LEGO-Anything uses coding agents to generate 3D scenes from a single image
What happened
A research team proposes an Image-to-Code method in which a coding agent writes an executable Blender program from a single photo and iterates until the scene matches the original image. To evaluate it, the team built LEGO-Bench, containing 208 images of 104 indoor and outdoor scenes and 443 registered assets, scored on validity, geometric accuracy and appearance similarity. According to a report on 3 October 2026, the method can generate 3D scenes from photos, but the agent cannot tell whether it got them right.
Written by AI from the reports · updated 13 d ago
Timeline
Follow the reports to see the event from each side.
- The DecoderAI agents build 3D scenes from photos but have no idea if they got it right
A new approach called LEGO-Anything turns single photos into editable Blender code for 3D scenes. GPT-6 Astra leads the accompanying benchmark with up to 53 percent reconstruction accuracy. The biggest weakness across all tested agents is that they can't judge their own geometric accuracy any better than a coin flip.