facebook/sam-3d-objects
Updated • 8.77k • 668
Text-to-3D & Image-to-3D
Text-to-3D & image-to-3D AI models: meshes, textures & spatial AI assets from prompts or photos - TRELLIS.2, Hunyuan3D, SAM 3D & more.
Note Reconstruct textured 3D objects from a single image with SAM 3D Objects. Image masks select the objects to recover, while the model estimates geometry, appearance and pose for single-object or multi-object reconstruction.
stabilityai/stable-point-aware-3d
Image-to-3D • 2B • Updated • 4.78k • 453Note From reference image to textured, UV-unwrapped mesh: SPAR3D combines point-cloud diffusion with mesh prediction. Editable point clouds provide control over object geometry, with material parameters included for downstream rendering.
facebook/sam-3d-body-dinov3
Updated • 16k • 1.19kNote For human reconstruction, SAM 3D Body recovers a full-body 3D mesh from one image, including body, hand and foot pose. Optional keypoints and masks guide inference; outputs include mesh vertices and pose parameters.
microsoft/TRELLIS.2-4B
Image-to-3D • Updated • 2.43M • 1.5kNote Microsoft's second-generation image-to-3D foundation model (4B). TRELLIS.2 turns a single image into structured 3D assets ready for mesh workflows - the most battle-tested entry point here, with a text-to-3D sibling in this collection.
tencent/Hunyuan3D-2.1
Image-to-3D • Updated • 114k • 1.39kNote Tencent's established image-to-3D system: detailed meshes plus PBR texture painting in one pipeline. Among the most downloaded open 3D models - note the community license with regional restrictions before commercial use.
TencentARC/Pixal3D
Image-to-3D • Updated • 403Note Pixel-aligned image-to-3D from TencentARC: geometry and PBR textures via direct pixel-to-3D mapping. An MIT-licensed 2026 newcomer building on TRELLIS research - strong quality for a compact setup.
microsoft/TRELLIS-text-xlarge
Text-to-3D • Updated • 13.9k • 92Note The text-to-3D sibling of the TRELLIS family: generate 3D assets directly from text descriptions, no input image needed. The prompt-driven entry point of this collection alongside the image-conditioned models.
VAST-AI/TripoSG
Image-to-3D • 1B • Updated • 12.5k • 152Note VAST-AI's 1.5B rectified-flow base model for image-to-3D geometry. Focuses on clean, editable shapes - texture it afterwards with a paint model or your own pipeline. MIT-licensed.