Last week a user emailed me a field report. He had run claude-real-video — my open-source tool that turns a video into something an LLM can actually read — over his entire photo library: 2,181 videos in about four days. Then he sent me the bug list, worst first.
The worst one was a design flaw I had been shipping for months without noticing.
The blind spot
Frame dedup compares downscaled frames and drops a frame when too few pixels changed. Sounds reasonable — until the thing that matters is small in the frame.
A person filmed at phone-camera distance covers roughly 0.5% of the picture. Whatever they do, they can never change 8% of the pixels. So percentage-based dedup calls the crucial second "a duplicate" and deletes it. On his repro clip — a static shot where a vehicle knocks someone down in about one second — extraction produced 69 frames and dedup kept 4. The analysis described "a vehicle passes close to the camera" and missed the incident entirely. Setting the threshold to zero did not help. The math is structurally blind.
The fix
Percentages cannot see small subjects, so 0.7.16 adds a third check that ignores percentages: if a handful of cells change hard, the frame stays. That is the whole idea. On the same repro, the action now survives 10/10 frames and a vision model narrates the event correctly. On normal footage the kept-count barely moves, so you do not pay extra for it.
Everything else he reported — a crash on non-UTF-8 metadata, a flag name that means the opposite of what it says, 68 GB of intermediates piling up silently — shipped the same day, also in 0.7.16.
Try it
If you use Claude Code (Codex, Cursor and Gemini CLI work too):
npx skills add HUANGCHIHHUNGLeo/claude-real-video
Claude Code plugin marketplace:
/plugin marketplace add HUANGCHIHHUNGLeo/claude-real-video
/plugin install claude-real-video@claude-real-video
Or run it directly:
pip install claude-real-video
crv "your video URL or file"
Here is a 60-second demo:
日本のユーザーの方へ
日本からの利用者が増えてきて、うれしいです。インストールは pip install claude-real-video、そのあと crv "動画のURLまたはファイル" を実行するだけです。ご質問は日本語でも大丈夫です。
The takeaway
A bug list this long only comes from mileage. If someone runs your tool 2,181 times and writes down everything that broke, that is not criticism — that is the roadmap. Treasure those users.
Top comments (0)