minimax/h3-max/3d-to-video

H3 Max 3D-to-Video

Turn a 3D preview, blockout, or previs animation into a finished video. The input video guides scene layout, camera movement, and object motion, while reference images guide the final appearance.

How to use

1. Upload your 3D video. Use it to establish framing, spatial relationships, camera movement, and action. Simple geometry is fine—make important objects and movements easy to distinguish.
2. Optionally add reference images. Provide images showing how the subjects, environment, materials, and lighting should look.
3. Optionally add Scene Intent. Describe details that are unclear from the inputs, such as what a placeholder represents or the intended atmosphere.
4. Choose your output resolution and generate.

Scene Intent is optional. Use it to clarify the scene rather than repeat everything already visible.

Choosing effective references

• Match the scene content. Choose references relevant to the subjects and environment in your 3D video.
• Make important details visible. Clear views of a subject’s shape, clothing, surfaces, or materials provide stronger guidance than small or heavily obscured details.
• Keep references consistent. Avoid conflicting versions of the same subject or incompatible visual styles unless the difference is intentional.
• Give each reference a purpose. Complementary images can describe different subjects, environments, or details. More images are not necessarily better.
• Clarify ambiguous relationships. If needed, use Scene Intent to explain which reference corresponds to which object in the scene.

Think of the 3D video as guidance for where things are and how they move, and reference images as guidance for how they look. Results are generative and may differ from exact source geometry or reference details.

Using no reference images

You can generate without supplying reference images. The endpoint automatically creates visual references from the input video and any Scene Intent you provide.

Automatically generated references are billed at $0.10 per image, in addition to video generation and applicable input token charges. Supply your own references when you want more control over the final appearance.

Pricing

The total price includes a $0.50 processing fee per request, video generation, input token charges, and any automatically generated reference images.

• Duration: Each shot is rounded up to a whole second, with a minimum of 5 billed seconds per shot.
• Input video and reference images: $0.02 per 1,000 tokens beyond the 4,096 tokens included per shot.
• Automatically generated references: $0.10 per image.

For example, a single 5-second video at 768p starts at $0.90: $0.50 processing + $0.40 video generation. Input token charges and automatically generated references are additional, so the final total may be higher.