SDXL 1.0 Tips in A1111 Low VRAM and other Errors and Refiner use cases for Stable Diffusion XL
#stablediffusion #A1111 #AI #Lora #koyass #sd #sdxl #refiner #art #lowvram #lora
This video introduces how A1111 can be updated to use SDXL 1.0, the various errors we get using Low VRAM of 8GB and how we may avoid them and run it on 8GB in Automatic 1111.
00:00:00 introduction, what we will see in this video
00:01:46 how to upgrade A1111
00:02:34 where to download SDXL and Refiner models from?
00:03:45 testing SDXL in A1111
updates(--medvram-sdxl is best for SDXL now)
00:05:27 testing the refiner model in img2img
00:06:44 using the refiner as a base model for txt2img
the Video will also show how Refiner can be used as a standalone without base model to avoid switching from base model to refiner and waste so much time in the process and still get high quality results.
Conclusion:
-8GB can run SDXL, for me, setting --medvram option made it possible
-refiner cannot use LoRAs
-base model cannot use most of previous SD models LoRAs
-SDXL have better grasp of human/animal anatomy way much better than all previous models.
-refiner can be used standalone and produce great results better than SD 1.5 and faster than base mode using the right work flow using 512x512 or 640x640 or 768x768 or any multiples of 128 in general
-cannot still use most of the extensions such as Controlnet-after detailer.
-very slow compared to SD 1.5 on 8GB RTX 3070 laptop graphics card compared to 1.5
if you get NaNs running A1111 with --no-half-vae option may fix it
stability ai SDXL models
https://huggingface.co/stabilityai
https://huggingface.co/stabilityai/st...
https://huggingface.co/stabilityai/st...
A1111 download page for fresh installation
https://github.com/AUTOMATIC1111/stab...
Computer Specs:
Laptop: Legion 5 Pro
Processor :AMD Ryzen 7 5800H , 3201 Mhz
System RAM: 16.0 GB
Graphics GPU: NVIDIA GeForce RTX 3070 Laptop GPU 8GB