If you are choosing a Stable Diffusion front end today, the Automatic1111 vs WebUI Forge question has a short version: Forge is built on top of the same WebUI, so the screens look familiar either way. The decision is about upkeep, memory handling and which model families you want to run, and the project pages answer part of it. They give no speed benchmarks, so none appear below.

What you need before you start

  • An Nvidia card is the main target for both. The Automatic1111 README lists separate guides for AMD, Intel and Apple Silicon.
  • Git, and Python. That README names Python 3.10.6 on Windows and says a newer Python does not support torch.
  • Disk space for checkpoints, and time to read the memory notes in our GPU and VRAM guide.

Step 1. Know what each project is

Automatic1111 WebUIThe original Gradio-based WebUIReleases page lists 1.10.1 as the latestMemory control through launch flagsWindows install with Python 3.10.6 or a zippackageWebUI ForgeA platform built on top of the WebUIBased on WebUI 1.10.1, synced every 90 daysGPU Weight slider and swap toggles in the UIOne-click package with Git and Python included
How the two projects describe themselves, side by side

The Forge README says it aims to make development easier, optimize resource management, speed up inference and test experimental features. It is named after Minecraft Forge. It also says it syncs with the original WebUI every 90 days or when important fixes arrive. As of November 2026 the Automatic1111 releases page lists 1.10.1 as its latest release, a single bug fix for CPU upscaling. Version 1.10.0 before it added Stable Diffusion 3 support, with Euler recommended.

Step 2. Compare the parts that differ

AreaAutomatic1111WebUI Forge
InstallClone the repo and run webui-user.bat, or use the sd.webui.zip package, update.bat, then run.batOne-click package with Git and Python, or a Git clone and webui-user.bat
Memory--xformers, --medvram, --lowvram launch flagsGPU Weight slider, queue and async swap toggles
Newer modelsStable Diffusion 3 added in 1.10.0Flux support: NF4 and GGUF formats
ExtensionsBuilt-in extensions folder plus a folder for your own extensionsREADME reports Adetailer and integrated extensions working in its last tests
Known gapsNone listed on the READMEControlNet Union and ControlNet Flux not implemented, OFT LoRAs broken

Source: the Automatic1111 README, wiki and releases page for the left column, the Forge README for the right (both listed in the sources of this note).

On the extension row, be careful. Forge's own status table has test dates mostly in August 2024 or earlier, so "working" describes that time, not today. If one extension decides your choice, check its page first.

Step 3. Install Automatic1111

  1. Install Python 3.10.6 and tick "Add Python to PATH". Install Git.
  2. Clone the repository, or download sd.webui.zip from the v1.0.0-pre release, run update.bat, then run.bat.
  3. With the clone route, start webui-user.bat as a normal user, not as administrator.
  4. For a small card, set the flags in webui-user.bat on one line, for example set COMMANDLINE_ARGS=--xformers --medvram. The wiki describes --medvram as sacrificing some performance for low VRAM and --lowvram as sacrificing a lot of speed.

Step 4. Install WebUI Forge

  1. Download the one-click package. The README recommends the build for CUDA 12.1 and PyTorch 2.3.1.
  2. Extract it, run update.bat first because older versions may carry unfixed bugs, then run.bat.
  3. The CUDA 12.4 and PyTorch 2.4 build is labeled fastest, but the README warns MSVC and xformers may be broken. Start with the recommended build.
  4. Manual route: git clone https://github.com/lllyasviel/stable-diffusion-webui-forge.git, then webui-user.bat.

One mismatch to know about: the install lines mention update.bat and run.bat, while the repository file list shows webui.bat and webui-user.bat. If a script is missing, use the one the file list shows.

What usually goes wrong

  • Python newer than the version the project names. Install the listed one beside it and point the launch script at it.
  • Flux slowness in Forge. The README says lowering GPU Weight solves 99% of problems in its linked discussion, so try that before anything else.
  • Moving extensions from one to the other and expecting identical behavior. Forge integrates ControlNet itself, so read its notes before adding a second copy.

Which to pick

Choose Automatic1111 if you want the original, you follow its extension guides and you run SD 1.x or SDXL style models. Choose Forge if memory is tight or you want Flux, and you accept a younger platform with a status table to read. Nothing stops you from keeping both in separate folders, which is the cheapest way to settle it on your own card. For an example of Forge packaged inside another tool, see our Proteus V0.3 note.