Verified project record

ashawkey/stable-dreamfusion

stable-dreamfusion generates 3D models from text prompts or 2D images using NeRF and Stable Diffusion backends. It is a work-in-progress research repository requiring a CUDA-capable GPU, manual dependency building, and coding knowledge.

Project overview

Project type: Model Runtime · Image & Vision
Deployment: Refer to project documentation
License: Apache-2.0

Best for

Developers and researchers on Ubuntu systems with supported CUDA GPUs (e.g., V100 or ~24G VRAM) exploring open-source NeRF and diffusion-based 3D generation.

Key capabilities

Generates 3D models from text prompts using DreamFusion powered by Stable Diffusion.
Generates 3D models from input images using the Zero-1-to-3 diffusion backend.
Exports generated 3D models as meshes with obj, mtl, and png texture files.
Provides a GUI for free-view control to visualize generated 3D models.
Supports negative text prompts to refine the 3D generation process.
Leverages monocular depth estimation to aid image-to-3D generation.

Limitations and risks

This project is a work-in-progress and contains lots of differences from the original paper.
The current generation quality cannot match the results from the original paper, and many prompts still fail badly.
Adopters may face OpenGL installation issues leading to eglInitialize() failure.

Getting started

Setup is considered hard. It requires manually cloning the repository, installing requirements, building CUDA extensions, and downloading pre-trained models. The first success path involves running main.py with a text prompt.

Evidence and sources

README: A pytorch implementation of the text-to-3D model **Dreamfusion**, powered by the [Stable Diffusion](https://github.com/CompVis/stable-diffusion) text-to-2D model.
README: This project is a **work-in-progress**, and contains lots of differences from the paper. **The current generation quality cannot match the results from the original paper, and man…
README: python main.py --text "a hamburger" --workspace trial -O
GitHub project description: Text-to-3D & Image-to-3D & Mesh Exportation with NeRF + Diffusion.
README: To use image-conditioned 3D generation, you need to download some pretrained checkpoints manually: * [Zero-1-to-3](https://github.com/cvlab-columbia/zero123) for diffusion backend.

Reviewed at: 2026-07-16T10:09:43Z