LL3M: Large Language 3D Modelers
7 days ago
- #3D modeling
- #LL3M
- #Blender
- The method involves three phases: initial creation, automatic refinement, and user-guided refinement.
- LL3M corrects implausible configurations and simplistic geometry automatically in the initial phase.
- User-guided refinement allows interactive and iterative 3D asset generation.
- LL3M supports diverse shape generation with detailed parts and intricate arrangements.
- Each mesh is generated through interpretable, editable Blender code.
- LL3M can apply style concepts (e.g., steampunk) to different initial meshes.
- The system enables material editing on specific parts of the mesh via shader nodes.
- Multiple successive edits of the same 3D asset are possible while preserving identity.
- Generated Blender code is well-documented, with clear variable names and structured logic.
- Users can intuitively edit parameters in Blender nodes and structures.
- LL3M generates multiple objects and arranges them with appropriate spatial relationships.
- Parenting relationships enable scene graph behavior in Blender.
- Each part in the scene hierarchy is assigned a meaningful semantic name.