Every AI video model launch used to come with the same asterisk: great picture, shame about the sound (or lack thereof). MiniMax just showed up and quietly removed the asterisk, dropping a model that generates video and matching stereo audio in the same breath — and it's giving away the recipe.
One Model, Every Modality, No Foley Booth
Shanghai-based MiniMax released H3, a general-purpose multimodal model that reads text, images, video, and audio as one unified context and outputs clips up to 15 seconds long at native 2K resolution (2560x1440) with synchronized stereo sound generated in the same pass — not stitched on afterward. It can also edit existing footage and transfer motion from a reference clip onto new material, all from natural-language instructions.
Pricing lands at $0.14 per second for 2K output and $0.10 per second for 768p, and MiniMax says it plans to release the model's weights within days — meaning anyone will be able to download, inspect, and modify the thing that just matched some of the best closed video models on the market.
The Open-Weights Flex Nobody Saw Coming
The headline spec sheet is impressive, sure, but the real story is MiniMax handing out the keys. While plenty of Western labs treat their best video models like a walk-in vault, MiniMax is planning to open-source weights for a model that generates synced audio-video natively — a combination that's still genuinely hard to pull off well, let alone give away for free.
That's a meaningful shift in leverage: open-weights video generation this capable puts serious creative tooling in the hands of indie developers, researchers, and yes, probably a wave of deepfake anxiety right behind it. Native audio-video sync also quietly raises the bar for every competitor — "video generator" is starting to mean "video and sound generator," full stop, and the labs still bolting audio on as an afterthought just got a deadline.
MiniMax isn't just shipping a good model — it's shipping a good model and then handing you the source code, which is the kind of move that tends to reshuffle a market fast.
Source: MarkTechPost