Unique3D
github.com
- Free plan
- Windows
- Linux
- Web
- Self-hosted
What Unique3D is
Unique3D creates textured 3D meshes from a single image. Its method derives four orthographic views from the input and progressively increases their resolution, then generates corresponding normal maps with a normal diffusion model. The project describes mesh generation in 30 seconds from single-view images and says it can produce a high-fidelity mesh from one orthogonal RGB image in under 30 seconds. Its ISOMER reconstruction algorithm combines color and geometry cues and can produce meshes with millions of faces. Users can try the online demo for free with a registration invitation code, or run a local interactive Gradio demo after downloading model weights. The online demo is described as slower but more stable than the Gradio demo. The project also links to a ComfyUI implementation. Linux setup targets Ubuntu 22.04.4 LTS and CUDA 12.1, while Windows installation may require Visual Studio Build Tools. Occluded images can produce weaker reconstructions when the four views do not capture the whole object.
Who it suits
Unique3D suits people who want to turn an image into a textured 3D mesh, including researchers interested in collaboration on 3D generative models. Users should be comfortable with its model-weight download and platform setup requirements.
Good to know, and what to watch
- Generates textured meshes from a single image.
- Supports image-to-3D and AI texturing.
- ISOMER can reconstruct meshes with millions of faces.
- Links to a ComfyUI implementation.
- Local interactive Gradio demo is available.
- Online demo requires a registration invitation code.
- Local inference requires downloading model weights.
- Occlusions can worsen reconstruction quality.
- Windows setup may require Visual Studio Build Tools.
Plans
Every plan and priceOnline demo
Free plan
The verdict
Unique3D offers an image-to-mesh workflow with multi-view generation, normal maps, and GLB export. Its setup requirements and sensitivity to occluded inputs are worth considering before choosing it.
What the maker publishes
- ComfyUI integration
- The repository links a ComfyUI-Unique3D implementation for ComfyUI support. source
- Generation speed
- The project says it can generate textured meshes from single-view images in 30 seconds. source
- Input limitations
- The README warns that images with occlusions can produce worse reconstructions because four views may not cover the full object. source
- Intended users
- The project invites research collaboration and describes its mission as creating a 4D generative model with 3D concepts. source
- Linux requirements
- The README setup targets Ubuntu 22.04.4 LTS and CUDA 12.1. source
- Local inference
- The repository provides instructions for running an interactive local Gradio demo after downloading model weights. source
- Mesh reconstruction
- Its ISOMER reconstruction algorithm combines color and geometric priors and can reconstruct meshes with millions of faces. source
- Multi-view generation
- The method generates four orthographic views from one input image and progressively upscales them. source
- Normal maps
- The method uses a normal diffusion model to generate normal maps corresponding to the multi-view images. source
- Online demo
- The online demo is described as free to try with a registration invitation code, and as slower but more stable than the Gradio demo. source
- Purpose
- Unique3D generates high-fidelity textured 3D meshes from a single image. source
- Research result
- The project page says Unique3D generates a high-fidelity textured mesh from one orthogonal RGB image in under 30 seconds. source
- Security
- The repository security page says the project has no SECURITY.md policy and no published security advisories. source
- Windows setup
- The README describes a Windows installation path that may require Visual Studio Build Tools. source
Checked against the maker's pages on 7 Oct 2026.