MiniMax H3
H3 (Hailuo-03) LoRA trainers for keyframe, image, reference, and text video-audio modes.
Studio is in early access; availability resolves per project after sign-in.
- Endpoints
- 10
- Eligible
- 6
- Input schemas imported
- 10 / 10
- Delivery
- fal.ai
- Weights
- Not stated
- Developer
- Not stated in source
Overview
Four trainers teach MiniMax H3 video-audio LoRAs: first-last-frame, image-to-video-audio, reference-to-video-audio, and text-to-video-audio. The keyframe trainer covers all conditioning signatures in one run with per-mode probability knobs. Joint audio training maps clips with soundtracks to matching audio and silent clips to silence, automatically per clip. Inference image-to-video, reference, and text-to-video endpoints plus LoRAs join the trainers.
Capabilities
- Four-mode trainer set
- Keyframe-signature coverage
- Per-clip joint A/V training
- H3 inference plus LoRA tiers
Best for
- Custom keyframe behaviors
- Audio-matched styles
- H3 specialization
Use cases
- FLF2VA LoRA runs
- Reference-driven adaptation
- T2VA style teaching
Endpoints
4 under review · 6 eligible. Cataloged is not the same as executable: Studio resolves which endpoints a project can run.
| Endpoint | Task | Catalog status | Inputs |
|---|---|---|---|
minimax/h3/flf2v/trainer | unknown | Under review | 17 fields · requires training_data_url |
minimax/h3/i2v/trainer | unknown | Under review | 15 fields · requires training_data_url |
minimax/h3/image-to-video | image to video | Eligible | 10 fields · requires prompt |
minimax/h3/image-to-video/lora | image to video | Eligible | 11 fields · requires prompt, loras |
minimax/h3/ref2va/trainer | unknown | Under review | 16 fields · requires training_data_url |
minimax/h3/reference-to-video | reference to video | Eligible | 11 fields · requires prompt |
minimax/h3/reference-to-video/lora | reference to video | Eligible | 12 fields · requires prompt, loras |
minimax/h3/t2v/trainer | unknown | Under review | 15 fields · requires training_data_url |
minimax/h3/text-to-video | text to video | Eligible | 9 fields · requires prompt |
minimax/h3/text-to-video/lora | text to video | Eligible | 10 fields · requires prompt, loras |
Questions
Which trainers?
Keyframe, image, reference, text video-audio.
Silent clips?
Train against silence automatically.