01
One character, many scenes
A host, mascot, or lead who has to be recognisably the same person in every clip of a series.
Multimodal Video Tool
Reference to Video
Keep the same character when the scene changes
Upload one to three references and the subject in them carries into the shot: face, clothing, proportions. Then say where they are and what happens. The references handle who it is, so the prompt does not have to.
Creative workspace
Reference to Video
50
What is Reference to Video?
Reference to video is generating a clip from example images rather than from a prompt alone. You supply reference pictures of a character, a product, or a look, and every shot the model makes keeps them recognisable.
How the people who built the model describe it:
“fal's H3 Max is a post-trained variant of MiniMax H3, tuned for stronger prompt adherence and better aesthetics while co-optimized with our custom inference stack for higher throughput with no compromises on output quality”
Reference to Video at a glance
- You bring
- 1-3 references
- They fix
- Who and what it looks like
- The prompt fixes
- Where and what happens
What you can make with the Reference to Video tool
These are the jobs people actually bring to Reference to Video, and the settings each one needs.
02
One product, many settings
The same bottle, shoe, or device in a kitchen, on a street, in a studio, without rebuilding the shot each time.
03
Several stills, one shot
Take the subject from one frame and the place from another, and let the motion between them be generated.
Reference to Video
A workspace shaped for the model
Reference to Video puts the handful of controls that change the result most in front of you, and keeps the rest out of the way.
01
More angles, better likeness
Two or three references of the same subject beat one. A single frame shows the model one side of a face, and it invents the rest. The extra views are what stop it.
02
Describe the scene, not the subject
The references already carry identity, so the whole prompt can go on place, action, camera and light.
03
Not the same as image to video
Image to video animates the picture you gave it, framing and all. This takes only the subject out of it and puts that somewhere else.
04
Sound arrives with the shot
Ambience, effects, and dialogue are generated alongside the picture rather than added to it afterwards.
05
Close, not identical
Likeness is strong but not guaranteed. It drifts first on faces far from camera, unusual angles, and fine detail like text on a logo. Plan on a second run rather than a surprise at delivery.
Reference to Video
Choose with context
Where Reference to Video is the right call, and where a different model does the job better.
Starts from
A subject you need to keep
The prompt decides
Everything except who
Pick this when
The same face or product has to appear again
How to keep a subject consistent across shots
01
Upload one to three references
Same subject, different angles, clearly lit. Three views of one face work better than three different faces.
02
Write the scene around them
Place, action, camera move, sound. Leave out what they look like; that is what the references are for.
03
Run it, then run it again
Likeness varies between runs. Keeping the best of two or three costs less than trying to describe your way there.
04
Reuse the same references
For a series, keep the reference set fixed and change only the prompt. That is what makes episode four look like episode one.
Questions about reference to video
Start Creating With Reference to Video
Set up the brief now, then continue in the shared Studio workspace.