MiniMax H3 Reference Video — Character Consistency

How does MiniMax H3 reference video work?

MiniMax H3 reference-guided generation (Ref2VA) lets you attach reference images or a reference video to any generation. The model uses the references to lock the identity of characters, products, and styles, then renders your prompt around them. The result: the same spokesperson, mascot, or sneaker across every clip in your campaign.

On H3 Video, using a reference is simple: upload the reference alongside your prompt, choose your duration and resolution, and generate. Character consistency is one of the capabilities the community praises most about the MiniMax H3 model.

Reference video vs reference images

Reference images are best for locking a look: a character design, a product angle, a style frame. Reference video goes further, also preserving motion qualities and timing — useful when you need a new shot to cut seamlessly against existing footage.

Combine references with the six-layer H3 prompt formula (subject, action, scene, camera, lighting, audio) to control everything the references do not define.

Best practices for consistent characters

Use clean, well-lit references that show the subject clearly. Keep the subject description in your prompt aligned with the reference ("the same red-jacketed cyclist") rather than contradicting it. Generate a establishing shot first, then reuse it as the reference for follow-up shots in the same scene.

H3 Video FAQ