Veo real-provider reference clip
Copyright 2025-2026 wiltodelta
Licensed under the Apache License, Version 2.0. See LICENSE.txt.

Source: wiltodelta/remove-ai-watermarks, commit
540d5648cc271110439be8aeaf6626cc4d909b4e
data/fixtures/visible/veo/provider-original.mp4
The source maintainer documents a Google Flow / Veo 3.1 Fast generation and
expressly distributes this generated output under the repository license.
The original clip was not generated by videowatermarkremover.app or Gemini App.
No separate upstream NOTICE file exists in the pinned source tree.

Modifications by videowatermarkremover.app on 2026-10-21:
- A five-second silent H.264 excerpt was prepared from the eight-second licensed
  source for the site's real canvas workflow. The preparation shortened and
  encoded that input and omitted audio.
- applied-mask.png is the exact 1280 x 720 black-and-white PNG exported by the
  site's canvas and applied to the accepted edit: white rectangle x=1235, y=687,
  width=39, height=26. It is not a recreated or illustrative mask.
- result.mp4 is the actual output of this website's masked AI edit, using
  wavespeed-ai/video-eraser through MegaRouter, with input/output review allowed.
- Both MP4s were stream-copied into new containers for public display to remove
  private job-id and other nonessential container metadata. Encoded video packet
  hashes are unchanged from the accepted input and downloaded output. No further
  filtering, retouching, cropping, re-encoding or audio editing was performed.
- input-4s.png and result-4s.png are unretouched, lossless frames extracted at
  4.000 seconds from their respective MP4s.

The observed edit removes the visible wordmark but changes the mug's orientation
and other image details. This is one result, not a quality guarantee. It is not a
free FFmpeg-filter output or an official provider promotional example. SynthID
removal and provenance metadata retention have not been verified.
See SOURCE.md and provenance.json for source evidence and integrity records.