MiniMax H3 input limits, formats, and file requirements

MiniMax H3 accepts several media types, but every input has a technical boundary. Checking the MiniMax H3 input limits before upload prevents a creative review from being interrupted by the wrong codec, an oversized file, an unsupported ratio, or an API body that is too large.

The supplied launch manual separates first and last frame input from Omni Reference. The two entries serve different workflows and do not share the same file counts.

First and last frame requirements

First and last frame mode accepts zero, one, or two images. Zero images means text to video. One or two images guide the opening, ending, or both, depending on the selected controls.

Image width and height must each fall between 256 and 5,760 pixels. The supported image aspect-ratio range is 5:2 through 2:5. Output follows the original ratio of the frame image, so prepare both frames with matching dimensions when using two.

Omni Reference requirements

Omni Reference accepts up to nine images. It also accepts up to three video clips, with each clip lasting 2 to 15 seconds and all video references totaling no more than 15 seconds.

Up to three audio clips can be included. Each audio clip must also last 2 to 15 seconds, and their combined duration cannot exceed 15 seconds. Audio cannot be used by itself; at least one image or video must be part of the request.

The mixed request limit is 12 files. Per-type maximums do not increase that total. Nine images, three videos, and three audio clips would equal 15 files, so that full combination is not allowed.

Supported formats and codecs

The manual lists H.264/AVC and H.265/HEVC for video. Audio inside a video may use AAC or MP3. Supported still-image formats are JPG, JPEG, PNG, WEBP, HEIC, and HEIF. Standalone audio can be WAV or MP3.

File-size limits apply per item: 50 MB for video, 30 MB for images, and 15 MB for audio. The manual says the combined upload is not capped by one shared media limit, but API request bodies are limited to 64 MB.

For API workflows, URL input is the recommended approach. Host each source on a controlled HTTPS endpoint, verify that it can be read without an interactive login, and preserve the exact file served during the generation run.

Prompt and output limits

The prompt can contain up to 7,000 characters. H3 output duration ranges from 5 to 15 seconds at 24 FPS. Text-to-video and Omni Reference support 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16; Omni Reference also offers automatic ratio selection.

A preflight that catches most failures

Record filename, media role, codec or format, dimensions, duration, file size, and hosted URL before submission. Confirm the total number of files and the combined duration for video and audio references. Finally, test every URL from a clean session.

Frequently asked questions