MiniMax H3 Max, ready to run
Turn one prompt into H3 Max video
H3 Max is the post-trained MiniMax H3 that fal tuned for prompt adherence and aesthetics. Describe a shot or drop in an image, pick a length, and get 768P video back in seconds.
Creating an account is free. Generation is charged per second of output, and a render that fails returns its credits.
H3 Max video generator
Same controls as the model API, minus the API key.
Input
Post-trained for prompt adherence. Up to 768P.
H3 can rewrite your prompt before rendering. Off keeps your wording exactly.
This run costs 80 credits
16 credits per second at 768P
Result
Your video will appear here.
Made with H3
Five clips rendered with the same models this page runs on. Hover to play.
What is MiniMax H3 Max?
H3 is MiniMax’s video model. H3 Max is the variant fal post-trained on top of it, tuned for stronger prompt adherence and better aesthetics, and served on their own inference stack for higher throughput. In practice it follows a written shot more closely than stock H3 at the same resolution, which is why it is the default here.
H3 Max or H3?
H3 Max tops out at 768P and costs more per second there. Stock H3 is cheaper at 768P and is the only way to reach 2K or 4K. Pick H3 Max when the wording of your prompt matters; pick H3 when you need the resolution.
H3 Max at a glance
- Length
- 5 to 15 seconds
- Resolution
- 480P or 768P
- Inputs
- Text prompt, or an image with an optional end frame
- Aspect ratios
- 21:9, 16:9, 4:3, 1:1, 3:4, 9:16
An AI video generator that stays out of the way
Most AI video generators bury the model behind presets and a subscription wall. This one exposes exactly what the H3 API exposes: your prompt, the length, the resolution, and whether the model may rewrite what you wrote. You pay per second of output and nothing else.
Every H3 parameter, unhidden
Duration, resolution, aspect ratio, prompt expansion, seed and end frame — the same fields the model takes.
Pay per second, not per month
Credits are charged on output seconds and returned in full when a render fails.
Prompts in English or Chinese
H3 reads both. The interface switches with one click and remembers the choice.