Important update, the latest code commits done by lshqqytiger broke the DDIM image generation (May 8th 2023). The working version that works with the Ryzen 5600G APU can be found at my github forked repo (MAKE SURE TO USE THIS SPECIFIC BRANCH!!!): github.com/risharde/stable-diffusion-webui-directml/tree/master-without-36db52a
I downloaded version of july 28 and it works on my ryzen 4700u... Im even trying SDXL as we speak, but need to upgrade DRAM. 16GB means 8GB VRAM tops and its not enough for SDXL
Thanks so much for the dude! I'd been trying to get SD to work with rocm in Linux on my 5600G for over a week, and no dice. Used your tutorial and was up and running in an hour! Generations are about twice as fast on the higher resolutions, thanks again man 🙏
Thanks for the info, it was a lifesaver. Another tweak for making it run faster was going to settings, stable diffusion, "Checkpoints to cache in RAM" were set to 2. I sliced a third of the rendering time.
I used your method its pretty nice, but for today automatic1111 is better even in apu systems with your " --lowvram --precision full --no-half --opt-sub-quad-attention --opt-split-attention-v1 --disable-nan-check " codes. Just install the regular Automatic1111 for amd users, and go to settings, optimizations and set token merging values up to 1, i prefer 0,6. Its faster this way. And try custom realism models with euler 20 sample, default stable diffusion model is kinda slower. My apu is ryzen 3 2200g with 2x8gb ram and works fine. edit: do not use --xformers code.
@@kilobyte__ it was the case, yes. But things changed i guess and regular stable diffusion works really slow now. I am trying to reinstall this version instead with new tricks ill let you know if it works. (search for latent consistency lora and Lcm method)
The normal web ui is working the same way in my 5600g apu My commmand arguments is :set COMMANDLINE_ARGS= --lowvram --precision full --no-half --skip-torch-cuda-test
Just for context, i have an rtx2060s 8gb and with it i can make 4 images of around 768x512 30 steps in about 30 secs or so. So if you want to really use SD at its full potential you need an Nvidia gpu atm, if you going to buy new ish a rtx3060 12gb would a good budget option.
@@reptiloidtill yup, ive tried it on my rx580 8bg and at first, with no special commands i could do only like 400x400 images, 1 at a time or it would complain about memory. Tried a litle more and with a few commands i can do a bit higher res images but its painfuly slow. If didnt had tried it on nvidia first maybe i would not complain, but o well, once you has proven the fast lane its hard to use the slow lane lol.
@@maitotechlab9035I think that generation might not have any AI hardware. On my newer Rx6000er it works as expected, after a bit of trouble during setup. RX7000 seems to be on par performance wise with Nvidia, at least compared to similar priced ones. But still somewhat janky
Sir, but I've found it not using iGPU. 8845hs w 780m and in task manager it shows sd is only using cpu for generation. Any ideas how to make igpu run it?
Is there a way to make Easy Diffusion work with ryzen APU? I mean a way that i can do in order to not wait for the devs to implement it. Otherwise, is there a way to use the files I already have downloaded/installed with ED to speed up the process of using this ui?? In order to not download a second time the same files
I would give you your like. but something that never happened to me with the automatica github repositories happened with yours, the first time I run the webui and the windows firewall alert for python appears, which never happened to me with the original versions of automatica, but you deserve it a like because with your command line, if the original version works for amd without the command line the vram goes full and throws a vram error, Be careful with the repository
everything went smoothly until I tried to generate an image this error popped up, "RuntimeError: [enforce fail at ..\c10\core\impl\alloc_cpu.cpp:72] data. DefaultCPUAllocator: not enough memory: you tried to allocate 58982400 bytes." im using the same apu as you R5 5600g and 16gb ram. is there a way to put memory limit manually? couldn't find any help on other posts on github or reddit :(
Does this method also works for AMD GPU? I run SD on my pc with no problem with ryzen 5 2600 RTX 3060, I just bought 2nd pc with Ryzen 5 5600G in it and I'm planing to change side and buy AMD Gpu for it later, so I try this method first
Good question bro! Not too sure, I saw that but didn't try it as yet, in theory pip new versions should break anything so if you feel adventurous you could try it
Would love to, to answer any doubts but youtube will destroy my channel if it doesn't get engagement, already struggling as is with my content, sorry for the lame excuse
hey bro its giving me some error Couldn't launch python exit code: 9009 stderr: Python was not found; run without arguments to install from the Microsoft Store, or disable this shortcut from Settings > Manage App Execution Aliases.
@@Bhushanmilkhe Assuming you're running windows, did you install python? Make sure when installing you enable the checkbox on the first window of the installer to include python in PATH (Download python installer here: www.python.org/downloads/release/python-3106/). This should work - good luck!
@@rthd oh bro thanks for the help. But my stable diffusion was crashing after hitting the generate button it was saying out of memory. I have ryzen 5 3400g with vega 11 and 16gb ram
Did you adjust the command line arguments to what I mentioned I the video and use the latest repo link? That should have avoided the crashes, please let me know, thanks
so we can use your link for the actual thing to wokr on a APU, i have and APU ready to go for this!!!!! im gonna use your link !!!! and get back to you on this comments
Good question, I can't say for certain but I'd say there's a high probability that it should work. I do have a 4500U which in theory should be similar - if I get a chance I'll try it and get back to you or maybe you can try the code and let me know if it works for you!
Bro I got a problem, when I run webui-user.bat it get stuck in Can't run without a checkpoint. Find and place a .ckpt or .safetensors file into any of those locations.
