Continuity check
Find the consistency breaks in your AI-generated video
A score, a timeline of every break with the frames as evidence, and a written fix for each one.
- Max upload
- 5 GB
Drop a video here, or click to pick one
MP4 · MOV · WEBM · up to 5 GB
How it works
- 01
Drop it in
MP4, MOV or WebM. Up to 5 GB.
- 02
It watches
The whole thing, shot by shot.
- 03
It compares
Every later shot against the first one.
- 04
You get a list
A score, marks on the timeline, and what to change.
What it compares things to
The first time you see a person, a prop or a room clearly, that is the version everything later has to match. Not the shot before it, because then the mistakes just pile up.
What it looks for
Six kinds of thing go wrong. Each one it finds gets a mark on the timeline, a timecode and two frames.
- Stuff
- Clothes, hair, props, what is on the table.
- Space
- Which way people face, and where they stand.
- Time
- Time of day, the light, the weather, the shadows.
- Action
- Poses, hands, what someone is holding across a cut.
- Camera
- Sudden jumps in exposure, colour or sharpness.
- AI weirdness
- Faces drifting, twins appearing, things teleporting.
Where the score comes from
Nobody is guessing a number. The analysis lists what it found, and the score is worked out from that list in code: bigger problems cost more, and a long clip is not punished for being long. So the same findings always give the same score. Every card also says how sure it is. Read the method.
Clips it has already read
Three walkthroughs: the clip, the score, the shot that broke and the fix that was written for it, including one run where nothing broke. Read the case studies.
Who ends up using it
People checking generated shots before they cut them together, anyone making episodes, and people who build these models and want to compare two versions on the same clip. The same check scores the public benchmark.
From the notes
Continuity errors in film, and why AI video makes so many more of them
A hundred years of film built a whole job around remembering where the glass was. Generated video threw that away and gets the same errors at ten times the rate. Here is what changed and what to do about it.
Read it