MiniMax released H3, an open-weights multimodal video model. It combines text, images, video, and audio in one context. It generates 2K video with native stereo audio up to 15 seconds. It supports text-to-video, first-and-last-frame, reference-to-video, and precise video editing. It accepts up to 9 images, 3 videos, and 3 audio tracks as references. It renders legible text, UI, and graphics. It produces original score, dialogue, and foley timed to the picture. It allows localized edits while keeping the rest of the shot stable. fal.ai hosts the model via a serverless API. A faster variant, H3 Max, generates 768p clips in under 3 seconds. Commercial use of generated content is permitted.
What changed
Separate models were required for text-to-video, image-to-video, and reference-based generation.
What it unlocks
Generating 2K video with native stereo audio from combined text, images, video, and audio in a single context.
- 2K resolution
- 15 seconds max duration
- 7,000 characters max prompt
- 9 images, 3 videos, 3 audio tracks max references
What you need to act on it
- access to fal.ai API
- paid usage for MiniMax H3 (H3 Max offers free daily generations)
Sources