Experiment · Previz
Greek
AI video generation is the expensive step. Every regeneration to fix a bad angle, off timing, or a subject drifting out of frame burns credits and minutes. So don't discover the shot inside the video model — plan it cheaply in Blender first, then hand the model a previz it has to follow.
Build the scene and camera with Claude in Blender
The Higgsfield Bridge exposes Blender as callable tools — add primitives, set transforms, place cameras and lights, keyframe motion, run arbitrary bpy. That means Claude can build the scene: block out geometry, stage the actors, place the camera, set the lens, animate the dolly, key the timing. All of it is deterministic, all of it is free to iterate on, and none of it costs a single video-gen token.
Below is the scene coming together: greybox environment, staged characters, and the camera move for the dialogue beat — assembled conversationally, adjusted until the shot reads.
Hand the shot to the model as a contract
Once the blockout is locked, the Blender render becomes the reference the video model has to honor. Framing, camera path, action beats, duration — all decided already. The AI doesn't get to reinvent the shot; it gets to dress it. That flips the economics: instead of spending tokens to search for the right composition, you spend them once, on execution of a composition you've already validated.
Twenty seconds of previz below — entrance, dialogue beat, reaction, out — the exact shot the Blender scene defined, generated in a single pass.