Support Me: Ko-Fi
Help me with good review and 5-stars. Thanks.
Finetuned SDXL with high quality image and 4e-7 learning rate.
VAE: Here
Check my other SDXL model: Here
The Umbrella LoRA SDXL: Here
ComfyUI Workflow: Here
A1111 Settings: Soon
Description
FAQ
Comments (8)
Why are all the images I generate coming out as noisy, unfinished templates that look very scary no matter how I adjust the settings? My computer configuration is sufficient, and there are no issues running other models. I personally really like the results you have shown, even re-downloaded three times. I really want to know what the problem is
are you use ComfyUi or A1111 webui?
@GorillaMonsoon I used the web UI, and it seems to be the new updated version 1.5.1. The issues I was experiencing were actually due to my own settings, but now I have resolved them and everything is working fine. By the way, I wanted to mention that the models you have released are really great, and I really enjoy the style they produce
@killers23 there’s new version on dev branch of A1111 webui. if you don’t want to use comfyUI directly, you can try StabilityAI’s StableSwarm
you might be using an SD1.5 VAE. set your VAE back to Automatic in the settings.
this model is not yet good enough for SDXL imo, can output good images, but flexibility is not there
remember, it’s finetuned. not dreambooth or merged model. of course it’ll be overfitted due to low training steps comparing StabilityAI massive training steps for SDXL.
if you want to get better result for flexibility, just merge multiple times of multiple finetuned SDXL 0.9 and SDXL 1.0. And after that, use as pretrained model with your preferred dataset and tags. It’ll be much flexible for prompting than any sdxl finetuned model.