Awesome 3D Diffusion — a curated paper list

What it is: A GitHub repository that collects academic papers applying diffusion models (a type of AI used for image and 3D generation) to 3D content creation. It is maintained by researchers and meant as a reference hub for the field.

What it contains: The README itself is almost entirely a categorized list of papers, each with a link to its arXiv (academic preprint) page and citation info. Every entry includes authors and venue/year. The collection also links out to two related resources: a sister repository on 4D generation (3D plus time) and a survey PDF summarising the papers in this list. A BibTeX citation is provided for people who want to cite the survey in their own work.

How it is organized: The papers are grouped by technical approach. The three main branches are:

1. Methods starting from a pre-trained 2D diffusion model — using an existing 2D image-diffusion model as a guide, with subsections for turning text into single 3D objects, generating whole scenes or composed objects, turning one image into 3D, generating human or animal subjects, editing existing 3D models, adding textures, and producing multi-view images.
2. Methods without pre-trained 2D diffusion — training diffusion directly for 3D objects or 3D scenes.
3. Diffusion performed in 3D space itself — working directly with 3D representations like Gaussian splats, point clouds, meshes, volumes, implicit functions, triplanes, latent codes, or novel new representations.

It also has a section on diffusion for motion, specifically human motion.

How to contribute: The maintainers explicitly say the list is not complete and invite anyone to submit additions or corrections via a pull request directly — no need to open an issue first. This makes it an openly collaborative, living document rather than a static archive.