Multi-reference AI video is a generated clip composed from several images at once. You hand the model a face, a product and a set, and it keeps all of them across the take. Each model publishes its own ceiling on how many it will read, and the cards below carry that number straight from the platform catalog. Models that read one image only, or that read two as the first and last frame of a shot, are a different job and are not on this page.
Choose from the roster above. The card shows its cap.
02
Attach the set
Drop in the face, the product and the location.
03
Write and render
The exact price sits on the button. Press it once.
One shot, several sources
The hard part of generated video was never the first clip. It was the second one, with the same face in it. Reference images are how that got solved: you hand the model the material instead of describing it, and the take is built from what you gave it. What this page adds is the shortlist. Instead of opening every model page to find out which ones read more than one image, you read one roster that the platform catalog writes.
The cap is per model, and it is the wire's own
Every model here publishes its own ceiling and each card prints that model's number. The same figure gates the composer's attach controls and the spec sheet on the model page, so those surfaces cannot disagree. Length ceilings and resolutions come from the same feed. If a vendor widens the envelope, this page widens with it.
References, not frames
A model that takes two images because a shot has a first frame and a last frame is not doing this job, and carding it here would sell the wrong thing. The roster reads the roles each model publishes for its attachments and keeps the frame-only models off the page. Where a model names no roles at all, its images are references and it is listed.
The price is visible before you spend
Each card carries a floor price in credits and the same figure in dollars at the pay-as-you-go pack rate. Credits sell at several rates and plans buy them cheaper, which is why the dollar figure says which rate it came from. The binding quote is still the one on the Generate button, computed for the exact length and tier you picked.
Questions
Frequently asked questions
The roster on this page is the answer, and it is read from the platform catalog rather than written down. Every model listed accepts more than one reference on a single render, and each card names the vendor behind it, the cap it publishes, the longest clip it will carry and the price of a render. A model whose vendor raises that cap shows the new number here on the next catalog refresh.
It depends on the model, so each card states its own ceiling and no page-wide number is claimed. The figure is the maximum the wire itself accepts, read from the same feed the composer reads before it lets you attach. Go over it and the attach controls stop you in the composer rather than at dispatch, so a full set never costs you a failed render.
No, and this page keeps them apart. A first frame is the opening picture of the shot, and a first-and-last pair is the two ends of one move. A reference is material the model composes from: a face to keep, a product to place, a look to match. Models that publish only frame roles are doing the first job and are not carded here, even where they accept two images.
That is what they are for, and the model page for each one states its own behaviour rather than a shared claim. Several angles of the same face give the model more to hold onto across a take than one photograph does. Read the model page before you plan a shoot around a specific character, since vendors differ on what they keep and what they reinvent.
Free to try. A new account gets a one-time credit grant and no card is asked for. Packs start at $10/800⚡. Video costs more than stills, so the grant is there to judge the output rather than to fund a campaign.
