AI Cooks Up Complex 3D Animations and Worlds

Summary

Imagine an AI that can code, generate 3D models, create animations, and even build entire virtual worlds – that's the power demonstrated by Claude Cote when equipped with advanced APIs and granted permission to operate autonomously. The process involves installing Blender and a new official Blender MCP add-on, followed by setting up Claude Cote with 'dangerously skip permission' mode for uninhibited operation. While this bypasses manual approvals, a disclaimer advises caution and suggests starting with manual approvals if unsure. The system can leverage powerful APIs like Full AI for image and 3D generation, but requires credit replenishment. Demonstrated use cases include generating a complex animation of an angel character splitting and scattering like a tornado, which required five iterations to perfect. Another example shows the AI creating a dynamic bubble animation on a water surface using geometry nodes, allowing for adjustable bubble size and frequency. The system also explored world generation using tools like World Labs and ImageBlast, though with significant credit costs, up to five dollars per world. While attempting to create a Japanese-style world, the AI struggled with object placement and rotation, and vision capabilities were noted as weak for complex spatial reasoning. Testing with Codacs, a similar AI system, showed slightly improved vision and object scaling, but still not entirely production-ready. A challenging animation retargeting task, applying a Mixamo animation to a custom character rig, also highlighted the AI's iterative problem-solving, even recovering from a Blender crash.