-
-
Notifications
You must be signed in to change notification settings - Fork 445
FAQ
The functionality is integrated into the "Process Image" tabs, removing the need for a separate "PNG Info" tab.
Most command-line flags have been moved to UI → Settings. For a complete list, start the web UI with the --help flag.
Launch the app with the --debug flag and review the setup.log file for detailed insights.
1. Right-click on webui.bat and select "Create Shortcut".
2. Right-click the shortcut and open "Properties".
3. Add the options at the end of the "Target" field after a space. Example:
"C:\path\to\webui.bat" --medvram --autolaunch
4. Click OK and use this shortcut to start the app.
Add the --use-directml command-line flag when starting the app.
Render a smaller image (e.g., 512x512) and upscale it using the "Process Image" tab with an upscaler like SwinIR_4x. This method tiles the image to minimize VRAM usage.
Try a different VAE file. Add ,sd_vae to the Settings → User Interface → Quicksettings list to enable quick VAE selection. Place new VAE files in the models\VAE folder.
Run webui.bat --upgrade. This pulls the latest updates and applies them to your installation.
Use webui.bat --reinstall to reinstall required components or webui.bat --reset to reset to the latest version.
In Settings → Image Options, switch from PNG to JPG for smaller file sizes and disable "Always save all generated images." Save only desired images manually.
1. Render smaller images (512x512) and upscale later. 2. In Settings → Compute Settings, enable FP16 precision. 3. Use memory-efficient Cross-Attention Optimization methods like Sub-quadratic attention. Disable "SDP disable memory attention" if enabled.
Stable Diffusion models are typically trained on 512x512 images. Rendering at different aspect ratios or sizes may cause artifacts. Stick to standard sizes or upscale afterward.
"Hires Fix" upscales images during generation, avoiding post-generation artifacts from traditional upscalers. It complements, but does not replace, external upscalers.
CLIP Skip adjusts how detailed image generation becomes. Higher skip values can result in less specific but potentially higher-quality images. Some models benefit from specific settings. PS: CLIP skip is not needed while using SDXL in most cases.
There isn’t a definitive best, but "DPM++ 2M Karras" is commonly recommended as a reliable general-purpose sampler.
Create subfolders in the models directory and move ckpt files there. Restart the UI for changes to take effect. Use meaningful categories like "Photorealistic" or "Anime".
Add the --upgrade flag to your webui.bat launch parameters to automatically update on every startup.
This error is caused by a recent upstream Python library conflict.
Solution: Simply re-run webui.bat or webui.sh. The issue will resolve automatically, and SD.Next will work as expected.
The old clip-interrogator is incompatible with the newer transformers package required by SD.Next.
Solution:
If you manually installed clip-interrogator as an extension, remove it.
SD.Next now includes an updated version of clip-interrogator for new installations.
To manually update, run:
git submodule set-url extensions-builtin/clip-interrogator-ext Dahvikiin/clip-interrogator-ext.git webui.bat --upgrade
Then, re-launch SD.Next.
This issue is related to the Image Previews feature.
Solution: Ensure the preview method is not set to Full VAE. If the error persists, disable previews.
This happens when an extension has downgraded the required diffusers package.
Solution: Disable all user-installed extensions and re-launch SD.Next.
These errors occur because some legacy variables are not initialized when using specific --backend arguments.
Solution: Start SD.Next without any --backend arguments to initialize the variables. This issue will be addressed in future updates.
HIP SDK version 5.7.1 is recommended over version 6.1 due to its superior compatibility and more consistent performance in most scenarios.