SPEED UP Stable Diffusion with these tips!
Вставка
- Опубліковано 12 лип 2024
- Stability AI has just publicly released SDXL 1.0, and for those of us who don't have earth shattering rigs, things are a little... slow. Here are 8 tips to help speed up things a bit and get you back to making beautiful AI Art! Follow this guide if your struggling to run Stable Diffusion on automatic1111 with low VRAM!
RESOURCE LINKS:
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
- Automatic 1111 on Google Colab Tutorial: stable-diffusion-art.com/auto...
- Python 3.10.6: www.python.org/downloads/rele...
- Automatic 1111 GitHub: github.com/AUTOMATIC1111/stab...
Thanks for watching! it's my first shot at a UA-cam so apologies if it's a bit rough around the edges 😅 Let me know what you think and what else you'd like to see!
CHAPTERS
▬▬▬▬▬▬▬▬▬
00:00 START
00:10 About SDXL1.0
00:24 Optimise your workspace
00:46 medVRAM and xformers
01:24 No module 'xformers'. Proceeding without it.
01:50 Chrome Hardware acceleration
02:40 Disable A1111 Extensions
03:07 MiaoshouAI Assistant
03:42 Update A1111
04:10 Google Colab
▬▬▬▬▬▬▬▬▬ - Наука та технологія
excellent tutorial!
Hi, thanks a lot for the tutorials and effort you put in your videos they are great :D.
I was wondering if you could help me with this, I have downloaded "Miashou Assistant" but the butons for "Enable Auto Memory Release" and "Forc VRAM Release" are missing, do you know how to make them appear? :S
thanks
Is there any way around the following error?:
"RuntimeError: Could not allocate tensor with 83880960 bytes. There is not enough GPU video memory available!"
I can generate images in SD 1.5 using a resolution of 450x325, but the 1024x1024 requirements of SDXL appear to be too much for my 3 year old GPU...
It all comes down to available VRAM. The trick is to free up as much as you can before generating, which is the goal of the most of the steps in the video. I'm not sure how much VRAM you have, but try to make sure you have nothing in the background using your GPU, it's suprising what might actually be occupying that space, anything from Chrome to Slack! You can also try --lowvram instead of --medvram in your launch settings. SD also offloads work to RAM so make sure you reduce RAM usage in the backround too. I've heard people having success in ComfyUI with as little as 4GB VRAM, but I don't feel like digging out my GTX 970 to test xD
@@InciteAI Thank you for the tips and you time. o7
My GPU has 8gb of RAM and it's an AMD card so ComfyUI will be a no-go for me until either I upgrade to an nVidia card or they update for AMD support. Not holding my breath on the latter there.
👋
not bad presentation. nice clear voice!!! Clear and concise. Great Tip about MiaoshouAI Assist. That was a bonus!!! Render times to under 4minutes!!!! Processor AMD Ryzen 7 4800H with Radeon Graphics, 2901 Mhz, 8 Core 4gbVRAM. Would like to see more of tweaking, best settings, particulalry around using MiaoshouAI? and maybe some Comfy UI. Big question here is that Comfy portable appears to generate censored models? Any info on that?? SUBBED!!!
Glad you liked, and happy your getting faster generations! I've got some video's planned that should hit most of those topics! I haven't had that issue with ComfUI myself, even with the base SDXL model. I know SDXL is meant to respond better to simpler prompts, so if yours is really detailed, it might be worth trying to simplify it a bit?
how can i downgrade the python version . i tried but webui did not recognize
0:23 well default new instal should be fast. you can allways change something get faster removing quality xD
that look slower than my rtx 2060, but rtx 4090 its blink of eye. why i need make faster than blink LOL xD. its not just 4090 it whole new PC. price was 2x more. have to be faster bcoz usually things cost same and you get faster, but yeah 2k gpu blink of eye vs. what eva 2060 cost 500 back then LOL
everything is just bokeh and plastic on sdxl.
The base model certainly has a ...look, but as we see more community generated stuff I would hope we get more variations!