MiniMax H3 Max is fal’s post-trained version of MiniMax H3. It brings text-to-video, image-led generation, and video continuation into a family of creative workflows, with controls for framing, prompt expansion, and sound.
Post-trained by fal · Built on MiniMax H3
Describe a scene or establish its visual endpoints
Create a focused short scene
Develop the action beyond an existing ending
01 / TEXT-TO-VIDEO
Start with an event, then give it a viewpoint.
H3 Max generates a scene from a written prompt. Describe the subject, its movement, and the camera in the same brief: a red panda follows a falling blossom, shifts one paw, and looks up. A cinematic video starts to take shape when the camera has a clear reason to move.
Prompt expansion can be disabled, balanced, or quality-focused. Keep the key action explicit whichever option you choose.
02 / IMAGE KEYFRAMES
Let the images establish the visual direction.
The image route accepts an opening frame, an ending frame, or both. With an opening image, the canvas follows that image; with an ending image alone, it follows the final frame. For a story video, use the prompt to connect those visual states through a simple, readable action.
Keep the subject, setting, and lighting compatible across the two frames, then describe what changes between them.
03 / VIDEO CONTINUATION
Describe the next beat of the scene.
The extension route takes an existing clip and a prompt for what follows. A puppy can reach the ball, pause, then look toward someone outside the frame. The result can include the source and new footage together, or return the continuation for a short film you are assembling.
Continuation-only output includes 1.625 seconds from the source ending. Account for that overlap when joining clips.
Choose the canvas and delivery size before refining a scene. These settings refer to H3 Max’s text and image generation routes.
DURATION
Up to 15 secondsBuild a short scene around one clear action.
OUTPUT
480p · 768p · 1080p1080p uses latent refinement from a native 768p source.
TEXT CANVAS
Six aspect ratios21:9, 16:9, 4:3, 1:1, 3:4, and 9:16.
SOUNDTRACK
Generate or supply audioAn uploaded soundtrack can replace the output audio.
Choose whether the prompt or an existing image establishes the opening composition.
| Control | Text-to-video | Image-to-video |
|---|---|---|
| Visual starting point | A written scene description | First frame, final frame, or both |
| Canvas | Choose one of six aspect ratios | Follows the supplied keyframe |
| Direction | Describe subject, action, setting, and camera | Describe the movement connecting visual states |
| Useful starting point | A scene you want to explore from scratch | An established character, composition, or ending |
Explore Create workflows for product movement, character moments, and everyday stories.
Give every instruction a role in one focused scene.
01
A sunny coastal meadow. One friend holds a blue kite while the other stands nearby with the spool.
02
She raises the kite into the breeze. The camera tracks sideways as it lifts, then holds on the two friends watching it climb.
03
Wind through the grass, a flutter of kite fabric, and one spoken line: “Ready?” Leave room for the reply.
H3 Max is fal’s post-trained variant of MiniMax H3. Its main routes generate video from text or image keyframes, with a separate extension route for continuing existing footage.
MiniMax H3 is the underlying model. H3 Max is fal’s post-trained version, while H3 Max Turbo is a separate speed-oriented variant. Choose the specific route and settings for the task you need.
Yes. The image route supports a first frame, a final frame, or both. When the first frame is present, its proportions establish the canvas; an ending-only request follows the final image.
fal describes H3 Max output with native audio. The text and image routes also accept a supplied soundtrack: longer audio is trimmed to the video, and shorter audio is followed by silence.
The API offers 480p, 768p, and 1080p. Its resolution description identifies 1080p as latent refinement from a native 768p source.
Yes. The extension route can return the full extended video or the continuation. Continuation output begins with a 1.625-second repeat of the original ending before the new footage.
H3 Max is fal’s post-trained variant of MiniMax H3. Its main routes generate video from text or image keyframes, with a separate extension route for continuing existing footage.
MiniMax H3 is the underlying model. H3 Max is fal’s post-trained version, while H3 Max Turbo is a separate speed-oriented variant. Choose the specific route and settings for the task you need.
Yes. The image route supports a first frame, a final frame, or both. When the first frame is present, its proportions establish the canvas; an ending-only request follows the final image.
fal describes H3 Max output with native audio. The text and image routes also accept a supplied soundtrack: longer audio is trimmed to the video, and shorter audio is followed by silence.
The API offers 480p, 768p, and 1080p. Its resolution description identifies 1080p as latent refinement from a native 768p source.
Yes. The extension route can return the full extended video or the continuation. Continuation output begins with a 1.625-second repeat of the original ending before the new footage.
