Hugging Face has rebuilt the AUTOMATIC1111 Stable Diffusion pipeline as a native Gradio Workflow application, swapping the decade-old WebUI monolith for a modular, node-based graph that runs directly on Hugging Face Spaces.
Key takeaways
Hugging Face rebuilt AUTOMATIC1111's core Stable Diffusion pipeline as a Gradio Workflow app — a node-based, modular graph architecture that replaces the original WebUI.
The new app runs on Hugging Face Spaces, meaning no local install is required to access AUTOMATIC1111-style generation.
Individual pipeline stages — sampling, VAE decoding, LoRA loading — are exposed as discrete, rewirable nodes rather than buried in a settings panel.
The rebuild uses Gradio's Workflow component, which serializes the graph as JSON, making pipelines shareable and version-controllable.
Existing AUTOMATIC1111 extensions and checkpoints are not automatically compatible; the new architecture requires adapted integrations.
Why the old WebUI was hitting a wall
AUTOMATIC1111's original webui.py launched in 2022 as a rapid prototype and never fully shed that DNA. Settings were global, the extension API was brittle, and adding a new sampler or ControlNet pipeline meant patching a shared state object that other extensions also touched. For creators who run complex LoRA stacks or chained inpainting passes, race conditions and silent failures were a recurring cost. The Gradio Workflow rebuild addresses this structurally: each processing step is an explicit node with typed inputs and outputs, so the data flow is visible and auditable rather than implicit.
The rebuilt AUTOMATIC1111 interface exposes each pipeline stage — sampling, VAE decoding, LoRA injection — as a discrete, rewirable node in Gradio Workflow.
In the new architecture, a standard txt2img run looks like a small directed graph: a prompt-encoding node feeds a sampler node, which feeds a VAE decoder node, which outputs to a preview node. Want to insert a ControlNet pass? You wire in a conditioning node between the encoder and sampler — rather than toggling a checkbox that invisibly alters global pipeline state.
For creators who already use ComfyUI, this will feel familiar. The meaningful difference is that Gradio Workflow serializes the graph as plain JSON, which means a pipeline can be committed to a Git repo, diffed, and shared as a Hugging Face Space with a single URL — no zip file of custom nodes required. That portability alone changes how teams and tutorial creators distribute their setups.
According to the Hugging Face blog, the rebuilt app also exposes a Python API automatically, so any node in the graph can be called programmatically without scraping the WebUI's internal routes — a pattern that was technically unsupported in the original but widely abused anyway.
The compatibility gap creators need to plan around
The rebuild is not a drop-in replacement. AUTOMATIC1111's extension ecosystem — roughly 300+ community extensions covering everything from face restoration to regional prompting — was written against the original WebUI's hook system. None of those extensions load in the Gradio Workflow version without being rewritten as Gradio components or custom nodes. Hugging Face has not yet published a migration guide or a compatibility layer for the most-used extensions, so anyone dependent on, say, sd-webui-controlnet or adetailer should treat this as a parallel environment rather than an upgrade path for now.
Checkpoints and LoRA weights are fully reusable — the underlying diffusion models haven't changed, only the orchestration layer around them. Creators can point the new app at any Safetensors checkpoint they already use with the original WebUI.
Each pipeline stage is a typed node: changing samplers or inserting ControlNet conditioning is a wiring change, not a settings-panel toggle.
Because the app targets Hugging Face Spaces natively, creators without a GPU workstation can run it on a free or paid Space — though generation speed on a free CPU Space will be slow for anything beyond small resolutions. Paid Spaces with an A10G or T4 GPU bring performance closer to a local RTX 3090 setup. For creators already using Charmloop's AI image generator for quick iteration, the Gradio Workflow rebuild is more relevant as a customization and experimentation environment than as a replacement for a polished hosted tool.
Hugging Face hasn't announced a timeline for extension compatibility tooling or official migration documentation. The practical next step for power users is to watch the Hugging Face blog post for follow-up commits — the repo is public and active, and the node schema is stable enough to start building against now.
