theally
patreon
SD XL 0.9
🕑 Added 2023-06-25 17:23:18 +0000 UTCHey 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!)
If you're a researcher and want to try to get your hands on the weights early you can apply here at the Stability AI Huggingface page.
Comments
Oooh, it looks a little like when you force SD 2.x to show nakedness. But still cool!
Russ Matthews
This may be of interest in the IS IT NSFW discussion.
Russ Matthews
https://youtu.be/NPrXb8yq16A
I've trawled through the social media for all the StabilityAI employees I could find. They give away a lot of little details, but not a single mention of Mac support I'm afraid!
I noticed their announcement specifically called out Windows and Linux. Do we know anything about M1/M2 Mac support?
