Two models, two jobs
The Hailuo family plays both ends of the roster. MiniMax H3 is the multimodal flagship: it reads text, images, video clips and audio in a single context. Brief it like a creative director with a mood board, not like a search engine. Mixed-reference briefs that other models can't parse, "this product, in footage like this, sounding like that", are exactly what H3 was built for. Hailuo 2.3 is the opposite end: the family's budget workhorse, and proud of it.
The draft-cheap pattern
Here's how pros use Hailuo 2.3: every prompt idea, every camera direction, every hook gets drafted first. Nine out of ten takes die in the library. That's the process, and they die cheap. The survivor gets re-rendered on Kling 3.0 or Seedance 2.5 at full quality. Same composer, same prompt, one wallet: the draft and the final differ by one chip selection. If you're iterating toward an ad or a short, this pattern cuts your cost per keeper by half or more.
Why multimodal matters
Most models read a prompt. H3 reads a production brief. Attach the product photo AND the reference footage AND the soundtrack note, and the context fusion shows in the output. Style, motion and tone pulled from all three. For agencies translating client mood boards into motion, and for creators remixing their own footage with AI extensions, MiniMax H3 is currently the only model on the platform that takes the whole brief at once. The versions table above has both models, live, with real prices. Draft on 02, brief on H3, finish wherever the take is best.
