TIMI: Training-Free Image-to-3D Multi-Instance Generation with Spatial Fidelity
Researchers have developed TIMI, a new framework for generating 3D models from images that focuses on multi-instance scenarios and maintains spatial fidelity without requiring additional training. The system utilizes an Instance-aware Separation Guidance module to disentangle individual instances early in the process. Additionally, a Spatial-stabilized Geometry-adaptive Update module ensures that the geometric properties and relative positions of these instances are preserved. Experiments show TIMI outperforms existing methods in both overall layout and distinct object representation, while also offering faster inference speeds. AI
IMPACT This new framework offers a training-free approach to 3D multi-instance generation, potentially speeding up development and improving spatial accuracy in AI-driven 3D content creation.