Install Stable Diffusion 3 Locally: Step-by-Step with StableSwarmUI & ComfyUI
Vložit
- čas přidán 26. 06. 2024
- Welcome to another tutorial! In this video, I'll guide you through the process of installing Stable Diffusion 3 on your local machine using two different interfaces: StableSwarmUI and ComfyUI. Whether you're a beginner or an advanced user, this step-by-step guide will help you set up everything you need to start generating stunning AI images right away.
First, I'll show you how to install StableSwarmUI on Windows. We'll go through downloading the necessary files, setting up the folder structure, and running the installer. I'll also explain how to configure your settings and customize the interface to suit your preferences.
Next, we'll move on to ComfyUI. I'll provide detailed instructions on downloading, extracting, and setting up ComfyUI. You'll learn how to install custom nodes, manage workflows, and load Stable Diffusion models for the best results.
Join my Discord server for news, resources, and discussions about AI tools. Share your AI-generated images, and let's learn from each other!
Join my new discord channel:
/ discord
Join my facebook community / pixaromacommunity
If you find this tutorial useful, please leave a like and subscribe for more in-depth guides and tips!
Resources:
StableSwarmUI
github.com/Stability-AI/Stabl...
ComfyUI
github.com/comfyanonymous/Com...
ComfyUI Manager
github.com/ltdrdata/ComfyUI-M...
Get the SD3 models
huggingface.co/stabilityai/st...
License Info
stability.ai/license
Juggernaut X RunDiffusion - Version 10 - SDXL model
civitai.com/models/133005?mod...
Juggernaut - Reborn, Version 1.5 SD Model
civitai.com/models/46422/jugg...
Simple workflow templates
openart.ai/workflows/templates
Chapters
00:00 Stable Diffusion 3 and Swarm UI Install
04:47 Getting the SD3 Models
13:04 Stable Diffusion 3 and Comfy UI Install
Maybe you can try this AI tool bit.ly/4b7UCT1 to get stunning 8K/10K Stable Diffusion-generated images.
Unlock special offer bit.ly/4b9L1v3 for users who need this AI tool.
Wow, this tutorial is mind-blowing! Can't believe all the detailed steps and functionalities. Simply amazing! Great work!
Best Tutorials on CZcams!
Thank you for the very complete and comprehensive video and easy and clear teaching
Thanks a lot! 🙂
Thank's a lot
Bless you my Lord pixaroma Infinity, I will start tomorrow my Lord stable Diffusion.
Thanks!
Thank you so much for support, appreciate it
@@pixaroma "I" :p
@@halkive4619 It's hard with English, that's why I don't do tutorials with my voice. It's difficult to keep up with grammar when English is not your first language. :D
@@pixaroma oh man im sorry, I didnt mean it in a rude way. Im so weird....i kinda hesitated on commenting that :/ Mybad brother
@@halkive4619 don't worry about it :)
Stable Diffusion 3 models with names including "incl_clips" signify that these models have integrated text encoders. This integration means you do not need to add an extra node for those CLIP models, as the text encoding is already handled within the model itself. This differs from models like "sd3_medium," where an additional node for the text encoders are necessary. So if you use sd3_medium.safetensor then add the clip encoders, for others are not need it from my understanding.
CivitAI just banned the SD3 because of the weirds license of stabilityAI, so play around with sd3 but because of the license many people will not be able to improve it or use it for comercial work.
👋
How to use with defooocus AND automatic 1111 and please make a tutorial on it????
I will make it when they update the ui to support it, right now it doesn't support sd3
@@pixaroma plz i believe it work for defooocus if not for fooocus but maybe it will work for defooocus as a saftencor SD3 model so how plz teach and i use always not local defooocus but google colab defooocus hmmm??
@AaliDGr8 I don't use colab since my PC can handle it, the free colab can not handle many interfaces so i prefer to not pay for premium. Anyway the license or sd3 is quite strange, so you are better for now with sdxl.models
You could click on send to and select desktop which will create a shortcut on the desktop.
Good to know, thanks
I am a poor man but I will subscribe good fellow.
Hi, do you think it's worth switching for now since it's just the beginning of SD3 or better wait for more commonity models developpeD?
Also I wasn't following what's happening with forge, will it stop to receive updates? I really like the ui similar to A1111 but better. Is is still worth sticking with one of the 2 for future SD3 version or it's time to learn Comfy? Ngl Swarm looks quite weird. Which one would you reccomend between all 4 from now on?
I will add Thursday a video about that, i will probably continue with a1111 and comfyui
Best tutorial for SD3, is it possible do add clip skip to the ui like before?
If you are talking about comfyui you can see if the node they talk about work for you, i am not at computer this weekend github.com/comfyanonymous/ComfyUI/issues/2223
@@pixaroma yees, exactily thank u
Will you be switching to another UI now that forge is changing to an experimental repo?
Yeah, since without updates you can not do much for long, i mean it still works but probably other ui will do more improvements in speed and quality so we have to keep up. Alternative are comfyui, swarmui, and automatic1111 , probably i will have all 3 and see how it evolves. I am playing more with comfy ui I saw it can do more advanced stuff. Automatic 1111 doesn't support yet SD3 but because of the license of sd3 probably many will stick with sdxl for a while
Damn. Did not hear about this. ComfyUI looks powerful but seems like it requires more work to do what can be done faster in the more basic UIs. Maybe I'm wrong I'll have to check it out. Never heard about this SwarmUI until I watched Pixaroma's video.
There are also versions for nvidia\amd gpus on huggingface, like Stable Diffusion 3 Medium TensorRT. Is it better to use them or this one ?
I didn't got to test them, heard that tensor might interfere with other extensions or control net
The model from old SD works in SD3? and Thnk for your video!
SD3 is just another model like sd.15 or sdxl, you just need the interface updated so it supports that model , comfy ui and swamp update the ui, but a1111 and forge didn't update it yet.
Hi, I am using AMD Radeon GPU & StableswarmUI can't access my GPU , what to do? please help.. Error code"Could not allocate tensor with 377487360 bytes. There is not enough GPU video memory available!" while in task manager GPU utilization shows 2% , it is not using my GPU at all . The same thing happened in automatic1111 but using "--use-directml --opt-sub-quad-attention --no-half --disable-nan-check --autolaunch" in user.bat file resolved by problem , plz help in StableswarmUI
I know that on nvidia is always easier than on AMD gpu, I am a designer not a programmer so those code are too much for me, but when it say tensor i think they try to find tensor like on nvidia, that amd doesnt have. But what I am suggesting is to ask in their discord server, it has a help channel there and you might find someone to help try this discord.gg/q2y38cqjNw
this is probably a dumb question but, does the portable comfyUI version can be updated? im asking cuz i used to use auto1111 but I stopped using the program like a year ago, in those days I remember I had to install the program manually and i needed to use a cmd command in order to receive updates
Yeah it can, you have an update bat, but with comfyui manager you can install with the update button from the manager, I used the portable version in the video
@@pixaroma i was watching while typing and just saw the option in the manager 😅, thanks! very well explained tutorial ^^, btw in your opinion, is comfyui better than auto1111? (or the others), like, in terms of performance, speed, etc?
It seems that is faster then a1111 once they update a1111 to get better speed like forgeui had probably i will use that also because interface is more friendly and organized, on comfy UI you have more control but can be a mess sometimes with all the nodes. So it they fix it probably i will use a1111 again for normal use and comfy for more custom workflows
@@pixaroma thanks again ^^ i'll do the same then
شكرًا لك . الشرح رائع وممتع. يسعدني الاشتراك بالقناة
So I have a pretty beefy PC sitting at home - RTX 4090 paired with 1/1gbps symetrical network connection. Is there a webserver version of that UI that lets me using this PC as raw horsepower for my webapps instead of paying third parties?
Honestly I dont know, I only run it locally, but I asked chatgpt same question and gave me a lot of solution but I didnt really understand all that. something like:
"Choose a Framework:
If you're looking for a specific UI like ComfyUI, check if they provide a server or API version that can be self-hosted.
If ComfyUI doesn't offer a server version, you can use frameworks like TensorFlow Serving, Flask, FastAPI, or Django to create your own web server for running AI models.
"
"Ensure your PC has all necessary software installed. This typically includes Python, CUDA, and the appropriate deep learning libraries (e.g., TensorFlow, PyTorch).
Install a web server framework (e.g., Flask or FastAPI). "
So really I can't help there, you need to do more research.
does it have seamless tiling on y only or x only here?
Sorry, I didn't have time to play around with it, after i come back from vacation i will play more with it
can i use for image to video model and where i will find that model?
I only tested one model in forge ui, this one, when i test more i will do more tutorials but so far i didn't find some good one, czcams.com/video/zXxK6IrRvD0/video.htmlsi=1c84r1Gi27C7Qbwe
How well does it do for video clips?
Not sure if can do yet, didn't test it, is for images is not made for video, and the other extensions like control net doesn't work yet
How much power does it use? Would we get slammed by the electric utility?
I was able to use it on a 6gb vram rtx2060
in 21:30 you changed the "sd3_medium" to "sd3_medium_include_clips" they are the same, just with the second one you can remove the TripleClipLoader node and pass the clip yellow line from the Load checkpoint node instead, the 6 gb model include two clips and the 15 gb one include all the 3 clips.
Thanks, yeah I found out about that later after I published the video. I didn't have time to do more tests, is the one with 3 clips better then that with 2 and that with one? Theoretical it should but not sure in practice
@@pixaromaI didn't download the 15 GB model yet. I only downloaded the 6 GB one and just started testing it right now.
i watched video in silent but why downloading clips seperately when there is clips included safetensors?
I found that about that later, I first found that about later, if you look at my channel is the first video with comfyui, I am learning more about each day, and when I got the workflow from hugging face didnt work until I had all those model, so I shared what it worked for me, later I saw what all meant. Thanks
[Error] Self-Start ComfyUI-0 on port 7825 failed. Restarting per configuration AutoRestart=true...
I keep getting this error, Do you know how to fix it.
Sorry i dont, did you try to update it from the update bat? Or try to reinstall comfyui
Does anyone know about support for Automatic 1111 ?
It doesnt support yet, they need to update it to make it work with SD3, so we wait for support.
How well does it do hands and teeth?
Really bad 😂 if is closeup portrait is ok, but for hands is really bad
@@pixaroma Will it take ADetailer? Definitely helps hands and faces
Not sure yet, I didn't n't have time to test it, sorry
anyone else getting "Invalid operation: ComfyUI execution error: Cannot handle this data type: (1, 1, 3),
Maybe it has something with do with amd, Nvidia usually works better and in video i used Nvidia but for amd are different steps on most ui and some dont support all AMD cards, but since i don't have amd maybe someone else can answer your question
works with amd gpu?
with some amd cards comfyui works, but need some extra things, if you search on their page amd you can find some info, always more problems on amd than on nvidia github.com/comfyanonymous/ComfyUI
automatic1111 can or no?
Not yet, I am waiting for it to be updated, hope will be soon.
What for SwarmUI? No portable, plus SDK(((
You can use comfyui is portable, or wait for automatic1111 to be updated.
I haven’t seen full video but I believe u must have hifi system to run offline… and I wonder why it would be free?
You need like 8gb of vram to work, and a good Nvidia card help, sdxl and 1.5 model are free for commercial use but sd3 only for personal use, you need a license for commercial like 20$ per month
When I click Generate, : Error
Invalid operation: All available backends failed to
load the model.
Try to select the models from the list maybe
retretert
Retreated?
Hi I have the next issue:
Prompt outputs failed validation
TripleCLIPLoader:
- Value not in list: clip_name1: 'clip_g.safetensors' not in []
- Value not in list: clip_name2: 'clip_l.safetensors' not in []
- Value not in list: clip_name3: 't5xxl_fp8_e4m3fn.safetensors' not in []
Queue size: 0
Extra options
even I delete and put it directly to clip node, it doesn't work
You need those models in clip folder if you are using sd3 medium, and if you have it in that folder and still say is not in the list just select it from Drop down where is that name