Note: This is not my model - this is a preview version of waifu-diffusion-xl https://huggingface.co/hakurei/waifu-diffusion-xl
This model exists under the SDXL 0.9 RESEARCH LICENSE AGREEMENT due to the repository containing the SDXL 0.9 weights.
wdxl-aesthetic-0.9 is a checkpoint that has been finetuned against our in-house aesthetic dataset which was created with the help of 15k aesthetic labels collected by volunteers. This model also used Stability.AI's SDXL 0.9 checkpoint as the base model for finetuning.
Description
FAQ
Comments (13)
Quick question, do I use this with the SDXL refiner or just on its own ?
I've been using it with the refiner - but the sample workflow provided on the WDXL repo doesn't have it! I've been getting better results running through the refiner though, for sure.
@theally awesome, thank you 😊.. just tested it out with the refiner.. works well.
Just curious, why is this one double the size of normal SDXL models? Can it still run on 24GB cards?
@ConsumeEm it runs fine on 10GB (in Comfy), and it's probably so big because it's unpruned - contains training data which isn't used for inference. It's not a great model though, but it will improve over time I'm sure.
Can this be used in Automatic11111? If yes, where do I install it? Thanks :)
It can’t be used in the current release (1.4) of A1111, but it can be used in the 1.5.0-RC release candidate. It goes in models/stable-diffusion just like any other checkpoint model. SDXL has its own VAE and earlier VAEs don’t work right with it, so you need to either have “None” set for your VAE to use the built in VAE, or have an SDXL-compatible VAE for it to use.
@cmdicely850 Thank you very much for the info :)
Let me be the first weeb to say that I hate the art direction this went in. "When you try to be everything to everyone, you end up being nothing"
Yeah, I agree - big disappointment. WD 1.4 was huge for the scene when Stable Diffusion first came out, and I was expecting big things - maybe they've refined it since I last looked? Will check it out!
Maybe I'm missing something but to be honest this is the hardest anime SDXL model I've ever tried to use. Almost impossible to make something good with it without loras which defeats the purpose of a dedicated checkpoint. Other anime models just work, this one you have to grind like horse during wild wild west movie just to get scraps.
This is true! When the WD 1.5 version came out it was great, so I was expecting big things from WDXL, but... meh!
hey, this was released before SDXL 1.0, it shouldnt have been uploaded here lol; WDXL 1.0 will happen