MiniMax H3 Unifies Text, Image, Video, and Audio Generation, With Open Weights Coming

MiniMax detailed H3, a general-purpose omni-modal model that the company says understands text, images, video, and audio in a single system, then generates video with native stereo audio at up to 2K resolution and 15 seconds per clip. MiniMax says it plans to open the model weights in the coming days, subject to applicable laws and regulations.

• One model spans four modalities. MiniMax positions H3 as understanding and generating across text, image, video, and audio rather than chaining separate tools.
• Stereo audio is generated with the video, not added after. The company describes the sound as native and jointly produced alongside the picture.
• Weights are slated to go open. MiniMax says it will release the model weights within days, putting a multi-input generator into self-hosting range.

The model covers a matrix of reference and editing operations, not just text-to-video

MiniMax lists a wide set of generation paths for H3. Beyond text-to-image, text-to-video, and text-to-audio, the company says the model handles image-to-image reference and editing, image-to-video reference and editing, audio-to-audio reference and editing, and audio-video-to-audio-video reference and editing. MiniMax also lists multi-shot modeling among the capabilities.

According to MiniMax, H3 is strongest at instruction following, accurate text and brand presentation, and video-to-video motion transfer. The company frames those as the traits that make outputs usable for finished work rather than tests, describing H3 as aimed at production-ready content generation.

Native stereo audio and a 2K ceiling define the output specs

H3 generates video with native stereo sound that MiniMax says is produced jointly with the visuals. The model defaults to 2K resolution and also offers a 768p option, with clip length running up to 15 seconds.

The jointly generated audio track puts H3 alongside a run of recent video models that fold sound into the generation step. We covered Wan2.5-Preview, which the Alibaba team positioned as a native multimodal architecture generating synchronized audio with video.

Kling and PixVerse have pushed the same way. Kling 3.0 added multi-speaker native audio and 15-second clips.

PixVerse took a similar route with PixVerse V6, which targeted multi-shot coherence and built-in audio.

On cost, MiniMax makes comparative claims without publishing exact rates. The company says H3 at 2K runs at less than a third of the price of mainstream models, and that its 768p option costs less than half the price of mainstream models' 720p. MiniMax does not name the models it is comparing against, so the figures stand as the company's positioning rather than a verified benchmark.

MiniMax points H3 at advertising, product design, and title sequences

The company lists a set of commercial and creative use cases for H3: advertising, branding, e-commerce, product design, UI/UX, gaming, film opening titles, product websites, and animated posters. The emphasis on accurate text and brand presentation tracks with that list, since ad and product work depends on rendering logos and copy correctly rather than approximating them.

For filmmakers, the film-opening-title and animated-poster use cases are the most direct hooks. Motion transfer and multi-shot modeling point at sequences that need consistent movement across cuts, and the instruction-following claim addresses the gap that usually separates a demo clip from a shot that matches a brief.

Open weights would move H3 from hosted demo to self-hosted tool

MiniMax says it will open the model weights in the coming days, subject to applicable laws and regulations. If the release lands, H3 would join the group of capable generation models that teams can run and fine-tune on their own infrastructure rather than access only through an API, a distinction that matters for studios weighing data control and per-clip cost.

Every performance, resolution, price, and capability figure here comes from MiniMax's own announcement, and the weights are not out yet. The specs describe what the company claims H3 does at launch. Whether the open-weights release ships on the stated timeline, and how the output holds up against the pricing comparisons, will be clearer once the model is in hand.