VIGA Writes Blender Code, Renders It, Critiques It, and Fixes It — Until the 3D Scene Matches the Photo
A new agentic system from Impossible Inc. generates editable 3D worlds from a single image by iteratively writing and debugging Blender scripts, marking a significant step toward autonomous 3D content creation.
Impossible Inc. has released VIGA, an AI agent that takes a reference image and autonomously constructs a matching 3D scene in Blender through an iterative code-render-compare loop. As @minchoi described it: "Give it any image and it writes Blender code, renders, compares, and fixes itself until the scene matches." The system even adjusts camera angles during its self-critique phase to identify discrepancies from multiple viewpoints.
Unlock the full briefing
Get every story in today's briefing, the full archive, and the daily AI intelligence brief.
All stories today
Full archive
Daily brief
Cancel anytime. Payments powered by Stripe.