Stable Diffusion WebUI vs Fooocus

Last reviewed Aug 26, 2026 · 742 words

Fooocus was written as an argument against Automatic1111. Its author, lllyasviel, is the researcher behind ControlNet, and Fooocus takes the position that most of the dozens of controls on an A1111 txt2img tab are things the software should decide for you. So the two are not rivals so much as a thesis and its rebuttal: A1111 is a workbench holding every knob the community ever asked for, Fooocus is a prompt box with strong defaults and a hidden Advanced tab. Both are Gradio apps, both are Python, both list 8 GB minimum RAM and want an Nvidia card, and both have been in maintenance mode long enough that choosing between them means choosing which kind of frozen you prefer.

What the defaults do for you

Type "portrait of a lighthouse keeper, film still" into Fooocus and it silently expands the prompt with its GPT-2-based expander, applies a style preset, picks the sampler and scheduler, sets a resolution from a list of SDXL-native aspect ratios, and refines with a second pass. The result looks like a competent Midjourney output because Fooocus was tuned to produce exactly that. Type the same prompt into a fresh A1111 install and you get a 512x512 image from whatever checkpoint you loaded, at 20 steps of the default sampler, and it will look like 2022 until you learn what the sliders mean. That learning is the whole A1111 experience: negative prompts, CFG, hires fix, the right VAE, and a growing folder of LoRAs.

What A1111 has that Fooocus refuses to

The extension list. ControlNet as an extension with every preprocessor, ADetailer for face fixing, Dynamic Prompts for wildcards, regional prompting, training helpers, and hundreds more, on top of built-in inpainting with mask blur, mask padding, and inpaint-area controls that Fooocus reduces to a brush and an "improve detail" option. A1111 also still runs SD 1.5 checkpoints, which matters if you have a library of 1.5-era models and LoRAs; Fooocus is SDXL-only. The catalogue's 164,668 stars against Fooocus's 52,538 are the size of that accumulated tooling, and it is real, even if much of it was last updated in 2024.

Both are ageing, differently

A1111's releases have slowed to a trickle and it never gained Flux support; the active successor is Stable Diffusion WebUI Forge, a fork by the same lllyasviel that kept the A1111 interface and did add newer models. Fooocus itself went into limited long-term support in 2024 with bug fixes only. Neither will run the video models or newer image architectures that dominate the image generation category now. A1111 vs ComfyUI covers why ComfyUI became the default and ComfyUI vs Fooocus covers the beginner side of that; this comparison is for people who have decided they want a form, not a graph.

Installing and living with each

A1111 is a git clone plus webui.sh or webui-user.bat, which builds a Python venv and pulls several gigabytes of dependencies; the catalogue's "heavy Python dependencies" con is where most support threads start, usually around torch and CUDA versions. It serves on port 7860. Fooocus is a 7z archive on Windows or a clone plus one pip install on Linux, downloads its default SDXL checkpoint on first run, and serves on port 7865. Its "low maintenance" highlight is accurate for the same reason it is frozen: nothing changes, so nothing breaks. On VRAM, Fooocus's automatic offloading gives it a documented 4 GB Nvidia floor, while A1111 needs --medvram or --lowvram launch flags before an SDXL model is comfortable on the same card.

Decision table

What I'd do

For a single user who just wants pictures, Fooocus. For anyone who has reached for a slider in the last month, I would skip plain A1111 and install Forge instead, which keeps the same interface and most extensions while running newer models faster on the same GPU. The one case where I would still stand up original A1111 in 2026 is an extension that has not been ported.