
A matte ceramic coffee mug on a dark slate table, soft morning window light, slow turntable move, shallow depth of field, photoreal product commercial, no text overlays
Turn a prompt or an image into a MiniMax H3 video up to 15 seconds long, rendered in 2K. Draft for free, then sign in to generate.
Small set of prompts you can send straight into the studio handoff.

A matte ceramic coffee mug on a dark slate table, soft morning window light, slow turntable move, shallow depth of field, photoreal product commercial, no text overlays

Night city street in light rain, neon reflections on wet asphalt, stabilized tracking shot, cinematic color grade, natural motion, no logos

Quiet Japanese tea room, soft shoji light, steam rising from a ceramic bowl, slow push-in, contemplative mood, photoreal, no on-screen text
MiniMax H3 is MiniMax's omni-modal video model. It reads text, images, video and audio as context and generates clips of 4 to 15 seconds at 24 FPS, with 2K output through its H3-Regenerate-2K stage. MiniMax publishes the weights on Hugging Face and also offers H3 in the Hailuo AI app and the MiniMax API.
This site is an independent online studio for MiniMax H3: no local GPU or setup, 2K output, and credit-based generation in your browser.
| Duration | 4–15 seconds (official spec) |
|---|---|
| Frame rate | 24 FPS |
| Resolution | 2K in this studio |
| Aspect ratios | 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 |
| Inputs | Text, images, video and audio references |
Three steps from prompt to video.
Describe the scene and camera, or upload a still to use as the first frame. You can draft before signing in.
Pick one of six aspect ratios and a clip length up to 15 seconds.
Sign in, generate the clip in 2K, then download the MP4.
Four ways to start a clip in the studio.
Four jobs creators use MiniMax H3 for — each maps to a mode in the studio.
Turntable moves, hero reveals and ad beats from a single product still — image to video keeps the pack shot on-model.
Hook-style clips for TikTok, Reels and Shorts in 9:16, generated from a prompt or a reference frame.
Camera moves, mood and lighting written into the prompt — 24 FPS clips for boards, mood films and edits.
Omni reference keeps a face, outfit or product consistent across clips for episodic and brand work.
Write the scene, choose the frame and length, and download the result — all in the browser.
Running H3 locally gives you full control, but you need the weights, a high-memory GPU and the official workflow. The online studio skips setup: open the page, write a prompt and generate in 2K.
Read the local setup guide| Option | Setup & cost | Best for |
|---|---|---|
| This online studio | None — runs in the browser, credits per clip | Creators who want 2K clips without a GPU |
| Local run (open weights) | High-memory GPU, model weights and the official workflow; free per clip after that | Tinkerers who need full control |
| MiniMax API & Hailuo app | Official channels — API key plus integration work, or the consumer app | Developers and casual mobile use |
Writing prompts and choosing settings is free. Generating a video uses credits after you sign in: a 5-second 2K clip costs 130 credits, and each extra second adds 26. See the pricing page for current plans.
View pricingMiniMax H3 is an omni-modal video generation model from MiniMax, with open weights on Hugging Face. It creates 4–15 second videos from text, images, video and audio references, up to 2K at 24 FPS. This site is an independent online studio where you can use it without a local GPU.
Write a prompt or upload an image in the generator at the top of this page, choose an aspect ratio and a length up to 15 seconds, then sign in and generate. The finished clip downloads as an MP4.
Writing prompts and choosing settings is free. Generating a video uses credits after you sign in: a 5-second 2K clip costs 130 credits, and each extra second adds 26. See the pricing page for current plans.
Text to video, image to video (a first frame with an optional last frame), and omni reference, which accepts up to 9 images, 3 video clips and 3 audio clips as references.
The studio renders in 2K. You can pick 21:9, 16:9, 4:3, 1:1, 3:4 or 9:16. In image-to-video mode the frame follows your input image.
Yes. MiniMax publishes the weights on Hugging Face under the MiniMax H3 Community License, with Diffusers support and a local workflow for H3-Base. Local runs need a high-memory GPU; see our download and local setup guide.
It depends on your plan's terms on this site and on MiniMax's license for the model. Review our Terms of Service before you publish generated videos.
Our prompt guide explains MiniMax H3's official three-part prompt structure, camera terms and dialogue tags, and the prompt examples page has ready-to-use starters.
Download and local setup, a getting-started guide, prompt examples and the official model card.
A focused gateway for creators who want MiniMax H3-quality motion without building a full studio stack.