TL;DR: MiniMax H3 is a next-generation, open-weights, multimodal video model, and we're a Day-0 launch partner. It's already live on Civitai, so you can generate with it right now. Open weights land in a few days.
What H3 is
H3 is MiniMax's newest video model. Open weights, general purpose, and multimodal, which means it doesn't just read a text prompt - it works with audio, reference images, and existing footage too. MiniMax took it GA this week, the API is live now, and the open weights are coming in a few days.
We got early access as a launch partner, and we've had it running on Civitai since yesterday morning. So this isn't a "coming soon" post. It's a "go try it" post.
What it can do
Text-to-video with precision control: describe the shot and steer it, not just prompt-and-pray.
Video-to-video object removal: take existing footage and cleanly pull objects out of it.
Audio-driven lip-sync: feed it audio and video together and get matched lip-sync. Audio input is free.
Animatic-to-footage: turn a rough animatic into finished-looking video.
Multi-video reference: point it at several reference clips to guide the result. Your first 5 reference images are free.
Try it now
H3 is in the generator today. Pick it, prompt it, and share what you make. No waitlist, no early-access form.
Want to see what it's capable of first? MiniMax put together a model card and a set of demo cases here.
Open weights are coming
The API is live now. MiniMax has said the open weights will follow in a few days. When they land, the usual open-model workflows open up on top of everything H3 already does through the generator. We'll keep you posted.
Thanks to MiniMax
Big thanks to the MiniMax team for making us a Day-0 partner on this one. You can read their launch announcement here.
Go generate with H3. Push it somewhere weird. Show us what H3 can do.