lockstep: one photo now recasts anyone in a video
each::labs released lockstep, a model that swaps the people in a video from one photo each and keeps the motion, lip sync and camera.
what it is
each::labs, a platform that serves lots of image, video and audio models through one api, released its own video model: lockstep.
it does one job. it swaps the people in an existing video for characters you supply, and keeps the original motion, lip sync and camera work.
how it works
per each::labs:
- one source video
- one to eight reference photos, one per character
- an optional prompt, where you tag each photo and say who replaces whom
- output at 480p, 720p or 1080p, same length and timing as the source
it's available in their playground, through the api and inside workflows.
the claims
each::labs calls it the best recast model on the market and says it costs a fraction of what others charge. those are their claims. we haven't seen independent tests.
we tried it
we ran one test ourselves: a 15 second clip at 720p with one swap.
it cost $1.35 and took about 13 minutes per run. the first run failed with no output.
why it matters for a creator business
recasting existing footage is how a lot of short-form content gets made. keeping the timing right has always been the hard part.
the daily post, by email. nothing else.
one email a day max. unsubscribe anytime.