Skip to content

MiniMax H3 / Image to video

MiniMax H3 architecture clips from interior images

A single interior image can support a short camera move, but it does not contain a complete 3D building. Treat H3 image-to-video as a presentation tool for a limited viewpoint. It should not be advertised as an accurate walkthrough of rooms that are absent from the source.

Keep the movement within the image's evidence

A small push-in toward a window or a gentle lateral drift can add life to an architectural visual. Moving through a doorway into an unseen room requires the model to invent a layout. For a property listing or design approval, that invention may be misleading.

Prepare straight vertical lines and a clean source composition. If the input already has distorted walls, the animation may amplify the problem. Keep furniture and structural elements easy to distinguish from reflections.

Build a controlled interior shot

  1. Choose an approved render or photograph and crop it for the intended presentation format.
  2. Upload it to H3 Image to Video.
  3. Ask for a small movement while preserving the room layout, windows and furniture.
  4. Compare wall edges, floor patterns and door positions at multiple points in the result.

Example prompt

PROMPT RECIPE / ADAPT TO YOUR SOURCE

Preserve this living room's layout and furniture. The camera makes a slow, short push toward the large window without entering another space. Curtains move very gently in a light breeze. Straight walls and stable window frames, soft daytime ambience, one continuous shot.

Present it honestly

Use the clip as a visualization and identify conceptual material where appropriate. For a multi-room tour, generate separate shots from approved views and connect them with clear edits. Do not imply that a generated transition proves how rooms connect. If spatial accuracy is the primary requirement, a measured 3D walkthrough or real video capture is the more suitable production method.

REVIEWED JOLLYAI IMPLEMENTATION

The workflow behind the guide

One uploaded image + motion brief → image and text conditioning → H3 FL2VA turbo sampling → video/audio decoding → combined video file.

Inputs, presets & access
One image; aspect follows the source. Limited free access and longer PRO presets. Current ordinary pixel budget: 0.4 MP.
Implementation note
Local H3 turbo workflow. Six sampling steps; simple scheduler; turbo adapter; video/audio shifts 12/3. Current environment: pruned W4A8 FL2VA weights with a 4B encoder and matching projection. R2V uses a separate reference checkpoint. These settings describe the reviewed JollyAI deployment, not every H3 implementation.

Configuration reviewed September 20, 2026. Your current playground controls and plans page determine availability. Prompt recipes above are suggestions, not guarantees or claimed test results.

Open Image to video ↗

Sources and useful next steps

The implementation notes describe JollyAI's reviewed local integration. Upstream capabilities and licences may differ from the settings available here.

Found an outdated setting or a reproducible problem? Contact JollyAI with the workflow and job reference. Keep account tokens and private input files out of public reports.