I started Stable Diffusion with EasyDiffusion. Nice tool, but not enough for my needs. Automatic1111's WebUI got my attention quickly, and then here I am. Started using it a bit, and it works perfectly for simple gens. Although, I'd like to experiment different styles and models, and some of them need the hires options, which unfortunately started generating a lot of errors on my end telling me to give it more Vram basically. I re-read the doc, and online answers, which all point to adding commandline args, and in the end I tried different variations of all those : --xformers --lowvram --medvram --precision full --no-half --no-half-vae --opt-split-attention
There's always something that goes bad though. First, my it/s slowed by like 2/3 times peaking at >8s/it... Quite slow for a 16gb VRAM Quadro P5000. Second, I don't have the same error, sure. I get new ones : "NansException", telling me to add yet another commandline --disable-nan-check, which only helps at generating grey squares over 5 minutes of generation. Quite inefficient, I do it faster by hand.
My best results are with my webui-user.bat looking like this :
@echo off
set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers --no-half --autolaunch
set PYTORCH_CUDA_ALLOC_CONF=garbage_collection_threshold:0.9,max_split_size_mb:1024
set ATTN_PRECISION=fp16
call webui.bat
Traceback :
Traceback (most recent call last):
File "C:\Users\Shadow\stable-diffusion-webui\modules\call_queue.py", line 57, in f
res = list(func(*args, **kwargs))
File "C:\Users\Shadow\stable-diffusion-webui\modules\call_queue.py", line 37, in f
res = func(*args, **kwargs)
File "C:\Users\Shadow\stable-diffusion-webui\modules\txt2img.py", line 56, in txt2img
processed = process_images(p)
File "C:\Users\Shadow\stable-diffusion-webui\modules\processing.py", line 526, in process_images
res = process_images_inner(p)
File "C:\Users\Shadow\stable-diffusion-webui\modules\processing.py", line 684, in process_images_inner
devices.test_for_nans(x, "vae")
File "C:\Users\Shadow\stable-diffusion-webui\modules\devices.py", line 156, in test_for_nans
raise NansException(message)
modules.devices.NansException: A tensor with all NaNs was produced in VAE. Use --disable-nan-check commandline argument to disable this check.
