Rebuilding AUTOMATIC1111 with Gradio Workflow
Hugging Face rebuilt AUTOMATIC1111’s features into a single Gradio Workflow canvas called Workflow1111, integrating eleven media pipelines with seventy-three nodes for tasks like text-to-image and image-to-video.
Hugging Face has transformed AUTOMATIC1111’s extensive feature set into a unified Gradio Workflow canvas named Workflow1111, consolidating eleven media pipelines using seventy-three interconnected nodes. The workflow integrates state-of-the-art models for text-to-image generation, high-resolution fixes, image-to-image editing, prompt matrices, and image-to-video conversion, all accessible via a single interface. Users can run pipelines by signing into their Hugging Face account or providing an access token, with model calls consuming their own quota.
Workflow1111’s canvas is built from four operator types: Python functions, model calls via InferenceClient, references to other Gradio Spaces, and dataset rows from the Hugging Face Hub. Each node on the canvas represents one operator, with inputs and outputs connected as edges. The text-to-image pipeline includes controls such as negative prompts, steps, CFG, seed, width, height, and model selection, while a prompt-builder function node appends style presets before generating the image.
The high-resolution fix pipeline uses a FLUX.1-Kontext model node to refine and upscale text-to-image outputs, while the image-to-image tab allows users to upload an image and describe desired edits, with a Qwen3-4B model node converting the response into a structured list of tags for rendering. The workflow also supports automatic mask generation for inpainting using a DETR object detector, which identifies objects and creates masks locally without requiring manual painting.
The prompt matrix feature combines a base prompt with multiple suffixes to generate variant images in parallel, mimicking AUTOMATIC1111’s functionality. Background removal and upscaling are handled by Space nodes that call external Gradio Spaces, while preprocessing annotators like Canny and line art are implemented as local NumPy functions. The workflow preserves generation metadata in PNG files and supports image-to-video conversion, with all pipelines designed to run efficiently even without a local GPU.