Opening
Active · Research
This Lab Entry studies existing Blender and AI-agent approaches alongside the emerging Generative Scene System.
The question
Where can existing tools help turn a visual reference into an editable 3D environment, and where do they leave gaps? The question spans visual interpretation, procedural construction, agent coordination, and judging whether a scene works.
An agent may help interpret a scene, select available capabilities, coordinate work, and inspect feedback. Blender construction still depends on reliable tools and human direction. Reusable primitives and generators matter because they give the agent dependable operations to work with instead of a fresh improvisation each time.
The survey compares approaches to Blender control, scene awareness, reuse, visual inspection, and validation. Different tools may solve different parts of the problem. The research asks what can be adopted, what needs adaptation, and where a missing capability deserves its own experiment.
What we have learned
The useful question is less “Can an AI make a world?” and more “Which parts can be described, built, checked, and revised with confidence?” A scene representation between the reference and Blender may help separate creative interpretation from construction. Spatial boundaries and feedback may make revisions more local and understandable.
These are architectural directions, not a claim that an autonomous scene builder is already operating. Findings are being brought into small Blender experiments. Detailed capability definitions and implementation choices remain in the project workspace.