• The program is tested to work on Python 3.10.6. Don't use other versions unless you are looking for trouble.

• The program needs 16gb of regular RAM to run smoothly. If you have 8gb RAM, consider making an 8gb page file/swap file, or use the --lowram option (if you have more gpu vram than ram).
• The installer creates a python virtual environment, so none of the installed modules will affect existing system installations of python.
• To use the system's python rather than creating a virtual environment, use custom parameter replacing set VENV_DIR=-.
• To reinstall from scratch, delete directories: venv, repositories.
• When starting the program for the first time, the path to python interpreter is displayed. If this is not the python you installed, you can specify full path in the webui-user script; see Command-Line-Arguments-and-Settings#environment-variables.
• If the desired version of Python is not in PATH, modify the line set PYTHON=python in webui-user.bat with the full path to the python executable.
  • Example: set PYTHON=B:\soft\Python310\python.exe
• Installer requirements from requirements_versions.txt, which lists versions for modules specifically compatible with Python 3.10.6. If this doesn't work with other versions of Python, setting the custom parameter set REQS_FILE=requirements.txt may help.

Some error has occurred but the window closed too fast we can't see the issue, we need to prevent webui-user.bat form closeing instantly.

1. Right click to edit webui-user.bat.
2. Add a command pause at the end of the file.
3. Save the modification and run webui-user.bat again.

webui-user.bat should now pause allowing you to see the issue.

@echo off
set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=

call webui.bat
pause

Running as root is not recommended, but can be overriden by adding -f to your arguments.

./webui.sh -f

When running on video cards with a low amount of VRAM (<=4GB), out of memory errors may arise. Various optimizations may be enabled through command line arguments, sacrificing some/a lot of speed in favor of using less VRAM:

• Use --opt-sdp-no-mem-attention OR the optional dependency --xformers to cut the gpu memory usage down by half on many cards.
• If you have 4GB VRAM and want to make ~1.3x larger images, use --medvram.
• If you have 4GB VRAM but you get an out of memory error with --medvram, use --lowvram --always-batch-cond-uncond instead.
• If you have 4GB VRAM and want to make images larger than you can with --medvram, use --lowvram.
• If you have 4GB VRAM and get an out of memory error when loading a full weight model, use --disable-model-loading-ram-optimization (added in v1.6.0)

This is one of the most frequently mentioned problems, but it's usually not a WebUI fault, there are many reasons for it.

• WebUI uses GPU by default, so if you don't have suitable hardware, you need to run on CPU.
• Make sure you configure the WebUI correctly, refer to the corresponding installation tutorial in the wiki.
• If you encounter this issue after some component updates, try undoing the most recent actions.

If you are one of the above, you should delete the venv folder.

If you still can't solve the problem, you need to submit some additional information when reporting.

1. Open the console under venv\Scripts
2. Run python -m torch.utils.collect_env
3. Copy all the output of the console and post it

Video cards Certain GPU video cards don't support half precision: a green or black screen may appear instead of the generated pictures. Use --upcast-sampling. This should stack with --xformers if you are using. If still not fixed, use command line arguments --precision full --no-half at a significant increase in VRAM usage, which may require --medvram.

This is the same problem as the one from above, to verify, Use --disable-nan-check. With this on, if one of the images fail the rest of the pictures are displayed.

It is either a model cause - resource

Merge cause - resource

Or GPU related.

• NVIDIA 16XX and 10XX cards should be using --upcast-sampling and --xformers to run at equivalent speed. If issue is persisting, try running the vae in fp32 by adding --no-half-vae If this fails, you will have to fall back to running with --no-half, which would be the the slowest + using the most gpu memory.

• AMD cards that cannot run fp16 normally should be on --upcast-sampling --opt-sub-quad-attention / --opt-split-attention-v1. The fallback order should ideally be the same as the one above. Following that, if it continues to fail, AMD users may need to utilize some trick like "export HSA_OVERRIDE_GFX_VERSION=10.3.0" specific to their GPU. It would be ideal to do a thorough google search + all of github search to find the HSA_OVERRIDE_GFX_VERSION right for your specific GPU.

(These are word-of-mouth troubleshooting tips. Test with a fp32 4gb SD1 model)

Your installed xformers is incompatible with your GPU. If you use Python 3.10, have a Pascal or higher card and run on Windows, add --reinstall-xformers --xformers to your COMMANDLINE_ARGS to upgrade to a working version. Remove --reinstall-xformers after upgrading.

If you use Windows, this means your Python is too old. Use 3.10

If Linux, you'll have to build xformers yourself or just avoid using xformers.

Windows defender/antiviruses sometimes blocks Gradio's ability to create a public URL.

1. Go to your antivirus
2. Check the protection history:
   
3. Add it as an exclusion

Related issues:

This issue has been noted, 3 times. It is apparently something users in china may experience. #8537