BentoDiffusion
A collection of example projects for self-hosting Stable Diffusion family models as HTTP services with BentoML, covering image and video generation from text prompts.
BentoDiffusion is a collection of example projects showing how to serve models in the Stable Diffusion family with BentoML, for generating and manipulating images or video clips from text prompts. Each subdirectory covers a different model, such as SDXL Turbo or other Stable Diffusion variants, with its own service definition.
Starting the BentoML service inside a model's directory launches an HTTP service exposing endpoints such as text-to-image generation, reachable through a web UI, curl, or a Python client. A service can also be deployed to BentoCloud with a single command after logging in, or packaged as an OCI-compliant image for custom infrastructure.
Testing locally is recommended on an Nvidia GPU with at least 12 GB of VRAM. It is written in Python under the Apache-2.0 license.