Fix Automatic1111 Git Pull Break: Venv Reset and Rollback

TL;DR

• Delete venv first — the official A1111 docs list venv deletion as the canonical repair path; it is faster than hand-pinning packages.
• Models and settings survive — they live in models/, config.json, ui-config.json, styles.csv outside the virtual environment ([see /INTERNAL:models-folders]).
• Forge requires update.bat — skip it and you run a stale build with unfixed bugs ([see /INTERNAL:forge-vs-a1111]).

Why git pull Breaks Automatic1111

A git pull updates the Python code and extensions but does not rebuild the virtual environment. If upstream changed a dependency version — PyTorch, xformers, Gradio, anything in requirements_versions.txt — the old wheels inside venv now mismatch the new import statements. The launcher prints a wall of RequirementError or ImportError and exits.

The A1111 Troubleshooting wiki states it plainly: "If you encounter this issue after some component updates, try undoing the most recent actions. … you should delete the venv folder."¹ That is the entire rollback playbook from upstream.

What the Civitai Articles Get Wrong

Fastest Repair: Delete venv and Relaunch

1. Close the WebUI console window.
2. In the repository root (stable-diffusion-webui/), delete the venv folder.
3. Double-click webui-user.bat (Windows) or run ./webui.sh (Linux/macOS).
4. The launcher recreates the virtual environment and installs the exact versions pinned in requirements_versions.txt.

Why this works: A1111 isolates every Python package inside venv. The repo itself only holds code, extensions, and a few scripts. Your checkpoints, LoRAs, embeddings, and UI settings live in models/, embeddings/, config.json, ui-config.json, styles.csv — outside venv. Deleting venv never touches them.⁶

Community confirmation: Multiple guides (Civitai 1024, YouTube, Stack Overflow) recommend venv deletion as the first step after a broken pull.⁷

If the repair keeps fighting you, moving on is a valid answer — see leaving A1111 in 2026.

When You Must Reinstall PyTorch Manually

On slow or air-gapped connections, the launcher's automatic pip step can time out. A1111 exposes two environment variables for this:

Edit webui-user.bat / webui-user.sh and add:

set TORCH_COMMAND=pip install -U torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

The cu121 index hosts official wheels for torch, torchvision, torchaudio, and xformers⁸ — no third-party mirrors needed.

Once it runs again, the acceleration stack version matrix explains which xformers build matches your PyTorch.

xformers on Windows: Current State (2024+)

Forge: Different Update Workflow, Same Repair Logic

Forge bundles Git, Python, and a pinned PyTorch in a one-click 7z package. The critical difference:

Forge docs warn: "Running update.bat is important, otherwise you may be using a previous version with potential bugs unfixed."¹²

If update.bat was skipped, re-extract the 7z or delete venv and run update.bat → run.bat.

Common Errors and Fixes

Decision Table: Repair vs Reinstall

FAQ

How do I roll back Automatic1111 after a bad git pull? Upstream does not publish a numbered git reset --hard playbook. The documented recovery is: undo recent actions, then delete venv.¹ If you know the good commit hash, git reset --hard <hash> works, but you must still delete venv afterward.

Will deleting venv erase my models and settings? No. Models live in models/Stable-diffusion/, LoRAs in models/Lora/, settings in config.json, ui-config.json, styles.csv — all at repo root, outside venv.⁶ See also: Models and stable diffusion model folders — what survives a reinstall?

Why does the Civitai guide say to add git pull to webui-user.bat? That is a community workaround, not an official feature. The wiki never mentions it.²

Do I need Visual Studio Build Tools for xformers on Windows in 2024? No. Since January 2023, an official wheel installs via pip automatically when you launch with --xformers.⁴

What Python version does Automatic1111 require on Windows? 3.10.6 exactly. The installer checks for it; other versions cause xformers and torch wheels to fail.¹

Forge vs Automatic1111 — which is easier to repair? Both reset via venv deletion. Forge adds a mandatory update.bat step; A1111 relies on git pull. No primary source claims Forge is faster to repair.¹²

Where do I put --xformers on Windows? Edit webui-user.bat, find set COMMANDLINE_ARGS=, append --xformers (space-separated). Example: set COMMANDLINE_ARGS=--xformers --skip-torch-cuda-test --autolaunch.⁵

My generated images are black/green after an update — what flag fixes it? Try --no-half-vae first; if that fails, --upcast-sampling or --no-half.¹⁴

Sources

1. <https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Troubleshooting> — official recovery steps, venv deletion, GPU troubleshooting, ZIP git fix, NaN/black-screen mitigations, xformers reinstall flag.
2. <https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-NVidia-GPUs> — Windows install prerequisites, COMMANDLINE_ARGS examples, default local URL.
3. <https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Command-Line-Arguments-and-Settings> — environment variables and CLI flags.
4. <https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Xformers> — Jan 2023 wheel transition, Windows manual-build history.
5. <https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Command-Line-Arguments-and-Settings> — COMMANDLINE_ARGS usage examples including xformers.
6. <https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Command-Line-Arguments-and-Settings> — --models-dir, --ckpt-dir, --styles-file, --ui-config-file, --ui-settings-file paths proving user data lives outside venv.
7. <https://www.civitai.com/articles/1024> — community repair guide recommending venv deletion and manual pip diff.
8. <https://download.pytorch.org/whl/cu121> — official PyTorch CUDA 12.1 wheel index listing torch, torchvision, torchaudio, xformers.
9. <https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Troubleshooting> — "If you use Windows, this means your Python is too old. Use 3.10" (xformers wheel requirement).
10. <https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Optimizations> — benchmark table: xFormers 6.5/7.5 it/s vs Doggetx 4.1/6.6 it/s on RTX 3060.
11. <https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Troubleshooting> — --reinstall-xformers flag usage.
12. <https://github.com/lllyasviel/stable-diffusion-webui-forge#installing-forge> — Forge one-click package, mandatory update.bat, run.bat, advanced git-clone method.
13. <https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Troubleshooting> — fatal: not a git repository after ZIP extraction and git restore commands.
14. <https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Troubleshooting> — NaN, black/green screens, --no-half-vae, --upcast-sampling mitigations.
15. <https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Troubleshooting> — Torch unable to use GPU after updates; venv deletion fix.
16. <https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Troubleshooting> — reinstall from scratch: delete venv, repositories.
17. <https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Troubleshooting> — model corruption (MetadataIncompleteBuffer, PytorchStreamReader) and redownload guidance.