Haven't tested yet but my guess is your base resolution too low, and I'd imagine the glow is from noise offset. If you can't generate at higher res this might not be a good model.

@Aitasai Damn I was worried about that, I can't run Hires fix any higher than that with my 1660Ti as it runs out of Vram x-x I guess with that there's no point using Hires if it just ruins the image like that. Would this still work without Hires? Or is that pretty much needed? Regardless thanks for the help finding the issue!

@Aitasai Were would I look to check settings for HiRes fix? whenever I try to run it higher than 326x328 I get a message in CMD that says:

"torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 2.00 GiB (GPU 0; 6.00 GiB total capacity; 3.35 GiB already allocated; 243.73 MiB free; 3.94 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF"

That was at base res of 512x512 And it says that everytime it gets to the HiRes portion of generation.

@Aitasai Tried it with those settings in commandline Arg and it still comes up with teh same message, I had medvram already so I added in the precision and no-half. I also had Xformers in there too.

Would changing the batch size and batch count help? I usually try to run it with Batch count 2, batch size 4 so I don't have to run multiple times for more than one image.

Update: well turns out it does but I have to set the Size to just 1 for it to work, but the image does look 1000x better now xD Guess it was my fault for trying to do larger Sizes, I just have to run the Batch count as the higher number instead. Sorry for being dumb xD But thanks for the help getting this sorted out, will help with not only this but other models I've been trying to use :D

@Hybridwolf You can also search for "VRAM usage polls" on the settings and increase it, setting it to max should be fine. It (should) help prevent VRAM related crashes.

I can't remember if it needs xformers or don't work with xformers, but increasing it won't hurt anyway.

求群名啊，想一起交流

@Aitasai  there is one sample image i think u didnt use 84000 :D probably because of that i felt like :)

Check your rendering size. I think they were trained on 512. I normally use native 768x1024. But I was getting blurry pics on both models with that. I changed to 512x768 and then upscale to 691x1036

@Rachey13x  I tried this but I'm still getting some blurry artifacts and odd contrast, could you list your exact settings ?

@Rachey13x ty , for me the brighter one works best. still getting some blurry stuff, i used 512x512 -> and 512x768

@DetaKongz you might want to change ur sampling method, try Euler a at 25 -35 steps that's usually works out for me 512x768

yes

@MoltenHeart ty, any other good pointers with the content of this page? or a good tutorial

@belmont725 Well I know a bit but I'm no expert. Are you familiar with this webpage? https://huggingface.co/hollowstrawberry/stable-diffusion-guide/blob/main/README.md#index It has a lot of good info on it.