MiniMax H3 Max Video Generation Model

Overview

MiniMax H3 Max is a short-form video generation model built from the MiniMax H3 lineage. It turns a written shot description into a finished video clip with synchronized audio. You can also anchor the motion with a starting image, and optionally lock the ending frame with a second image.

This workflow is useful when you need many variations fast. You can iterate on camera moves, subject actions, lighting, and sound cues without hand animating frames.

What you can build

• Social video concepts and quick ad variations with sound
• Product shots that animate a packshot into a moving scene
• Keyframed transitions that move from a defined first frame to a defined last frame
• Short cinematic b-roll clips for edits and mood boards
• Storyboard-style shot exploration before you upscale or re-render elsewhere

Inputs

• A required written description of the clip you want. Include subject, action, camera motion, lighting, and sound. Put spoken lines in quotes if the speaker is on camera.
• A required clip length in whole seconds, from 5 to 15.
• A required output resolution choice: 480P or 768P.
• An optional first-frame image to start the video from. When you provide one, it sets the opening frame.
• An optional last-frame image to end the video on. This works best when paired with a first frame.
• An optional aspect ratio for text-to-video generations: 21:9, 16:9, 4:3, 1:1, 3:4, or 9:16. If you provide a frame image, the output follows that image’s shape.
• An optional prompt expansion choice that controls how much the system rewrites your description before generation. You can disable it to use your text as written.

Outputs

The model returns a single generated video clip as an MP4 file. The clip includes an audio track that matches the scene intent described in your prompt, such as ambience, sound effects, music, or dialogue.

Limitations

• Clip length is limited to short generations. This listing supports 5 to 15 seconds per run.
• Output resolution in this listing is limited to 480P or 768P.
• If you need 2K output, use MiniMax H3 instead of H3 Max.
• The returned clip length can be slightly longer than the requested duration on some deployments.
• When you provide a first or last frame, the output uses that image’s canvas shape. Mismatched framing can cause unwanted cropping or composition shifts.
• Prompt expansion can change phrasing and introduce details you did not write. Disable it when you need strict wording.
• Low-quality inputs can degrade results. Blurry images, heavy compression, text artifacts, or watermarks can cause flicker and warped details.

Safety & compliance

Hosted deployments of H3 Max often apply automated content filtering. Some providers, such as fal, run formal Trust & Safety programs with systems to detect CSAM, block non-consensual intimate imagery, and moderate harmful requests.

Don’t use this model to generate illegal content, sexual content involving minors, or non-consensual sexual imagery. Avoid prompts that request impersonation or deceptive real-person content without consent.