source: Hugging Face Blog: Rebuilding AUTOMATIC1111 with Gradio Workflow
level: technical
hugging face has released workflow1111, a gradio workflow app that rebuilds most of automatic1111's stable-diffusion-webui feature set on a single canvas. the app uses 73 nodes to create 11 media pipelines, including text-to-image, hi-resolution fix, image-to-image, prompt-matrix grids, vlm interrogate, detection-to-inpaint masks, controlnet-style annotators, background removal, png info storing, and image-to-video. users can run any pipeline by signing in with a hugging face account or providing an access token, and model calls use the user's own quota.
the canvas uses four operator kinds: fn for python functions, model for inferenceclient calls, space for other gradio spaces, and dataset for hub dataset rows. roughly two-thirds of the canvas runs in-process without network calls, with 32 of 36 operator nodes being fn nodes. every output node becomes a rest endpoint, exposing nine endpoints like /image and /png_info, and can also be exposed as mcp tools for ai assistants. the app runs without a gpu by calling inference providers or spaces, but a fn node can load a local model to run on your own gpu.
workflow1111 is compared to comfyui, but gradio workflow covers similar ground with nodes that can be hardware you don't own, typed rest endpoints generated from the graph, and oauth for visitors to run workflows under their own identity. custom nodes are plain python functions, so they can do anything python can. the app started with a simple gr.workflow(bind=[your_function]).launch() and can be duplicated from the space to rewire pipelines. this approach lets users build multi-model pipelines that open in a browser, sign in, use right away, and call from code.
why it matters: it shows how ai pipelines can be built and shared as modular graphs with automatic api endpoints, making complex image and video workflows accessible without gpu ownership.
source: Hugging Face Blog: Rebuilding AUTOMATIC1111 with Gradio Workflow