Choose a source clip
Start with a short shot where the person is easy to identify. Clear faces, limited occlusion, and a simple camera move make the replacement easier to evaluate.
MiniMax H3 Recast · A practical guide
MiniMax H3 RecastReplace the person.
Keep the performance.
MiniMax H3 Recast uses reference photos to replace people in an existing video. Start with the shot you already have, then change the cast while aiming to retain its movement, camera, and timing.
Vismuse opens a general video creation workspace. For the dedicated H3 Max Recast endpoint, see the access guide.
THE SHORT ANSWER
What is MiniMax H3 Recast?
Recast is for a specific task: replacing a person in footage with a character supplied in a reference photo. It differs from generating a new scene from text, and from animating a still image.
The H3 Max Recast model on fal is designed around the source shot: its movement, camera, edit cuts, and audio. Treat those as preservation goals and inspect the output, rather than expecting a frame-for-frame copy.
If you are asking, “Can I replace someone in a video using a photo without rebuilding the scene?”, this is the workflow to investigate.
From source clip to new cast
How to replace a person in a video
Prepare the inputs first. A clear replacement target matters more than a long prompt.
Prepare the new character
Use a sharp reference photo with even lighting. Include the hairstyle and wardrobe you want to transfer; a full character replacement may change more than the face.
Define who changes
On a dedicated Recast service, supply the video and reference photos. For multiple people, state which person each photo replaces. Describe people by their position or clothing.
Review the whole shot
Compare the output with the source at the beginning, middle, and end. Check identity, hands, clothing, motion, camera framing, background details, and sound before keeping a take.
Example replacement instruction
Replace the person in the dark jacket with the character in Reference Photo 1. Keep the walking pace, gestures, camera path, scene layout, and shot timing consistent with the source. Leave the other people unchanged.
Adapt the target description to your clip. For a simple single-person shot, the dedicated endpoint also supports an optional prompt.
Choose the right kind of change
Recast vs. face swap vs. motion reference
| Workflow | What changes | When to use it |
|---|---|---|
| Face swap | Mainly facial identity | You want to keep the existing body and outfit. |
| Character recast | The person’s broader appearance, potentially hair and clothing | You want a different cast in an existing shot. |
| Reference-based generation | A new video guided by images, movement, and a written brief | You want a new scene or a creative reinterpretation. |
These are workflow distinctions, not a claim that every tool preserves the same details.
Online access & API
Where can you use MiniMax H3 Recast?
Run the specific model on fal
The minimax/h3-max/recast playground accepts a source video and reference photos. Developers can use the same endpoint through its API.
- Published source length: 5–30 seconds; no individual shot over 15 seconds.
- Published output options: 768p and 1080p.
- Multiple references: clarify which person each photo replaces.
Create from a prompt and character images
Use Vismuse’s AI Video Generator to develop a character-led video from a written brief and image references. Upload character images, then describe the action, camera movement, and visual style you want.
This is a general video generation workflow. It does not promise the dedicated endpoint’s source-audio retention or exact scene preservation.
Open AI Video GeneratorHow much does Recast cost?
Checked October 11, 2026: fal’s model page lists $0.30 per output second at 768p and $0.45 at 1080p. A 10-second output would be $3.00 or $4.50 at those rates. Verify the current rate before running a job; these are fal prices, separate from Vismuse credits.
Questions, answered
MiniMax H3 Recast FAQ
What is MiniMax H3 Recast?
It is a video character replacement workflow that takes existing footage and reference photos of the new cast. The H3 Max Recast endpoint on fal is designed to retain the source performance, camera movement, cuts, and audio while changing the people on screen.
Where can I use MiniMax H3 Recast online?
fal provides a browser playground and API for minimax/h3-max/recast. This guide links to that dedicated endpoint. The Vismuse button opens our general AI Video Generator for reference-based video creation.
Does Recast replace the whole person or just the face?
Recast can change a character's broader appearance, including hair and clothing. A traditional face swap typically concentrates on facial identity while keeping the original body and wardrobe. Choose based on how much of the person you want to change.
Will the original background and movement stay exactly the same?
The dedicated endpoint is designed to preserve the source shot and motion, but that is not a guarantee of identical pixels. Compare the generated frames with the original, especially around hands, fast movement, overlaps, and detailed backgrounds.
Can I replace more than one person?
The published fal workflow accepts a reference photo for each new person. Its default mapping follows the main people from left to right; an optional instruction can clarify a different mapping. Keep each person's identity and reference role explicit.
Does it keep the original audio?
fal documents source-audio preservation for its dedicated H3 Max Recast endpoint. Audio behavior depends on the workflow you actually run; do not assume a general reference-video generator will copy the original soundtrack.
Is MiniMax H3 Recast free, and does it have an API?
fal offers API access and charges for generated video. At our October 11, 2026 check, its model page listed $0.30 per output second at 768p and $0.45 at 1080p. Check the live model page before running a job. Those are fal rates, not Vismuse prices.
Can I use Vismuse for the same workflow?
Vismuse's AI Video Generator creates video from text prompts and image references. Describe the movement and camera direction in your brief. It is a separate workflow from the dedicated fal Recast endpoint described here. Use it to create a new character-led video, and review how closely the result follows your references.
Sources & further reading
Model behavior and published limits were checked against fal’s model page and tutorial on October 11, 2026. This page explains the documented workflow; it is not a completed generation benchmark.
Your next video starts here
Give your character a scene worth stepping into.
Bring a clear brief and the references that define your idea.