Build a new scene
Use an agent environment with GPT-6 Astra access, Python and Blender. Supply the photograph and any known dimensions. New scenes require observation and custom modeling.
Read the getting-started guide ↗An open-source scene-building workflow
A room photo.
An editable world.
Build 3D scenes you can keep working on. A community GPT-6 Astra workflow for observing photographs, modeling in Blender, and reviewing the result.
Look closely
Photograph ↔ reconstruction
Drag the divider, or use the arrow keys.
SOURCE PHOTO
RECONSTRUCTION
↔
Bedroom — known bed length and candidate furniture specifications help constrain scale.Existing V5 examples
Each room starts from its own photograph. This preview compares appearance, not measured 3D ground truth. Absolute scale needs dimensions or explicit priors.
Keep the scene editable
Explore the space.
Work on the parts.
- Editable geometryFurniture is built from separate components for continued changes.
- Optional interactionThe film shows Blender Bullet hinge motion and Cloth: two dynamic garments and three static ones.
- Evidence stays attachedSpecifications, measurements, revisions and known limitations are documented in the repository.
Start with the right path
See it here. Reproduce it with the tools.
Replay the included example
Follow the documented environment setup and frozen reconstruction recipe. This separate MuJoCo replay example is distinct from the V5 showcase above.
Open the replay instructions ↗What would you reconstruct?
Inspect the workflow, try an example, and share what works or where it breaks.
Star the project to follow its progress ↗