Learn how to get Free YouTube subscribers, views and likes
Get Free YouTube Subscribers, Views and Likes

Do THIS to speed up SDXL image generation by 10x+ in A1111! Must see trick for smaller VRAM GPUs!

Follow
Keyboard Alchemist

#SDXL, #automatic1111, #stablediffusiontutorial,
Is your SDXL 1.0 crawling at a snails pace? Make this one change to speed up your SDXL image generation by 10x or more in Automatic 1111! I have a 8GB 3060Ti GPU and it was unbearably slow, until I put in this command line argument, and it sped up my image generation with the base model plus refiner by 10x to 14x for a single image generation. It's actually very simple. In this video tutorial, I will show you how to speed up your image generation, and how to install and use the Refiner model within Txt2Img without manually switching models.

UPDATE: With the release of A1111 webUI version 1.6.0, there is a new command line argument that we can use: 'medvramsdxl' instead of 'medvram'. 'medvramsdxl' will only enable model optimization for SDXL models. This is nice because if you wanted to generate images with SD v1.5 models and do not need to enable 'medvram', you won't need to manually edit your webuiuser.bat file (having 'medvram' enabled when you don't need it tends to slow down your image generation). I hope this helps, cheers!

If you have not installed SDXL 1.0 base and refiner models, watch my previous video for a stepbystep guide    • How to Install and Use SDXL 1.0 Base ...  

Chapters:
00:00 Intro
00:38 Add the magic words to 'webUIuser.bat' file
01:48 Image generation test with SDXL1.0
02:05 Only 41 seconds for a single image (11x faster)
02:20 Trying 'lowvram'
03:00 Does this negatively impact my Image Quality?
03:30 Comparing images generated with medvram ON or OFF
04:25 Install Refiner Extension
05:00 How to use Refiner Extension
06:07 Refiner works better with less steps
06:32 Images generated in SDXL 1.0 with the Refiner Extension

Useful links
A1111 Command Line Arguments:
https://github.com/AUTOMATIC1111/stab...

Refiner extension:
https://github.com/wcde/sdwebuirefiner

sdxlvaefp16fix on Hugging Face:
https://huggingface.co/madebyollin/sd...
The VAE fixes an issue where you will get an entirely black image when decoding in float16 precision.


**If you enjoy my videos, consider supporting me on Kofi**
https://kofi.com/keyboardalchemist

posted by olyhc1hr