Watermark video with FFmpeg: logo overlay, transparency, and broadcast-bug positioning

When to use this

You watermark video for brand recognition (broadcast bug, logo bug), copyright assertion (anti-piracy visible mark), or pre-release content tracking (different watermark per recipient = forensic identification). The visible watermark trade-off: it permanently alters the frames, costs encoder compute, and can't be removed without re-encoding from a clean source. For piracy tracing without visible alteration, see forensic invisible watermarking (out of scope here — vendors like NexGuard, Witbe, Verimatrix specialize).

Command variants

ffmpeg -i input.mp4 -i logo.png \
  -filter_complex "[1:v]scale=120:-1[wm];[0:v][wm]overlay=W-w-20:20" \
  -c:a copy \
  output.mp4

ffmpeg -i input.mp4 -i logo.png \
  -filter_complex "[1:v]format=rgba,colorchannelmixer=aa=0.5[wm];[0:v][wm]overlay=W-w-20:20" \
  -c:a copy \
  output.mp4

ffmpeg -i input.mp4 -i logo.png \
  -filter_complex "[1:v]format=rgba,colorchannelmixer=aa=0.4[wm];[0:v][wm]overlay=(W-w)/2:(H-h)/2" \
  -c:a copy \
  output.mp4

ffmpeg -i input.mp4 -i logo.png \
  -filter_complex "[1:v]format=rgba,colorchannelmixer=aa=0.5[wm];[0:v][wm]overlay=W-w-20:20:enable='between(t,0,10)'" \
  -c:a copy \
  output.mp4

What each parameter does

• overlay=W-w-20:20

  Position formula. W = main video width, w = overlay width. (W-w-20) places overlay 20px from right edge. The second argument is Y position (20px from top).

• scale=120:-1

  Scale logo to 120px wide, height auto-calculated to preserve aspect ratio. Use -1 for auto on either dimension.

• format=rgba,colorchannelmixer=aa=0.5

  Convert to RGBA + scale alpha channel by 0.5 = 50% opacity. The standard semi-transparency idiom.

• enable='between(t,0,10)'

  Watermark only between seconds 0 and 10. Useful for time-limited overlays.

• -c:a copy

  Audio passes through unchanged. Watermarking is a video-only operation.

What this outputs

A re-encoded video file with the watermark permanently rendered into the frames. File size and quality match the encoder settings (defaults to libx264 medium-preset). The watermark cannot be removed without re-encoding from a clean source.

Pitfalls

1. Logo PNGs without alpha channels render with a white box around the logo — always use transparent PNGs for proper compositing.
2. Logo too small at high resolution (scale=120 looks tiny on 4K) or too big at low resolution (looks oversized on 360p). Calibrate logo size relative to source resolution.
3. Forgetting -c:a copy re-encodes the audio along with video, which doubles the compute cost. Always stream-copy audio for watermarking.
4. Filter syntax is whitespace-sensitive: spaces in `[1:v] scale=120:-1 [wm]` vs `[1:v]scale=120:-1[wm]` matter. Stick with the no-space convention.
5. enable expression syntax requires single quotes around the expression and proper escape if used in shell scripts.
6. Re-encoding always introduces some quality loss. If you watermark video that was already encoded, you encode it twice — quality compounds.

At production scale

Watermarking is libx264-bound at production scale. NVENC GPU-accelerated watermarking works but has occasional alpha-blending artifacts at edges. For broadcast-grade output, stick with libx264 + filter_complex. At >100K minutes/month with watermarking as a standard step, the compute cost roughly doubles vs unwatermarked encoding. Consider conditional watermarking: only apply when the customer's use case requires it (e.g., screener delivery vs final master).

MpegFlow models watermarking as a parameterized DAG stage so the watermark file, opacity, position, and timing are per-job inputs. This lets you watermark differently per recipient (forensic tracing) without rebuilding the pipeline per customer.

• FFmpeg
• watermark
• overlay
• video-operations
• Broadcast

• Recipe
  Burn Captions Into Video
• Architecture
  Broadcast Grade Vod Transcoding