CalvinBackup/remove-ai-watermarks
1
0
Fork 0
mirror of https://github.com/wiltodelta/remove-ai-watermarks.git synced 2026-08-27 16:02:28 +02:00
Code Issues Packages Projects Releases Wiki Activity

feat(visible): auto fill prefers LaMa > MI-GAN > cv2, warn on cv2 fallback

Browse Source 
The auto backend now resolves best-first: LaMa (highest quality, recovers the
textured/structured backgrounds the classical fill smears) > MI-GAN > cv2. Both
learned backends share the same onnxruntime availability check, so auto cannot
tell them apart and always prefers the better one; a memory-tight deployment
that cannot afford LaMa's ~4.7 GB peak pins MI-GAN explicitly via
`--backend migan` / `backend="migan"` (the deployment's call, not the library's).
cv2 stays the no-deps floor and now emits a one-time quality warning when auto
falls back to it, since it smears texture/structure.

Motivated by a v0.12.1 reverse-alpha vs 0.14 localize->fill head-to-head:
reverse-alpha recovered structured backgrounds more cleanly than any inpaint;
LaMa closes most of that gap, MI-GAN can ghost/hallucinate, cv2 is weakest.
doubao/jimeng removal is identical between versions; gemini strict coverage is
4pp lower (all recovered via assume_ai) with cleaner clearance and no
outside-box damage.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
Victor Kuznetsov
2026-07-09 18:24:20 +03:00
co-authored by Claude Opus 4.8
parent 9756189eaf
commit c858006e93
7 changed files with 78 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/remove_ai_watermarks/cli.py
+6 -6
View File
@@ -296,10 +296,10 @@ _visible_backend_option = click.option(
"backend",
type=click.Choice(["auto", "cv2", "migan", "lama"]),
default="auto",
help="Fill backend for visible-mark removal (localize -> fill). auto: MI-GAN when "
"the 'migan' extra is installed, else cv2. cv2: classical inpaint (no deps). "
"migan: MI-GAN ONNX (light, needs 'migan'). lama: big-LaMa ONNX (best quality, "
"needs 'lama').",
help="Fill backend for visible-mark removal (localize -> fill). auto: best available, "
"LaMa > MI-GAN > cv2 (a learned backend needs the 'lama' or 'migan' extra; else cv2, "
"with a warning). cv2: classical inpaint (no deps, smears texture). migan: MI-GAN ONNX "
"(light, ~1 GB, the memory-tight pick). lama: big-LaMa ONNX (best quality, ~4.7 GB).",
)
@@ -556,8 +556,8 @@ def cmd_visible(
Finds a known mark in its usual place (Gemini sparkle / Doubao-Jimeng-Samsung
text) via the watermark registry and removes it by LOCALIZING the mark to a mask
and filling that mask with the chosen ``--backend`` (auto: MI-GAN if the 'migan'
extra is installed, else cv2). ``--mark auto`` removes every detected mark in one
and filling that mask with the chosen ``--backend`` (auto: best available, LaMa >
MI-GAN > cv2). ``--mark auto`` removes every detected mark in one
pass. For arbitrary logos/objects, use ``erase``.
"""
from remove_ai_watermarks import watermark_registry as registry