generative-models

Why we included this project

This is the repo where Stability AI ships the reference implementations of its diffusion models, so it's the obvious starting point if you want to run Stable Diffusion XL, Stable Video Diffusion, or the newer SV3D and SV4D pipelines from source rather than through a hosted API. The code is organized around each released checkpoint, which makes it straightforward to reproduce published results, fine-tune a base model, or adapt the sampling scripts to your own data. The most recent additions, SV4D and SV4D 2.0, turn a short video of a single object into multi-view novel-view video and 4D asset generation, which is handy for 3D content pipelines and visual effects work. The code is MIT licensed and the weights come from Hugging Face, so a small team can run local inference on a single GPU without much setup.