Turn text, first and last frames, or reference images, video and audio into video with MiniMax H3.
Reference images (optional)
Reference videos (optional, up to 15s in total) β its length is billed on top of the output
Reference audio (optional, up to 15s in total)

MiniMax H3 gives you a browser-based studio for short AI video. Write a prompt, add a first and last frame, or upload reference images, a video and audio, and the page picks the right generation mode for what you provided. It runs online, so there is nothing to install.
Describe the subject, the action, the camera and the mood, choose an aspect ratio of 16:9, 9:16, 1:1, 4:3, 3:4 or 21:9, and MiniMax H3 renders a clip of 4 to 15 seconds. Longer prompts with a timeline of beats work well for planned shots.
Add a first frame to animate a picture, and add a last frame as well to define how the shot ends. The model then moves between the two images. When you provide frames they define the shape of the video, so the aspect ratio option is hidden.
Upload up to 5 reference images, one reference video and one reference audio file to guide a generation. Use them to keep a character consistent, borrow camera movement from a clip or match a piece of audio. Reference videos can be MP4 or MOV files of up to 50MB and 15 seconds.
Choose 768P for quick drafts and 2K when you want more detail. The total credits for your choices are shown on the Generate button before you start.
MiniMax H3 follows detailed prompts well, so it can help to describe a clip as a short timeline: what happens at the start, in the middle and at the end, plus the camera movement and the lighting. When you use references, say what each one is for, for example which image is the character and which clip supplies the camera movement. Keep the prompt within 5,000 characters.
The credits for a clip are taken when it is submitted and shown on the Generate button beforehand, including the length of any reference video. If a video fails it is retried automatically once without extra charge, and if it still fails the credits go back to your balance. You can leave the page while a video renders and come back to it.
MiniMax H3 is a flexible choice for planned shots. Use text to video for a new idea, first and last frames for a transition between two pictures, and reference mode for a character that has to stay consistent or a shot that should borrow movement from a clip.
MiniMax H3 is billed per second: 19 credits per second at 768P and 30 credits per second at 2K. If you add a reference video, its length is billed at the same rate as the output. A 5 second text to video clip costs 95 credits at 768P or 150 at 2K.
The short version: clips of 4 to 15 seconds, 768P and 2K, six aspect ratios, up to 5 reference images (10MB each), one reference video (50MB, 15 seconds) and one reference audio file (15MB, MP3 or WAV). Credits for failed generations are returned automatically.
MiniMax H3 sits in the same workbench as the image models and the other video models. Create a character with an image model, then use it as a reference or first frame here. Credits are shared across models, and video models unlock with any credit pack or subscription.
Describe the video. Optionally add reference images, a reference video and reference audio, or switch to first and last frame mode to add start and end images.
Pick 768P or 2K, set a duration between 4 and 15 seconds, and choose an aspect ratio if you are not using frames. The Generate button shows the total credits, including any reference video length.
Click Generate and wait a few minutes. The MiniMax H3 video appears below the controls, where you can play it and open it to download. If a generation fails, the credits are returned.