Video Remove Watermark Gemini

Video Remove Gemini Watermark

Remove Gemini / Veo visible watermarks from video with VeoWatermarkRemover (GeminiWatermarkTool-Video) — the video build of GeminiWatermarkTool. Uses reverse alpha blending, adaptive per-frame intensity, and audio passthrough.

Scope: Gemini diamond, small "Veo" text (Google Flow), and legacy pre-Gemini-3.5 Veo text watermarks. Does not remove SynthID (invisible watermark), arbitrary logos, or non-Gemini marks.

Rules

When this skill applies, read and follow [skill-dependency-manager](../../rules/skill-dependency-manager.md) — run scripts as documented, install missing tools into .dependency/.

• Run remove_watermark.py through .dependency/python/python.exe. The script calls the bundled GeminiWatermarkTool-Video binary only. Never use host python or tools outside .dependency/.
• Do not customize removal. No custom alpha tuning — the wrapper handles paths/batch; pixel processing is delegated to GeminiWatermarkTool-Video.
• populated: false is not a reason to skip. Install first, set populated: true, retry the same command.
• Never overwrite sources. Output goes into a no-watermark/ subfolder beside each source video (same directory as the input file).
• Never copy or move input videos. Pass the user's actual file or directory path (absolute paths are fine).
• Never use Cursor attachment cache paths. Chat-attached files under .cursor/projects/.../assets/ are not the user's source. Use the path the user provides or ask if unclear.

Setup (first run)

From project root, download VeoWatermarkRemover v0.6.4-demo for your platform and extract to .dependency/gemini-watermark-video-tool/:

| Platform | Release asset | |----------|---------------| | Windows | GeminiWatermarkTool-Windows-x64-Video.zip → GeminiWatermarkTool-Video.exe | | Linux | GeminiWatermarkTool-Linux-x64-Video.zip → GeminiWatermarkTool-Video | | macOS | GeminiWatermarkTool-macOS-Universal-Video.zip → GeminiWatermarkTool-Video |

Register in .dependency/manifest.json:

"gemini-watermark-video": {
  "populated": true,
  "bin": ".dependency/gemini-watermark-video-tool/GeminiWatermarkTool-Video.exe"
}

Use GeminiWatermarkTool-Video (no .exe) on Unix. Dependencies: portable GeminiWatermarkTool-Video binary only (separate from the image GeminiWatermarkTool binary).

Quick Start

Default: create /no-watermark/ next to each input file (never overwrites sources):

# Single file — output beside the source
# C:\Users\...\Downloads\clip.mp4 → C:\Users\...\Downloads\no-watermark\clip.mp4
.dependency/python/python.exe .cursor/skills/video-remove-watermark-gemini/scripts/remove_watermark.py C:\Users\...\Downloads\clip.mp4

# Project-relative path
.dependency/python/python.exe .cursor/skills/video-remove-watermark-gemini/scripts/remove_watermark.py video/intro.mp4
# → video/no-watermark/intro.mp4

# Directory batch (flat)
.dependency/python/python.exe .cursor/skills/video-remove-watermark-gemini/scripts/remove_watermark.py video/cutscenes

# Recursive batch
.dependency/python/python.exe .cursor/skills/video-remove-watermark-gemini/scripts/remove_watermark.py video --recursive

Custom output folder name:

.dependency/python/python.exe .cursor/skills/video-remove-watermark-gemini/scripts/remove_watermark.py video/intro.mp4 \
  --output-subdir clean

Shared output directory:

.dependency/python/python.exe .cursor/skills/video-remove-watermark-gemini/scripts/remove_watermark.py video/cutscenes \
  --output-dir out

Detection

GeminiWatermarkTool-Video auto-detects the watermark type per file — no flags needed for current outputs:

| Watermark | Resolution | Notes | |-----------|------------|-------| | Gemini 3.5 diamond | 1080p, 720p (landscape + portrait) | Default profile; standard + compact 720p variants | | Small "Veo" text | 1080p, 720p (Google Flow) | Auto-detected alongside diamond | | Legacy "Veo" text | Pre-Gemini-3.5 | Use --legacy |

Multi-frame probe (12 frames) prevents false SKIP on intro fade-ins. Tick-exact transcoder timing preserves frame count, duration, and FPS.

Options

| Option | Default | Notes | |--------|---------|-------| | Output folder | /no-watermark/ per video | Sibling folder beside each input file | | --output-subdir | no-watermark | Subfolder name beside each source file | | --output-dir | (none) | Shared output root under input path | | --legacy | off | Pre-Gemini-3.5 larger Veo text watermark | | --ml | off | Opt-in ML intensity assist for tricky clips | | --variant | auto | Force 720p-1 (48×48) or 720p-2 (44×44) when auto-detect fails | | --sigma | engine default | AI denoise sigma — lower for anime/illustration (e.g. 15), higher for photo (e.g. 25) | | --json | off | Print structured metadata per video | | --pattern | *.mp4 | Also matches .mkv, .mov, .webm | | --recursive | off | Search subdirectories |

# Default — auto-detect diamond or small Veo text, audio preserved
.dependency/python/python.exe .cursor/skills/video-remove-watermark-gemini/scripts/remove_watermark.py video/clip.mp4

# Pre-Gemini-3.5 legacy Veo text watermark
.dependency/python/python.exe .cursor/skills/video-remove-watermark-gemini/scripts/remove_watermark.py video/old.mp4 --legacy

# ML assist for clips with varying intensity (opt-in since v0.6.4)
.dependency/python/python.exe .cursor/skills/video-remove-watermark-gemini/scripts/remove_watermark.py video/clip.mp4 --ml

# Anime / illustration content (lighter AI denoise)
.dependency/python/python.exe .cursor/skills/video-remove-watermark-gemini/scripts/remove_watermark.py video/clip.mp4 --sigma 15

# Force 720p variant when auto-detect fails (rare)
.dependency/python/python.exe .cursor/skills/video-remove-watermark-gemini/scripts/remove_watermark.py video/clip.mp4 --variant 720p-2

Direct CLI equivalent (single file):

.dependency/gemini-watermark-video-tool/GeminiWatermarkTool-Video.exe -i input.mp4 -o output.mp4

Agent workflow

1. Confirm source — inputs should be Gemini / Veo / Google Flow generated videos with a visible corner watermark.
2. Use the user's real path — e.g. C:\Users\...\Downloads\clip.mp4. Do not use Cursor chat attachment paths under .cursor/projects/.../assets/.
3. Run on the user's path directly — do not copy videos into video/ or other folders.
4. Inspect output — cleaned files appear in /no-watermark/ beside the originals.
5. Skipped videos — exit code 1 means no watermark detected; check --json for applied: false.
6. Residual frames — some clips with heavy motion may leave residue on a few frames; see manual touch-up workflow (ffmpeg decompose → image GWT GUI fix → ffmpeg recompose).
7. Revert — delete the output folder; sources are never modified.

Troubleshooting

| Issue | Fix | |-------|-----| | GeminiWatermarkTool-Video not found | Download v0.6.4-demo release into .dependency/gemini-watermark-video-tool/ | | Watermark not removed (legacy clip) | Re-run with --legacy | | Legacy clip damaged by default mode | Pre-Gemini-3.5 needs --legacy; diamond and legacy text are different profiles | | Auto-detect SKIP on 720p | Try --variant 720p-1 or --variant 720p-2 | | Output too blurry (anime/illustration) | --sigma 15 | | Intensity varies across clip | --ml | | Windows SmartScreen blocks exe | More info → Run anyway, or Unblock-File | | Watermark not removed | Video may not be a Gemini/Veo visible watermark, or unsupported resolution (4K, square, etc.) |

Related

• Script: [scripts/removewatermark.py](scripts/removewatermark.py)
• Video engine: allenk/VeoWatermarkRemover
• Image engine: [image-remove-watermark-gemini](../image-remove-watermark-gemini/SKILL.md)
• Parent project: allenk/GeminiWatermarkTool
• Algorithm write-up: Reverse Alpha Blending (Medium)

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: godot-fun
• Source: godot-fun/godot-framework
• License: MIT

Install and usage instructions live in the source repository linked above.