Video Watermark Remover

Video Watermark Remover

Remove watermark/logo/text overlays from videos using ffmpeg's delogo filter.

Core Logic

Step 1: Extract sample frames Extract frames at 1-second intervals to locate watermark positions:

for t in 1 2 3 4 5; do
  ffmpeg -y -ss 00:00:0$t -i "$INPUT" -frames:v 1 "/tmp/wm_frame_$t.jpg" 2>/dev/null
done

Step 2: Analyze with vision model Use the image tool to analyze the sample frames. Ask for:

• Exact watermark positions (which corner/edge)
• Pixel bounding box: x, y, width, height for a WxH resolution video
• Whether watermarks are in the same position across all frames (fixed) or move (dynamic)

Step 3: Build segmentation strategy

Fixed watermark (same position throughout):

• One delogo pass on entire video

Dynamic watermark (position changes at specific times):

• Identify segment boundaries (e.g., 0-3.5s bottom-right, 3.5s+ left edge)
• Process each segment separately with its own delogo coordinates
• Concatenate segments back together

Step 4: Process and verify

# Single segment
ffmpeg -y -i "$INPUT" -vf "delogo=x=$X:y=$Y:w=$W:h=$H" -c:a copy "$OUTPUT"

# Multiple segments
# Process each segment separately with -ss/-t, then concat

Verify each segment's result with the vision model before final delivery.

Key Parameters

Finding exact coordinates:

• Vision model gives normalized coordinates on 0-1000 scale — convert to actual pixels
• For a 720x1280 video: actual_x = normalized_x * width / 1000
• Always add 5-10px padding to the detected region to ensure full coverage

Dynamic Watermark Segmentation

When watermark moves between distinct positions:

1. Identify transition points — Note which frames have watermarks at which positions
2. Create segments — Each position change = new segment boundary
3. Process each segment — Apply correct delogo coordinates to each time range
4. Concatenate — Use ffmpeg concat with a manifest file:

   file 'seg1.mp4'
   file 'seg2.mp4'
   

   ffmpeg -y -f concat -safe 0 -i concat.txt -c copy output.mp4
   

Example segmentation:

Handling Multiple Watermarks

If multiple watermarks exist simultaneously (e.g., corner logo + username handle):

• Identify which the user wants removed vs. preserved
• Apply multiple delogo filters in the same vf string:

  -vf "delogo=x=540:y=1195:w=165:h=55,delogo=x=30:y=480:w=160:h=210"
  

Output Delivery

• Save to /root/.openclaw/workspace/downloads/
• Send via message tool with media parameter
• Confirm with user before final delivery

Limitations

• delogo fills the region with a blurred/averaged patch — works best on simple/static backgrounds
• Semi-transparent watermarks, complex AI-generated watermarks, and moving watermarks over textured backgrounds may still show traces
• For AI inpainting (natural content-aware fill): this skill cannot do it — inform the user and suggest inpainting pipeline if delogo result is unsatisfactory