Description

An AI video generator built on MiniMax H3, at [h3max.pro](https://h3max.pro).
A prompt or a single image goes in; 5 to 15 seconds of video comes back.

## The models

Four fal endpoints, two models, two input modes each:

| Model | Endpoints | Resolutions | Notes |
| ------ | -------------------------------------- | ------------------ | ----------------------------------------------------------------------------- |
| H3 Max | `minimax/h3-max/{text,image}-to-video` | 480P, 768P | fal's post-trained H3: better prompt adherence, costs more per second at 768P |
| H3 | `minimax/h3/{text,image}-to-video` | 480P, 768P, 2K, 4K | Stock MiniMax H3: cheaper at 768P, and the only way to reach 2K or 4K |

Both accept 5 to 15 seconds. Image mode takes an optional end frame.