There will be native sdxl support and in the hires. fix the model can be changed. See:
Use the SDXL Refiner Model for the high-res fix pass. Hopefully, we won't need this extension for a long time.
You need to enable high-res fix.
32GB RAM (Not VRAM)
SDXL requires a lot of RAM so the more the better. With 32GB, it will use ALL of your RAM. Close everything you don't need to have open. <br>
Also, use this pull request with 32GB RAM to reduce RAM usage.
Use this VAE (fp16)Set batch count to 1 (for 32GB RAM)Other settings as you would normally
My settings:
Upscaler: LatentUpscale by: 1.25Denoising: 0.53
The default values can be changed in the settings
Of course, this extension can be just used to use a different checkpoint for the high-res fix pass for non-SDXL models.