Public benchmark

AI video continuity benchmark

One good-looking shot is the easy part now. Keeping the same people, clothes and rooms intact across many shots is where generators still struggle.

Each one gets the same briefs, the same prompts and the same check your own videos get. Then we see who holds it together.

Live. Updates as runs come in.

Two people on a sofa in front of a big screen split into twelve takes of the same woman at a table, her coat and the light changing from take to take — Same brief, twelve takes. The coat changes colour and the light wanders off on its own. That drift is what this table measures.

benchmark_entries · publishedLoading

Method

How this is done

Every generator gets the same briefs. Our agent then checks each video exactly like an upload. Two reads of the whole video run side by side, one for the people and one for the room, every appearance is held against its first clear look, and the score is worked out in code from what came back. Read how.

Run by VideoConsistency. If you built one of these and think a score is wrong, ask for a re-run with the published prompts, or point at the finding and the frames and argue with me.

Every brief has

01The same people coming back across many shots
02More than one location
03Some changes the story explains
04Some changes it does not

Keep reading

Find the breaks and fix them before anyone else watches

Sign in with your email and drop the file in. The report comes back in a few minutes for most videos, with a fix for every break. Plans start at $19 a month for 15 minutes of video and 10 seconds of shot fixes.

Check and fix a video How fixing works