20 times faster for SDXL model with Stable Diffusion A1111-small trick You Want to Know

After the official release of SDXL model 1.0 for use, it seems that Stable Diffusion WebUI A1111 experienced a significant drop in image generation speed, especially when compared to ComfyUI. However, this slowdown might be due to incorrect configuration settings. For my RTX 2070 graphic card, I made some modifications, and the image generation speed increased incredibly by 20 times. It is much faster than ComfyUI now.
A1111 is actually quite easy to get the optimized parameters for generating good images. Therefore, the workflow could be adjusted as follows: Generate the base image using WebUI A1111 and then perform upscaling to the final resolution using ComfyUI.

Here are some commands that might be useful for you (use them at your own risk). If something goes wrong, the best solution may be to delete the venv folder and then run the app again.

webui-user.bat setup
sdp works as xformers, improve the performance, so do not use them at same time, especially for high-end Nvidia graphic card

1 use xformers (works for me, 20 times faster)
set COMMANDLINE_ARGS=--xformers --medvram --no-half-vae
//low VRAM

or set COMMANDLINE_ARGS=--xformers --no-half-vae --lowvram //very low VRAM (such as 4 G)


2 Use SDP, Torch2.0, don't add xformers (high-end graphic card, such as RTX 40 series, I have not tested)
set COMMANDLINE_ARGS=--opt-sub-quad-attention --no-half-vae --medvram

set COMMANDLINE_ARGS=--opt-sdp-attention --medvram

At the end of the video, I have a small gift for you. Hope you will like it.


I do labor work, you do initial creation
------from Tube Underdeveloped　

Subscribe to my channel
   / @tubeunderdeveloped  

Now you can download more than 120 unique models for free https://www.thingiverse.com/zhwang168...

Twitter
  / play4fun168  

Facebook
  / zhwang168  

#TubeUnderdeveloped #Howto #3dPrinting #AICorner