisionFrame, faces : List[Face]) -> None:
if numpy.any(vision_frame):
vision_hash = create_hash(vision_frame.data)
if is_vision_frame(vision_frame):
vision_hash = create_hash(vision_frame.tobytes())
FACE_STORE.setdefault(vision_hash,
{
'lock': threading.Lock()
@@ -29,8 +28,8 @@ def set_faces(vision_frame : VisionFrame, faces : List[Face]) -> None:
def resolve_lock(vision_frame : VisionFrame) -> threading.Lock:
if numpy.any(vision_frame):
vision_hash = create_hash(vision_frame.data)
if is_vision_frame(vision_frame):
vision_hash = create_hash(vision_frame.tobytes())
return FACE_STORE.setdefault(vision_hash,
{
'lock': threading.Lock()
+1 -1
View File
@@ -66,7 +66,7 @@ def run_ffmpeg(commands : List[Command]) -> subprocess.Popen[bytes]:
def open_ffmpeg(commands : List[Command]) -> subprocess.Popen[bytes]:
commands = ffmpeg_builder.run(commands)
return subprocess.Popen(commands, stdin = subprocess.PIPE, stdout = subprocess.PIPE, pipesize = 1024 * 1024)
return subprocess.Popen(commands, stdin = subprocess.PIPE, stderr = subprocess.DEVNULL, stdout = subprocess.PIPE)
def create_video_reader(video_path : str, frame_number : int, video_metadata : VideoReaderMetadata) -> subprocess.Popen[bytes]:
+2 -2
View File
@@ -28,8 +28,8 @@ if is_windows():
ONNXRUNTIME_SET['directml'] = ('onnxruntime-directml', '1.24.4')
ONNXRUNTIME_SET['qnn'] = ('onnxruntime-qnn', '1.24.4')
if is_linux():
ONNXRUNTIME_SET['migraphx'] = ('onnxruntime-migraphx', '1.25.0')
ONNXRUNTIME_SET['rocm'] = ('onnxruntime-rocm', '1.22.2.post1')
ONNXRUNTIME_SET['migraphx'] = ('onnxruntime-migraphx', '1.26.0')
ONNXRUNTIME_SET['rocm'] = ('onnxruntime-rocm', '1.22.2.post3')
def cli() -> None:
+3 -4
View File
@@ -98,8 +98,6 @@ LOCALES : Locales =\
{
'install_dependency': 'choose the variant of {dependency} to install',
'skip_conda': 'skip the conda environment check',
'workflow_mode': 'choose the workflow mode',
'workflow_strategy': 'choose the workflow strategy',
'config_path': 'choose the config file to override defaults',
'temp_path': 'specify the directory for the temporary resources',
'jobs_path': 'specify the directory to store jobs',
@@ -136,7 +134,7 @@ LOCALES : Locales =\
'voice_extractor_model': 'choose the model responsible for extracting the voices',
'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',
'temp_frame_format': 'specify the temporary frame format',
'temp_pixel_format': 'specify the temporary pixel 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',
@@ -149,6 +147,8 @@ LOCALES : Locales =\
'output_video_quality': 'specify the video quality which translates to the video compression',
'output_video_scale': 'specify the video scale based on the target video',
'output_video_fps': 'specify the video fps based on the target video',
'workflow_mode': 'detect or enforce the workflow mode',
'workflow_strategy': 'process the temporary frames in memory or on disk',
'processors': 'load a single or multiple processors (choices: {choices}, ...)',
'background-remover-model': 'choose the model responsible for removing the background',
'background-remover-color': 'apply red, green blue and alpha values of the background',
@@ -204,7 +204,6 @@ LOCALES : Locales =\
'clear_button': 'CLEAR',
'download_providers_checkbox_group': 'DOWNLOAD PROVIDERS',
'execution_providers_checkbox_group': 'EXECUTION PROVIDERS',
'workflow_mode_dropdown': 'WORKFLOW MODE',
'workflow_strategy_dropdown': 'WORKFLOW STRATEGY',
'execution_thread_count_slider': 'EXECUTION THREAD COUNT',
'face_detector_angles_checkbox_group': 'FACE DETECTOR ANGLES',
+1 -1
View File
@@ -4,7 +4,7 @@ METADATA =\
{
'name': 'FaceFusion',
'description': 'Industry leading face manipulation platform',
'version': 'NEXT',
'version': '3.8.0',
'license': 'OpenRAIL-AS',
'author': 'Henry Ruhs',
'url': 'https://facefusion.io'
@@ -9,4 +9,4 @@ FrameEnhancerInputs = TypedDict('FrameEnhancerInputs',
'temp_vision_mask' : Mask
})
FrameEnhancerModel = Literal['clear_reality_x4', 'face_dat_x4', 'lsdir_x4', 'nomos8k_sc_x4', 'real_esrgan_x2', 'real_esrgan_x2_fp16', 'real_esrgan_x4', 'real_esrgan_x4_fp16', 'real_esrgan_x8', 'real_esrgan_x8_fp16', 'real_hatgan_x4', 'real_web_photo_x4', 'realistic_rescaler_x4', 'remacri_x4', 'siax_x4', 'span_kendata_x4', 'swin2_sr_x4', 'tghq_face_x8', 'ultra_sharp_x4', 'ultra_sharp_2_x4']
FrameEnhancerModel = Literal['clear_reality_x4', 'face_dat_x4', 'nomos8k_sc_x4', 'real_esrgan_x2', 'real_esrgan_x2_fp16', 'real_esrgan_x4', 'real_esrgan_x4_fp16', 'real_esrgan_x8', 'real_esrgan_x8_fp16', 'real_hatgan_x4', 'real_web_photo_x4', 'realistic_rescaler_x4', 'remacri_x4', 'siax_x4', 'span_kendata_x4', 'swin2_sr_x4', 'tghq_face_x8', 'ultra_sharp_x4', 'ultra_sharp_2_x4']
+10 -10
View File
@@ -21,15 +21,6 @@ def create_help_formatter_large(prog : str) -> HelpFormatter:
return HelpFormatter(prog, max_help_position = 300)
def create_workflow_program() -> ArgumentParser:
program = ArgumentParser(add_help = False)
group_workflow = program.add_argument_group('workflow')
group_workflow.add_argument('--workflow-mode', help = translator.get('help.workflow_mode'), default = config.get_str_value('workflow', 'workflow_mode', 'auto'), choices = facefusion.choices.workflow_modes)
group_workflow.add_argument('--workflow-strategy', help = translator.get('help.workflow_strategy'), default = config.get_str_value('workflow', 'workflow_strategy', 'stream'), choices = facefusion.choices.workflow_strategies)
job_store.register_step_keys([ 'workflow_mode', 'workflow_strategy' ])
return program
def create_config_path_program() -> ArgumentParser:
program = ArgumentParser(add_help = False)
group_paths = program.add_argument_group('paths')
@@ -209,6 +200,15 @@ def create_output_creation_program() -> ArgumentParser:
return program
def create_workflow_program() -> ArgumentParser:
program = ArgumentParser(add_help = False)
group_workflow = program.add_argument_group('workflow')
group_workflow.add_argument('--workflow-mode', help = translator.get('help.workflow_mode'), default = config.get_str_value('workflow', 'workflow_mode', 'auto'), choices = facefusion.choices.workflow_modes)
group_workflow.add_argument('--workflow-strategy', help = translator.get('help.workflow_strategy'), default = config.get_str_value('workflow', 'workflow_strategy', 'memory'), choices = facefusion.choices.workflow_strategies)
job_store.register_step_keys([ 'workflow_mode', 'workflow_strategy' ])
return program
def create_processors_program() -> ArgumentParser:
program = ArgumentParser(add_help = False)
available_processors = [ get_file_name(file_path) for file_path in resolve_file_paths('facefusion/processors/modules') ]
@@ -309,7 +309,7 @@ def create_step_index_program() -> ArgumentParser:
def collect_step_program() -> ArgumentParser:
return ArgumentParser(parents = [ create_workflow_program(), 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() ], add_help = False)
return ArgumentParser(parents = [ 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_workflow_program(), create_processors_program() ], add_help = False)
def collect_job_program() -> ArgumentParser:
+2 -3
View File
@@ -5,7 +5,6 @@ from concurrent.futures import ThreadPoolExecutor
from typing import Deque, Iterator, List
import cv2
import numpy
from tqdm import tqdm
from facefusion import ffmpeg_builder, logger, state_manager, translator
@@ -15,7 +14,7 @@ from facefusion.ffmpeg import open_ffmpeg
from facefusion.filesystem import is_directory
from facefusion.processors.core import get_processors_modules
from facefusion.types import Fps, StreamMode, VisionFrame
from facefusion.vision import extract_vision_mask, read_static_images
from facefusion.vision import extract_vision_mask, is_vision_frame, read_static_images
def multi_process_capture(camera_capture : cv2.VideoCapture, camera_fps : Fps) -> Iterator[VisionFrame]:
@@ -31,7 +30,7 @@ def multi_process_capture(camera_capture : cv2.VideoCapture, camera_fps : Fps) -
if analyse_stream(capture_vision_frame, camera_fps):
camera_capture.release()
if numpy.any(capture_vision_frame):
if is_vision_frame(capture_vision_frame):
future = executor.submit(process_stream_frame, source_vision_frames, capture_vision_frame)
futures.append(future)
+5 -5
View File
@@ -65,7 +65,7 @@ Locales : TypeAlias = Dict[Language, Dict[str, Any]]
LocalePoolSet : TypeAlias = Dict[str, Locales]
WorkflowMode = Literal['auto', 'image-to-image', 'image-to-video']
WorkflowStrategy = Literal['disk', 'stream']
WorkflowStrategy = Literal['disk', 'memory']
CameraCaptureSet : TypeAlias = Dict[str, cv2.VideoCapture]
CameraPoolSet = TypedDict('CameraPoolSet',
@@ -322,8 +322,6 @@ JobSet : TypeAlias = Dict[str, Job]
StateKey = Literal\
[
'command',
'workflow_mode',
'workflow_strategy',
'config_path',
'temp_path',
'jobs_path',
@@ -378,6 +376,8 @@ StateKey = Literal\
'output_video_quality',
'output_video_scale',
'output_video_fps',
'workflow_mode',
'workflow_strategy',
'processors',
'open_browser',
'ui_layouts',
@@ -395,8 +395,6 @@ StateKey = Literal\
State = TypedDict('State',
{
'command' : str,
'workflow_mode' : WorkflowMode,
'workflow_strategy' : WorkflowStrategy,
'config_path' : str,
'temp_path' : str,
'jobs_path' : str,
@@ -451,6 +449,8 @@ State = TypedDict('State',
'output_video_quality' : int,
'output_video_scale' : Scale,
'output_video_fps' : float,
'workflow_mode' : WorkflowMode,
'workflow_strategy' : WorkflowStrategy,
'processors' : List[str],
'open_browser' : bool,
'ui_layouts' : List[str],
+2 -2
View File
@@ -18,7 +18,7 @@ from facefusion.types import AudioFrame, Face, Mask, VisionFrame
from facefusion.uis import choices as uis_choices
from facefusion.uis.core import get_ui_component, get_ui_components, register_ui_component
from facefusion.uis.types import ComponentOptions, PreviewMode
from facefusion.vision import detect_frame_orientation, extract_vision_mask, fit_cover_frame, merge_vision_mask, obscure_frame, read_static_image, read_static_images, read_video_frame, restrict_frame, select_video_frames, unpack_resolution
from facefusion.vision import detect_frame_orientation, extract_vision_mask, fit_cover_frame, is_vision_frame, merge_vision_mask, obscure_frame, read_static_image, read_static_images, read_video_frame, restrict_frame, select_video_frames, unpack_resolution
PREVIEW_IMAGE : Optional[gradio.Image] = None
@@ -277,7 +277,7 @@ def create_face_by_face(reference_vision_frame : VisionFrame, source_vision_fram
target_crop_vision_frame = extract_crop_frame(target_vision_frame, target_face)
output_crop_vision_frame = extract_crop_frame(temp_vision_frame, target_face)
if numpy.any(target_crop_vision_frame) and numpy.any(output_crop_vision_frame):
if is_vision_frame(target_crop_vision_frame) and is_vision_frame(output_crop_vision_frame):
target_crop_dimension = min(target_crop_vision_frame.shape[:2])
target_crop_vision_frame = fit_cover_frame(target_crop_vision_frame, (target_crop_dimension, target_crop_dimension))
output_crop_vision_frame = fit_cover_frame(output_crop_vision_frame, (target_crop_dimension, target_crop_dimension))
+1 -13
View File
@@ -4,21 +4,14 @@ import gradio
import facefusion.choices
from facefusion import state_manager, translator
from facefusion.types import WorkflowMode, WorkflowStrategy
from facefusion.types import WorkflowStrategy
WORKFLOW_MODE_DROPDOWN : Optional[gradio.Dropdown] = None
WORKFLOW_STRATEGY_DROPDOWN : Optional[gradio.Dropdown] = None
def render() -> None:
global WORKFLOW_MODE_DROPDOWN
global WORKFLOW_STRATEGY_DROPDOWN
WORKFLOW_MODE_DROPDOWN = gradio.Dropdown(
label = translator.get('uis.workflow_mode_dropdown'),
choices = facefusion.choices.workflow_modes,
value = state_manager.get_item('workflow_mode')
)
WORKFLOW_STRATEGY_DROPDOWN = gradio.Dropdown(
label = translator.get('uis.workflow_strategy_dropdown'),
choices = facefusion.choices.workflow_strategies,
@@ -27,13 +20,8 @@ def render() -> None:
def listen() -> None:
WORKFLOW_MODE_DROPDOWN.change(update_workflow_mode, inputs = WORKFLOW_MODE_DROPDOWN)
WORKFLOW_STRATEGY_DROPDOWN.change(update_workflow_strategy, inputs = WORKFLOW_STRATEGY_DROPDOWN)
def update_workflow_mode(workflow_mode : WorkflowMode) -> None:
state_manager.set_item('workflow_mode', workflow_mode)
def update_workflow_strategy(workflow_strategy : WorkflowStrategy) -> None:
state_manager.set_item('workflow_strategy', workflow_strategy)
+3 -3
View File
@@ -3,7 +3,7 @@ from functools import partial
from facefusion import process_manager, state_manager
from facefusion.types import ErrorCode
from facefusion.workflows.core import clear, setup
from facefusion.workflows.to_video import analyse_video, extract_frames, finalize_video, merge_frames, process_disk_frames, process_stream_frames, restore_audio
from facefusion.workflows.to_video import analyse_video, extract_frames, finalize_video, merge_frames, process_disk_frames, process_memory_frames, restore_audio
def process(start_time : float) -> ErrorCode:
@@ -22,8 +22,8 @@ def process(start_time : float) -> ErrorCode:
merge_frames
])
if state_manager.get_item('workflow_strategy') == 'stream':
tasks.append(process_stream_frames)
if state_manager.get_item('workflow_strategy') == 'memory':
tasks.append(process_memory_frames)
tasks.extend(
[
+9 -31
View File
@@ -1,5 +1,5 @@
from concurrent.futures import ThreadPoolExecutor, as_completed
from typing import Tuple
from concurrent.futures import ThreadPoolExecutor
from itertools import repeat
import cv2
import numpy
@@ -58,19 +58,8 @@ def process_disk_frames() -> ErrorCode:
read_static_video_frame(state_manager.get_item('target_path'), state_manager.get_item('reference_frame_number'))
with ThreadPoolExecutor(max_workers = state_manager.get_item('execution_thread_count')) as executor:
futures = []
for frame_number, temp_frame_path in temp_frame_set.items():
future = executor.submit(process_disk_frame, temp_frame_path, frame_number)
futures.append(future)
for future in as_completed(futures):
if is_process_stopping():
for pending_future in futures:
pending_future.cancel()
if not future.cancelled():
future.result()
for temp_vision_frame_ in executor.map(process_disk_frame, temp_frame_set.values(), temp_frame_set.keys()):
if not is_process_stopping():
progress.update()
for processor_module in get_processors_modules(state_manager.get_item('processors')):
@@ -84,7 +73,7 @@ def process_disk_frames() -> ErrorCode:
return 0
def process_stream_frame(frame_number : int, temp_video_resolution : Resolution) -> Tuple[int, VisionFrame]:
def process_memory_frame(frame_number : int, temp_video_resolution : Resolution) -> VisionFrame:
target_vision_frames = select_video_frames(state_manager.get_item('target_path'), frame_number, state_manager.get_item('target_frame_amount'))
target_vision_frame = get_middle(target_vision_frames)
temp_vision_frame = target_vision_frame.copy()
@@ -100,10 +89,10 @@ def process_stream_frame(frame_number : int, temp_video_resolution : Resolution)
if state_manager.get_item('temp_pixel_format') == 'bgr24':
temp_vision_frame = temp_vision_frame[:, :, :3]
return frame_number, numpy.ascontiguousarray(temp_vision_frame)
return numpy.ascontiguousarray(temp_vision_frame)
def process_stream_frames() -> ErrorCode:
def process_memory_frames() -> ErrorCode:
trim_frame_start, trim_frame_end = restrict_trim_frame(state_manager.get_item('target_path'), state_manager.get_item('trim_frame_start'), state_manager.get_item('trim_frame_end'))
output_video_resolution = scale_resolution(detect_video_resolution(state_manager.get_item('target_path')), state_manager.get_item('output_video_scale'))
temp_video_resolution = restrict_video_resolution(state_manager.get_item('target_path'), output_video_resolution)
@@ -119,19 +108,8 @@ def process_stream_frames() -> ErrorCode:
read_static_video_frame(state_manager.get_item('target_path'), state_manager.get_item('reference_frame_number'))
with ThreadPoolExecutor(max_workers = state_manager.get_item('execution_thread_count')) as executor:
futures = []
for frame_number in temp_frame_range:
future = executor.submit(process_stream_frame, frame_number, temp_video_resolution)
futures.append(future)
for future in futures:
if is_process_stopping():
for pending_future in futures:
pending_future.cancel()
if not future.cancelled():
_, temp_vision_frame = future.result()
for temp_vision_frame in executor.map(process_memory_frame, temp_frame_range, repeat(temp_video_resolution)):
if not is_process_stopping():
video_manager.write_video_frame(video_writer, temp_vision_frame)
progress.update()
+5 -5
View File
@@ -1,8 +1,8 @@
gradio-rangeslider==0.0.8
gradio==5.50.0
numpy==2.2.1
onnx==1.21.0
numpy==2.4.6
onnx==1.22.0
onnxruntime==1.26.0
opencv-python==4.13.0.92
tqdm==4.67.3
scipy==1.17.1
opencv-python-headless==5.0.0.93
tqdm==4.70.0
scipy==1.18.0
+10 -6
View File
@@ -50,34 +50,35 @@ def test_get_reader() -> None:
assert video_metadata.get('resolution') == (426, 226)
assert video_metadata.get('fps') == 25.0
assert video_metadata.get('frame_total') == 270
assert get_reader(get_test_example_file('target-240p-25fps.mp4'), 'read_video_frame') is video_reader
assert not get_reader(get_test_example_file('target-240p-25fps.mp4'), 'select_video_frames').get('id') == video_reader.get('id')
def test_conditional_seek_video_reader() -> None:
video_reader = get_reader(get_test_example_file('target-240p-25fps.mp4'), 'read_video_frame')
sequential_frames = {}
video_frames = {}
for frame_number in range(30):
sequential_frames[frame_number] = read_video_frame(video_reader)
video_frames[frame_number] = read_video_frame(video_reader)
for frame_number in [ 5, 17, 29 ]:
conditional_seek_video_reader(video_reader, frame_number)
assert numpy.array_equal(read_video_frame(video_reader), sequential_frames.get(frame_number)) is True
assert numpy.array_equal(read_video_frame(video_reader), video_frames.get(frame_number)) is True
def test_seek_video_reader() -> None:
video_reader = get_reader(get_test_example_file('target-240p-25fps.mp4'), 'read_video_frame')
sequential_frames = {}
video_frames = {}
for frame_number in range(30):
sequential_frames[frame_number] = read_video_frame(video_reader)
video_frames[frame_number] = read_video_frame(video_reader)
for frame_number in [ 5, 17, 29 ]:
seek_video_reader(video_reader, frame_number)
assert numpy.array_equal(read_video_frame(video_reader), sequential_frames.get(frame_number)) is True
assert numpy.array_equal(read_video_frame(video_reader), video_frames.get(frame_number)) is True
def test_drain_video_reader() -> None:
@@ -133,6 +134,7 @@ def test_collect_video_frames() -> None:
def test_close_video_reader() -> None:
video_reader = get_reader(get_test_example_file('target-240p-25fps.mp4'), 'select_video_frames')
read_video_frames(video_reader, 0, 4)
close_video_reader(video_reader)
@@ -176,6 +178,7 @@ def test_close_video_writer() -> None:
create_temp_directory(target_path)
video_reader = get_reader(target_path, 'read_video_frame')
video_writer = get_writer(target_path, 30.0, (426, 226), (426, 226), 30.0)
write_video_frame(video_writer, read_video_frame(video_reader))
assert close_video_writer(video_writer) is True
@@ -186,6 +189,7 @@ def test_clear_video_pool() -> None:
create_temp_directory(target_path)
video_reader = get_reader(target_path, 'select_video_frames')
video_writer = get_writer(target_path, 25.0, (426, 226), (426, 226), 25.0)
read_video_frames(video_reader, 0, 4)
write_video_frame(video_writer, read_video_frame(video_reader))
clear_video_pool()