Hugging Face released Workflow1111 on September 10, 2026, reconstructing the feature set of the AUTOMATIC1111 stable-diffusion-webui inside a single Gradio Workflow canvas. The implementation links eleven distinct media pipelines across 73 nodes, combining text-to-image generation, high-resolution fixes, prompt-matrix grids, background removal, and image-to-video animation. Users connect using Hugging Face accounts or access tokens to run models against their personal quotas.
Media pipelines rely on four operator types: Python functions, model calls through InferenceClient, external Gradio Spaces, and dataset rows. The text-to-image pipeline processes parameters such as CFG, seed, steps, and resolution, passing inputs through a prompt-builder function before querying models through Inference Providers. For upscaling and edits, a FLUX.1-Kontext node performs refinement instructions and handles image-to-image modifications, while a Qwen3-4B node refines rough user prompts into tag lists capped at forty items.
Local processing and endpoints
Image interrogation routes photos simultaneously to a Qwen2.5-VL vision-language model and a ViT classifier, which execute in parallel. Mask generation pairs a DETR object detector with local Pillow and NumPy processing, isolating detected subjects on the machine without transmitting image data after the initial detection query. Preprocessors for Canny, line art, sketch, luma-depth, and posterize run entirely on CPU, each finishing in approximately half a second. Across 36 operator nodes in the app, 32 are Python functions, and 22 run locally without network requests.
Every output node automatically functions as a typed REST endpoint, creating nine routes that include endpoints for images, edited images, grids, and metadata inspection. Enabling server settings also surfaces these nodes as Model Context Protocol tools. External artificial intelligence assistants, including Claude Code and Cursor, can trigger individual pipeline steps by sending requests with user access tokens directly to the server.
