got it working but its a bit laggy and slow but works . also how do i install manager into this i tried putting in from github but it wouldnt run said torch was missing and some lines of code line 77 or something flagged . and i couldnt restart after i closed it down said it couldnt find comfyui so i had to go through a few of the steps agin just to restart it
ERROR: Could not find a version that satisfies the requirement torch-directml (from versions: none) ERROR: No matching distribution found for torch-directml
I tried and tried for months to get ComfyUI to run stably on my AMD system (5900X/6900XT) with 32GB RAM and 16GB VRAM. It just wasn't worth it. DirectML is suuuuuper slow. I switched over to Linux and am running ComfyUI under ROCm, and it's about 10x faster than on Windows with DirectML. And I can generate much larger images without running out of VRAM.
There's several things missing. I attempted to run this but missing several modules you have on your Comfyui that are not installed by default. Also, no mention of lcm anywhere in this tutorial except in the beginning.
Sorry I didn't have time to add all the details. I think LCM has been added to ComfyUI main code. No additional installation is needed. For LCM videos, see ruclips.net/video/5NpffP9lJ9w/видео.html
thanks for trying to work on a different one. I can get fooocus almost there, but keep running out of memory, and can't scale the images down enough to not run out of memory, which is a bummer for 6700xt. can't wait to try this one.
It works! Thanks a bunch! Only question I have more is, would it be possible to make a shortcut to launch it? A .bat file or something, instead of activating the environment and typing out the line to boot it up :/
Running on 6700 12G, err show: comfyui\comfy\model_sampling.py:81: UserWarning: The operator 'aten::frac.out' is not currently supported on the DML backend and will fall back to run on the CPU. So finally DML allocator out of memory. How to handle the problem. Thanks.
I can't download torch-directml - says there's no matching distribution. I installed the current Anaconda and it's running Python 3.11.5 - is that the problem? Do I need an older version of Anazonda? UPDATE: Looks like manually installing the older version of Python in Anaconda fixed this particular error.
Your comment section says you're talented and knowledgeable. I am the opposite but I want in! How about an AMD tour de force update? For older AMD cards? Thanks.
usually, to start a command line, or any other program, on the Windows, search 'terminal'. The Git and Ananconda/Miniconda (for python virtual environment) are needed. They were included in my previous tutorial, see ruclips.net/video/_6yP3Gv04-w/видео.html. It includes how to start the command line
yes, on CMD, change directory to custom_node, then run git clone ... and cd the manager folder, run pip install -r requirements.txt@@hitmanehsan
Год назад+2
SOLUTION!!! I found out why this video does not work! You have to install python 3.8 and not a later version. With 3.8 it works perfect. If you have 3.10 like in his example you just get a lot of errors on every step of this installation. I guess he made the installation by himself with 3.8 or earlier and then updated before he made this toturial. So to save you a lot of time... be careful about what version of Python you install. To be sure install the 3.8 version and it will work a 100%
got it working but its a bit laggy and slow but works . also how do i install manager into this i tried putting in from github but it wouldnt run said torch was missing and some lines of code line 77 or something flagged . and i couldnt restart after i closed it down said it couldnt find comfyui so i had to go through a few of the steps agin just to restart it
ERROR: Could not find a version that satisfies the requirement torch-directml (from versions: none)
ERROR: No matching distribution found for torch-directml
were you using a virtual python environment?
I got the same problem and I was not in virtual python enviromment
A mi me paso tambien pero mi error era usar otra version de Python. Puse la 3.10.12 y ya funcionó.
Can I run it on ryzen 5 5600g igpu ?
I tried and tried for months to get ComfyUI to run stably on my AMD system (5900X/6900XT) with 32GB RAM and 16GB VRAM. It just wasn't worth it. DirectML is suuuuuper slow. I switched over to Linux and am running ComfyUI under ROCm, and it's about 10x faster than on Windows with DirectML. And I can generate much larger images without running out of VRAM.
my 6700XT on Windows (directml) is about twice slower than Linux ruclips.net/video/iCsUNPE7k2k/видео.html Your '10x' seems surprising to me.
Have you tried WSL2? It’s like some kind of virtual machine intergrated to Windows that allows you to run linux under Windows.
How did you get it working with ROCm? Is it a Linux-only thing?
There's several things missing. I attempted to run this but missing several modules you have on your Comfyui that are not installed by default. Also, no mention of lcm anywhere in this tutorial except in the beginning.
Sorry I didn't have time to add all the details. I think LCM has been added to ComfyUI main code. No additional installation is needed. For LCM videos, see ruclips.net/video/5NpffP9lJ9w/видео.html
Thank you ! But we need code in description sir, we like to paste
I also dont have "ckpt name" so im unable to generate a prompt
thanks for trying to work on a different one. I can get fooocus almost there, but keep running out of memory, and can't scale the images down enough to not run out of memory, which is a bummer for 6700xt. can't wait to try this one.
It works! Thanks a bunch! Only question I have more is, would it be possible to make a shortcut to launch it? A .bat file or something, instead of activating the environment and typing out the line to boot it up :/
can try adding the commands to the .bat file so it can activate environment and run the python file ?
Running on 6700 12G, err show: comfyui\comfy\model_sampling.py:81: UserWarning: The operator 'aten::frac.out' is not currently supported on the DML backend and will fall back to run on the CPU. So finally DML allocator out of memory. How to handle the problem. Thanks.
That's strange. My 6700 works without such issues. What python/pytorch version was it?
@@tech-practice9805 Followed all the steps, python with 3.10.12, pytorch should match with python. Thanks help!
I can't download torch-directml - says there's no matching distribution. I installed the current Anaconda and it's running Python 3.11.5 - is that the problem? Do I need an older version of Anazonda? UPDATE: Looks like manually installing the older version of Python in Anaconda fixed this particular error.
glad it works!
Your comment section says you're talented and knowledgeable. I am the opposite but I want in! How about an AMD tour de force update? For older AMD cards? Thanks.
how do you use a commandline?
usually, to start a command line, or any other program, on the Windows, search 'terminal'. The Git and Ananconda/Miniconda (for python virtual environment) are needed. They were included in my previous tutorial, see ruclips.net/video/_6yP3Gv04-w/видео.html. It includes how to start the command line
Thank you bro... it's working... it's working...!!! (Funciona... funciona...)
how can install manager ?
Comfy manager can be installed: to the custom_node directory, run 'git clone' the manager repo, then pip install its requirements.txt.
@@tech-practice9805 can u plz explain it more? i found custome_node folder.but what should i do there? use CMD on this folder ...?!
yes, on CMD, change directory to custom_node, then run git clone ... and cd the manager folder, run pip install -r requirements.txt@@hitmanehsan
SOLUTION!!! I found out why this video does not work! You have to install python 3.8 and not a later version. With 3.8 it works perfect. If you have 3.10 like in his example you just get a lot of errors on every step of this installation. I guess he made the installation by himself with 3.8 or earlier and then updated before he made this toturial. So to save you a lot of time... be careful about what version of Python you install. To be sure install the 3.8 version and it will work a 100%
Does anyone knows something about this?
it only use my APU not AMD GPU :(
try to disable the APU?
jesus you need to work on making tutorials man this video is no different than the comfyui github page
ungratfull
We all had no problem with this tutorial... Skill issue maybe?