MiniMax H3 Max Reference to Video
minimax/h3-max/reference-to-videoGenerate video from text and reference media with MiniMax H3 Max Reference to Video from fal through one Spiral API key. Makes a video of up to 15 seconds from reference images, videos and audio at up to 1080P; reference videos pass on their scene. It runs in the background and returns video. Pay as you go, with no subscription.
Pricing
$0.104 per second at 768p ($0.065 at 480p, $0.208 at 1080p). Images, video and audio share 4,096 included reference tokens; additional tokens: $0.026 per 1,000. Reference media costs are estimated; actual usage can vary.
Limits
- Runs
- In the background
- Returns
- video
- Longest queue wait
- 1 hour
- Run deadline
- 1 hour
- Longest run, end to end
- 930 minutes
- Largest inline input
- 1 MB
- Largest input file
- 2 GB
- Largest output
- 5 GB
Input
| Field | Kind | Default | Description |
|---|---|---|---|
prompt required | text | — | Text prompt for video generation. Refer to reference assets by their modality and order in the reference lists: Image 1, Image 2, Video 1, Audio 1, and so on. |
reference_image_urls | URLs or files | — | URLs of subject/style reference images, referenced in the prompt as Image 1, Image 2, and so on. Reference images, videos, and audio clips must add up to at most 12 files. |
reference_video_urls | URLs or files | — | URLs of reference video clips (2-15 seconds each, combined duration at most 15 seconds), referenced in the prompt as Video 1, Video 2, and so on. The model can reuse a video's people, setting and on-screen text as well as its motion; to move a person from an image with a video's motion, use a motion-control model. Reference images, videos, and audio clips must add up to at most 12 files. |
reference_audio_urls | URLs or files | — | URLs of reference audio clips (2-15 seconds each, combined duration at most 15 seconds), referenced in the prompt as Audio 1, Audio 2, and so on. Images, videos, and audio can be provided individually or together. Reference images, videos, and audio clips must add up to at most 12 files. |
image_url | URL or file | — | Optional URL of the image to use as the first frame. The video opens exactly on this image while the references keep the subjects consistent; the output canvas follows this image. |
middle_image_url | URL or file | — | Optional image to guide the video at middle_frame_time. Requires start and end images and native 480P or 768P resolution. |
middle_frame_time | number | — | Target time for the middle image, in seconds from the start. Rounded to the nearest frame at 24 fps; it must fall strictly between the first and last frames of the requested duration. Requires middle_image_url. |
end_image_url | URL or file | — | Optional URL of the image to use as the last frame. The video ends exactly on this image. |
duration | number 0.92 to 15 | 5 | Video length in seconds. The output can run up to about 0.7 s longer than requested. |
resolution | one of 480P, 768P, 1080P | 768P | The native generation resolution, or 1080P latent refinement from a native 768P source. |
seed | number (whole) | — | Random seed. A random seed is selected when omitted. |
prompt_expansion_mode required | text | balanced | How much effort to spend rewriting the prompt before generation. 'disabled' skips prompt expansion. 'balanced' returns in about a second. 'quality' spends up to ~30s on a richer prompt. |
aspect_ratio | one of adaptive, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 | adaptive | The aspect ratio of the generated video. |
reference_conditions | JSON | — | Ordered reference list; cross-modality order is semantic and is preserved end-to-end. Do not combine with reference_*_urls. |
Same request in code
curl https://api.spiral.ai/v1/runs \
-H "Authorization: Bearer $SPIRAL_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: $(uuidgen)" \
-d '{
"provider": "fal",
"model": "minimax/h3-max/reference-to-video",
"input": {
"prompt": "Image 1 is the female protagonist. Image 2 is her small dog. Keep the woman and dog consistent with their respective reference images while they walk together…",
"prompt_expansion_mode": "balanced"
}
}'Frequently asked questions
How much does MiniMax H3 Max Reference to Video cost?
$0.104 per second at 768p ($0.065 at 480p, $0.208 at 1080p). Images, video and audio share 4,096 included reference tokens; additional tokens: $0.026 per 1,000. Reference media costs are estimated; actual usage can vary. You pay as you go from one Spiral balance, with no subscription, and a failed run is not charged.
How do I use MiniMax H3 Max Reference to Video through an API?
Send `POST /v1/runs` with `"provider": "fal"`, `"model": "minimax/h3-max/reference-to-video"` and the input fields on this page, using one Spiral API key. It runs in the background: poll the run, or name a webhook to be told when it finishes.
Do I need an account with fal to use MiniMax H3 Max Reference to Video?
No. Spiral runs it on its own fal account and charges your Spiral balance.
Can my AI agent use MiniMax H3 Max Reference to Video?
Yes. Agents find and run it through Spiral's MCP server, CLI or agent skill, with the same API key.