Easiest way to Install Stable Diffusion XL 1.0 on Your Computer! (Best Tutorial)
Вставка
- Опубліковано 25 сер 2024
- In this video, I will show you how to install *Stable Diffusion XL 1.0 (SDXL 1.0)* on your computer in just a few minutes. SDXL 1.0 is a *latent text-to-image diffusion model* that can generate realistic and high-quality images based on text prompts. It is the **world's best open image generation model**, according to user preference tests. Whether you are interested in art, science, or entertainment, you will find this software amazing and fun to use.
▼ Link(s) From Today’s Video:
✩ Stable Diffusion XL: civitai.com/mo...
✩ Comfy UI: github.com/com...
✩ 7 Zip: www.7-zip.org/
► MattVidPro Discord: / discord
► Follow Me on Twitter: / mattvidpro
-------------------------------------------------
▼ Extra Links of Interest:
✩ AI LINKS MASTER LIST: www.futurepedi...
✩ General AI Playlist: • General MattVidPro AI ...
✩ AI I use to edit videos: www.descript.c...
✩ Second Channel: / @matt_pie
-------------------------------------------------
Thanks for watching Matt Video Productions! I make all sorts of videos here on UA-cam! Technology, Tutorials, and Reviews! Enjoy Your stay here, and subscribe!
All Suggestions, Thoughts And Comments Are Greatly Appreciated… Because I Actually Read Them.
-------------------------------------------------
► Business Contact: MattVidProSecond@gmail.com
The goal of this tutorial is to be the easiest method of installing Stable Diffusion. Let me know if you have questions!
Do you recommend I switch from automatic1111 to comfyui and put in the effort of learning how to use it? I have an RTX 3080 and SDXL seems to run noticeably slower for me than SD1.5 did.. the results are great.. but the performance in automatic1111 isn't
@@subn0rma1 It will be noticeably slower as it's working at higher resolutions. Try it without the refiner and a custom model as you really don't need the refiner and Stabilty AI kind of say it themselves. I can get a 1024x1024 in about 8-10 secs on my RTX 3060 with 12Gb VRAM. ComfyUI is definitely quicker though and easier on resources. Right now I'll be sticking with A111 as it has several QOL features I like that ComfyUI doesn't but of course, it's up to you. 🙂
@@subn0rma1the comfyui makes my pc burn unlike the automatic1111
Now do ROCm in Linux...just kidding, took me three days to get it set up properly. BUT hardware democratization AND open source are key. Totally enjoy all your content!
@@williammixson2541 Having an AMD card I have no choice but to go with ROCm, it's with it anyway
Quick Explanation for most of the advanced prompts:
Seed: basically like in Minecraft, if you use the same, you will get the same image. So if you got a great image and want to improve it, you can tweak the settings of that image by regenerating with different settings but the same seed.
Control_After_Generate: If the seed should stay the same or not after regenerating
Steps: How many iterations to generate. The more steps the longer it will take, but the higher quality you will get. (But beware that too many steps may make the image look a bit softened out, less photorealistic.) for SDXL I would recommend a value between 25-30
CFG: How close the prompt should be followed. Lower CFG means higher creativity but you may not get exactly what you wanted.
Sampler: Euler is good but I would recommend DPM++ 2M Karras its a bit slower but gives better quality
Thank you, this was super useful. We need a tutorial on how to get the most out of prompts.
can it handle old gpu like RX570?
@@PunxTV123 if you got the 8gb vram edition it should be no problem, just make sure that you got at least 24gigs of ram (not vram) in your computer
There is no "DPM++ 2M Karras" but some with "SDE" and "SDE GPU" and an "ancestral".
Edit: Oh ok, apparently "pp" means "++".
Great comment, got my images up to par quickly...
You completely ignored the refiner which is a big part of sdxl.
Expand a bit please? Is it just another thing to download?
@@IdK-er6rh yes it's a 2nd model. But you have to adjust the UI to be able to use it. Stable diffusion XL is supposed to generate about half of the steps using the base model and the remaining steps using the refiner. Which will improve the end result. There are premade comfyui setting for using both the base and refiner on civitai. A simple Google search should also help pointing you in the right direction.
Or get a better tutorial. I mean he tries to explain a lot however since it's not installing sdxl correctly I wouldn't consider this a good or even best tutorial. Sorry
@@nighty2445can you please recommend a better tutorial?
@@nighty2445please recommend a tutorial as u said, I am new to it , every thing look same😢
Pls...❤
can pls help tell how to use image as reference while generating
Thank you so much for making this video! I was trying to upgrade from SD to SDXL and I couldn't find a good setup video for XL anywhere until I hit this one. Beautiful work, I feel like anyone with the equipment wanting to get into Ai image generation could follow this easily. I love the fact you included the simplest things that most would take as common knowledge even down to showing 7zip. Any tutorial about software should be this positive in energy and thorough. Wonderful job! 😃
Doing the work of the people! Thank you! Please do create the more advanced tutorial too!!
Your tutorial was so good!!! It worked! Thank you for being so patient to go through all of these details... It looked really intimidating, but you broke it down so well. Thank you!
This is literally the best tutorial ever
mac version of this would be sweet...
Thanks, Matt, for explaining how to do this so clearly! I still think you are related to Carl Sagan! Thanks!!
I own a GTX 1060 3 GB, with ComfyUI I can use the SDXL model and generate 1024x1024 image in 3 minutes and 512x512 in 30 seconds, with xformers and --lowvram activated, it's amazing how optimized it is now!.
Thanks! This worked without a hitch. Am having fun now playing with it.
WORKS on 4GB Nvidia card after running updates. Thanks for the tutorial!
I thought I might be out of luck with a 4GB card. What specifically did you need to make sure was updated?
There are some update .bat files that you can run and it will pull latest updates@@Sensorium19
Great started video on SD XL! Good job Matt!
Two questions: is the refiner built in? 2: where do I use inpainting and img2img?
I believe it is built in. that is what I gathered from civit ai
@@MattVidPro The refiner is NOT built in! You need to download the refiner model safetensors file in addition to the base model file and load the workflow that uses both models from the examples page.
Thank you! Very straightforward and painless.
Ridiculously detailed (in a good way). Thanks!
🔥🔥🔥🔥video. Simple to the point and got a noob like me up and running. Thanks bro.
youre amazing brother, thank you so much for taking the time to make such an easy tutorial.
thanks for this knowledge!!
I'm going to be "that guy':
Will this model already allow uncensored generations, or will we have to find a different model?
And if so, where should we search?
boy what the hell
Thank you so much! Finally the installation issues have been resolved! Voila! :)
3:18 Skip to the install (aka) The next Step
4:18 Comfy UI Install
5:10 Downloading 7 zip and extract files
No more updates, I have a Radeon GPU Sad times.
:D
Mahalos! I was installing this before and there was a mess of Python dependencies and PATH requirements/proscriptions and--no more! Huge thanks
This is amazing... and only a 6.5GB SDXL file. Would love to see more on running these models locally.
It was super easy to install even for non-computer person! Thanks a lot!
Any way to use this with AMD GPU's?
Not a fan of Nvidia to be honest.
SD works better with Nvidia, but Linux works better with AMD. It's rare to find a one-size-fits-all solution. Usually everything has pros and cons.
Thanks for making this tutorial! It was easy to follow and fun to play around with once set up :)
Thanks Matt...Happy New Year !
Man, this is a million times easier than the A1111 in SD1.5...thanks bro!
Got it all working and generated some interesting results. Thanks for the excellent tutorial. 👍
Hey folks! If You have graphic cards with only 8GB VRAM (ie. RTX2070 Super) and problems with running SD XL on Automatic1111 (no image creation, VEEEEERY slowly iterating), then try this one (ComfyUI). I've installed it right now, and the whole example image is generated in few seconds (default settings, only changed the resolution as showed in this video to 1024x1024). *Thank You VERY MUCH, Matt :)*
does the --medvram flag not work anymore? (I'm on an AMD gpu so I can't use Xformers) for context I'm running an overclocked rx 6600 with 8gb vram.
Good tutorial, but one small note. Maybe I overlooked it or missed it, but I don't think you broached the issue of whether this works on AMD GPUs until after you went through the install process. I'd gotten all the way to starting comfy UI before realizing this was a NVIDIA only solution. Still a great tutorial and definitely the simplest and most thorough I've seen on here, if I had Nvidia I'm sure I'd be a happy camper right now!
For AMD / DirectML
google
/vladmandic/automatic
Hopefully soon they implement ROCm on Windows, since that out now
You missed it. He did address it in the beginning.
@@MokshaDharma He didn't in this one, he just addresses whether or not user has a GPU, not the brand. @ the time stamp if you see it
He mentions it towards the end @7:30 after people who are following along have already downloaded 8GB worth of files ;p
It's possible to run it without one. I'm using an AMD, and it runs adequately, although it's a bit sluggish, taking about 6 to 10 minutes for generation
On intel integarted GPU can i make it work
Thank you, it was very easy to get started with your explanation.
ive done everything like you have said. but when i go to click run nvidia gpu it goes into a windows prompt black screen. doesnt open like yours does
The same happens to me.
Is there any benefit to installing SDXL from scratch if you already have 1.5 installed or would just downloading the safetensor to your model folder accomplish the same thing?
Have you tried Happy Diffusion
thank you for sharing mate. i'm pretty new to ai software and this video helped me alot
How do you do img to img?
I'd like to run my old art through it.
Thanks Matt.
Since I have an AMD graphics card I can use a ROCm based interface with SDXL in Linux to be able to generate images with it at a fast speed.
I tried using my Intel cpu +32gb ram in ComfyUI on windows, and a 1024x1014 image takes about 12 minutes. 512x512 takes about 2 minutes.
What AMD card?
RTX 4070 has been generating images in like 10 seconds. But that’s also on SD 1.5
Do you know if comfy will work with a tesla gpu? as these are cheaper for high vram.
I got this video on my watch later, I am very interested.
Tank you!
But a question: Copying the models directly into the ComfyUI folder means, if I want to use the same model for A1111, I need to have a copy of that model in the A1111 folder as well?
I'm spit-balling but maybe you can create a file link in both folders that point to the model so you only need one copy? I managed to create a few links in W7, don't know if it's possible in your OS.
Where is the "fully uncensored" model found to download?
I need my glasses to see most of this video’s text lol
youtube's mobile zoom feature sitting here, depressed:
😢😢😢😢😢😢😢
thank you thank you thank you marvellous Matt
Hello, thank you for the video! I have a quick question. I guess right now what you are doing is using our local computer as a server to do the generation, so this is not a API request, is this correct? Also, I am wondering if we can create an API to integrate into a website for example? Thank you so much!
Does anyone know some prompts for getting SDXL's background in focus? (E.g., crisp edges on mountains and leaves in the background, as well as a subject in the foreground in focus.)
This was definitely the easiest to install. Still not as good as MidJourney with realistic people. Made them all really disfigured.
Still good for more abstract/fantasy images but a few things I noticed:
1. If using Chrome, the zoom is very finicky with the scroll wheel. Hold down Ctrl when scrolling to zoom easier.
2. Probably obvious to some, but be sure to keep the Command window open, accidentally closed mine and even though the web interface stays active, it will give you errors if you try to create anything.
3. This will take-over your computer unless you have a very high-end GPU. Don't plan on doing too many other things while this is running. This UA-cam video even slowed down when I tried running it. It ended up killing my fans and required a full shut down and restart to fix it. I use this computer for a lot of 3D and video editing, so it is no slouch, but still made it run very slow.
It won't let me move the stable diffusion file into the checkpoints folder
Love the shirt. It’s also my favorite band.
wow ty so much, this is very very simple ! ;-)
Thanks a lot for the video. Guess Mac users will have to wait a little further?
I wish that was changed...
Great! I am up and running. I really prefer the Auto1111 interface, but this comfy seems to be faster.
I'm running an RTX 3070, and the image I get is like white noise, only in colours, using the same prompt. What am I doing wrong?
Heya Matt. Was curious if you ever used RunPod for running SD. That's what I do since my laptop isn't really beefy. It does cost an hourly rate, but believe it is well worth it if you don't have the specs.
Thank you so much for a good explanation how to install this. :D
A question! I guess there is some updates now and then in Stable Diffusion. Do we have to do the same thing all over or do a normal "update"?
Does the linked download integrate the refiner? If not, do we download and install the refiner in the same spot?
Yes, the refiner model goes into the same folder as the base model. But for it to do anything, you also need to load the workflow that uses both models from the ComfyUI examples page.
@@jay_sensz Thanks, I will try to figure that one out. I wondered why I was getting such weird stuff when I would run with the refiner by itself. I have the two installed, but now will figure out the examples.
What if you have 7.9?
the operation keeps breaking when i try to move it from the zip file :/
Thanks man!
I just get a window that immediately pops up filled with errors. :(
I have no idea what it all means.
But 7 zip doesn’t appear and 7 zip file manager it appears as 7 zip console. And it fails to open either of the downloads. Someone help me!
👏👏❤️🔥👌👊 CPU 1 hour 38 minutes 😮😮
Can you make videos on AI 3d modeling tools? That would be incredible.
civitai's page is down. No models or even anything on the homepage. Do you have another source?
Nice T-shirt and video!)
I would love top see a more advanced video!!
Why the GPU, 7-Zip, Browser Downloads function explaining part, people who don't know those super basics will not comprehend how to use SD especially ComfyUI.
WHAT THAT'S AMAZING
I did all the jazz, but I got a queue size error "Failed to fetch" 🤔
After the last step, my Stable Diffusion won't open! From CD to dissolving screen. I have the equipment to open it, anyone else has the same problem?
tutorial on how to use 7zip and check if I have a Nvidia gpu thanks
Hummmm. I wonder if this was flesh Matt or AI Matt I’m seeing and hearing
Constantly get errors when Unzipping ComfyUI ZIP folder?
Thanks it work perfect
Thank for great detailed instructions. Worked perfectly. But, since I have Stable Diffusion installed locally, I don't need the ComfyUI interface and all that goes into installing it. Still, Thanks.
I'd say nowadays Pinokio way is even easier :)
it is true, that is wont work with low memory GPU. So need wait for easier version...
"[enforce fail at ..\c10\core\impl\alloc_cpu.cpp:72] data. DefaultCPUAllocator: not enough memory: you tried to allocate 26214400 bytes."
Installed it.
Granted, I only have 4gigs of gpu and it's not Nvidia. But there is still no excuse for how insanely slow this program is.
I set it to render ONE image of the default galaxy jar prompt, and at the default 515x512 size. It took just over an hour to render! Not onlynly that, but it also shot out 6 tiny versions of the prompt within that SINGLE 512x512 image.
There are clearly bugs in the program.
you look like a grown up version of Will from Stranger Things 😆
Aargh! Nothing about img2img or adding nodes!! There is a weird absence of any sort of (apparent) menu system in ComfyUI.
Can you include a picture of yourself for the AI to then create brand new photos of you?
I had a big problem installing this. My cat kept trying to catch my mouse!!! I've never had so many issues installing a model before!
Comfy UI from the link provided will not open on my machine. Any ideas?
huh how come your windows looks different?
idk when i tried to generat image an error popup says that
[enforce fail at ..\c10\core\impl\alloc_cpu.cpp:72] data. DefaultCPUAllocator: not enough memory: you tried to allocate 26214400 bytes.
why ?
i have Nividia MX130 2gb vram and 8gb ram i7 8gen
how do I fix Error occurred when executing CheckpointLoaderSimple: on Comfy UI
Is there a way we can use it on a phone
it not work. he not open any browser windows
just reamin in the prompt command window
Thanks
can pls help tell how to use image as reference while generating
Do you know of any open source models that can generate images with text?
Thank you. But how do i use multiple lora's or models at once ?
Hi Matt, awesome video. all went off without a hitch except when i got to running the Nvidia GPU batch file. It appears as only a text file and i cannot open it. When i dbl click it just opens with notepad.
Thanks 🙏
how to do image-to-image please
Bro how to create images in an automated way ? Via CLI or other way so can auto-generate images on demand through a programming language
cool, but what tf you talking about stable diffusion not working with AMD GPU?
My GPU is 512, can i install amd use this as well?