H3 via the Tokenator API
About H3
MiniMax H3 — это облегченная модель генерации видео с открытыми весами от MiniMax. Она разработана для точного многомодального редактирования и контролируемой генерации контента, включая редактирование по инструкциям, рендеринг текста и брендов, а также перенос движения из видео в видео. Модель подходит для коммерческих творческих рабочих процессов в рекламе, электронной коммерции, играх и дизайне интерфейсов, с нативным аудиовизуальным выводом для генерации на основе референсов.
Specification
| Full name | H3 |
| API ID | hailuo-3 |
| Model vendor | MiniMax |
| Type | Video generation |
| Clip duration | 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 seconds |
| Resolution | 768p |
| Aspect ratio | 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 |
| Audio | yes |
| Reference | accepted — you may attach an image, but the model works without one |
| First frame | accepted only together with the last frame |
| Last frame | accepted only together with the first frame |
| Billing | flat, 35 generations per clip: duration, resolution and aspect ratio do not change it |
| Defaults | 5 seconds, 768p, 16:9 |
| Generation time | up to 15 min |
| Supported API formats | /v1/videos/generations |
| Current status | currently unavailable |
| Data updated | 2026-10-13 |
Example request
curl https://api.tokenator.top/v1/videos/generations \ -H "Authorization: Bearer sk-your-tokenator-key" \ -H "Content-Type: application/json" \ -d '{ "model": "hailuo-3", "prompt": "a neon city in the rain, slow camera fly-through", "duration": 5, "resolution": "768p" }'
Compatible tools
The model can be used in any tool that accepts a custom base URL: Claude Code, Codex CLI, OpenCode, Cursor, Cline, Kilo Code, Cherry Studio, DeepSeek Harness, Hermes, Pi, OpenClaw.
See also
FAQ about H3
What is the API ID of H3?
hailuo-3 — put this into the model field of your request.
How do I request a video?
POST to /v1/videos/generations with model and prompt. The same key and the same base URL as for text models — only the path changes. The answer comes back when the clip is ready, with a link to the mp4 in data[0].url. The full request description, the parameters and curl, Python and Node.js examples are in the video generation documentation.
Which parameters does H3 accept?
prompt — the scene description, duration — 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 seconds, resolution — 768p, aspect_ratio — 21:9, 16:9, 4:3, 1:1, 3:4, 9:16, generate_audio for sound, frame_images to animate an existing frame. Values outside this list are refused with a clear error, and anything you leave out is filled from the model's defaults.
What does one clip cost?
Video is not billed in tokens but in generations from a video bundle: flat, 35 generations per clip: duration, resolution and aspect ratio do not change it. A 15-second clip costs the same as a 4-second one. The remaining generations are shown in your account and in the key dashboard.
How long does it take?
A render takes from tens of seconds to a few minutes. The connection is held with a keep-alive, and if the model answers with an async job the proxy waits for it and hands back the finished clip — you do not need to poll anything yourself.
What if a render fails?
The claimed generations are refunded automatically — a failed render costs nothing. If it keeps failing, try another aspect ratio or a shorter clip: not every combination is supported by every provider.
Can I try it without writing code?
Yes: the Studio tab in your key dashboard has an Images / Video switch where duration, resolution and aspect ratio are picked with buttons. Video bundles are on sale in your account.