theally

SD XL 0.9

Added 2023-06-25 17:23:18 +0000 UTC

Hey all! Quick update on SD XL, with all the information I've been able to gather from around the place - including the whispers and rumors! If you don't know what this is, it's StabilityAI's new model, and promises to be a huge leap forward in AI image generation tech.

It's already available to try on Clipdrop. You can generate a few (four or five) SD XL images here as an anon user, without an account.
It will be coming to DreamStudio Monday 26th June, and the weights (model) will be released to us mid July sometime.
It works in Auto1111 WebUI, and even better in ComfyUI - time for us to revisit that!
In requires a minimum of 8GB VRAM (Nvidia) or 16GB VRAM with an AMD card - sorry older GPUs!
It'll be released as .safetensors only, no .ckpt.
Kohya has a trainer ready for training LoRA and creating DreamBooth finetunes.
Textual Inversion compatible.
ControlNet compatible.
It has Noise Offset built in.
It can create 4k size images without upscaling.
The big rumor is that it will not filter out NSFW - we might be getting an uncensored model.
6 finger hands a thing of the past (I'll believe it when I see it!)