Rebuilding AUTOMATIC1111 with Gradio Workflow

A Hugging Face blog post walks through Workflow1111, a project that rebuilds most of AUTOMATIC1111's feature set as a single Gradio workflow canvas. The post says the app is a graph of eleven media pipelines built from seventy-three nodes, covering 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 the pipelines by signing in with a Hugging Face account or providing an access token, after which model calls use their own quota, the post says.

According to the post, nodes wrap four operator kinds: fn (a Python function), model (called through InferenceClient), space (another Gradio Space), and dataset (a row from a Hub dataset). The text-to-image pipeline includes negative prompt, steps, CFG, seed, width, height, and a model_id field, with a post-process node writing generation parameters into PNG metadata. Hi-resolution fix is described as a two-node detour through a FLUX.1-Kontext model node, and that same node doubles as image-to-image. Qwen3-4B generates prompts from rough input, capped at forty tags, and Qwen2.5-VL plus a ViT classifier handle interrogation.

The post states the app has 36 operator nodes, 32 of them fn nodes, and that 22 run entirely in-process without a network call. It says every output node becomes a REST endpoint, listing nine, and that launching with mcp_server=True exposes each output node as an MCP tool. It also describes FastVideo/fastvideo-fasth3-preview as a Gradio Workflow app running FastH3, a four-step distillation of MiniMax-H3, on ZeroGPU.