How to install New StableSwarm Beta.
HTML-код
- Опубликовано: 20 мар 2024
- github.com/Stability-AI/Stabl...
git clone github.com/ltdrdata/ComfyUI-M...
Prompt styles for Stable diffusion Automatic1111, Forge, ComfyUI & Vlad/SD.Next: / sebs-hilis-79649068
ComfyUI workflow for 1.5 models: / comfyui-1-5-86145057
ComfyUI Workflow for SDXL: / comfyui-workflow-86104919
Get early access to videos and help me, support me on Patreon / sebastiankamph
Chat with me in our community discord: / discord
Stable Diffusion for Beginners Playlist • Stable Diffusion Begin...
My Weekly AI Art Challenges • Let's AI Paint - Weekl...
My Stable diffusion workflow to Perfect Images • Revealing my Workflow ...
ControlNet tutorial and install guide • NEW ControlNet for Sta...
Famous Scenes Remade by ControlNet AI • Famous Scenes Remade b... 
Хобби
Super impressed with the UI/UX details being payed attention to by the dev team!
Awesome, multi-GPU support is a big deal. Looking forward to trying it out, thank you for sharing!
I wish you'd gone into the 'choose custom backend'. I'm pretty sure that'd allow me to skip all of the setup stuff I had to do to get my existing comfyui running.
I put my paths to my already existing models but im not gettign the ui to load them. Did you find a way around that?
can i map the inputs on the generate tab to comfy nodes? Otherwise this is just comfy with extra steps
big workflows get hard to organize so if i can hide my rats nest, that would be worth it
OMG, you can type CMD in the path of windows explorer and it will take you to that directory in command prompt.. omg forget this AI stuff my mind is so blown I can't i believe i never knew about that lol. Thank you! Great tip! Lol
Same reaction here! 😮
@@TimFromPerth and you can drag and drop a file/script in to cmd, and the path gets pasted instantly, another hot tip
I've been using computers heavily for 2 decades and only discovered it recently. It's probably a new thing.
If you really want your mind blown, try pressing Windows-V instead of CTRL-V..
@@blademarketing and Windows-C give you access to copilot 😯
A question few will ask...
Does it support Rocm or ZLUDA?
does it run cascade and playground?
I know Comfy is on the backend but maybe you could delve more into the benefits of using this over A1111 or get into more of the key features that make it worth the switch?
Yes, however as youtubers, most views will be for installation etc so people focus on that.
The "non-comfy" UI is a similar look to the A1111 interface, but it looks like they expose more options than A1111 has by default... I haven't installed Swarm myself, but that and the ability to switch back-and-forth into the Comfy view are the selling points.
The purpose of the stable swarm is to allow you to use multiple GPUs at the same time. If you have more than one Nvidia GPU go for it
@@ShahabEslamianI have multiple NVidia GPUs but they are all on various PCs on the Network. Does it work ? Also, I have one 3090, but the rest is a 2080Ti, a 1080 and a 1070. Is it really worth it ? Not sure it will make a big difference opposed to using the 3090 only.
@@SPT1 I don't think you can use it over the network because of the latency between the GPUs but check out their documentation. 3090 is already powerful enough for most applications so it might not matter to you anyway
Does it install comfy during this installation or should it already be installed?
that's the point if you have your comfy and auto111 already there?
Does it run under venv? So many problems with various other installs and thir individual requirements for different versions of Python.
Having a 3070 and an older CPU, XL renders took 40 minutes to over an hour on plain ComfyUI and A1111. Even 1.5 models took 20 to 30 minutes. I don't know what this UI does differently but even SDXL renders are completing in less than 25 seconds (once the model is loaded). This is wonderful. Thank you so much to the developers and to you for bringing it up.
Is that 8Gb VRAM? I ask because I have a 12Gb VRAM 3060 and SDXL was about 25secs an image for me on A1111. It's even quicker since I switched to Forge. Obviously, resolution and upscaling etc could make a big difference in causing those times in minutes but they still seem excessively large.
@@Elwaves2925 Yup! 8GB. I use the standard resolutions. 1024x1024 for sdxl and 512x512 for 1.5 models.
@@juancarlosgonzalez8950 Hmm, even for 8Gb that is really odd. Something is wrong somewhere, it seems like it's using your CPU, not the GPU but I have no idea why.
Have you tried SD WebUI Forge? It's basically A1111 but with much better optimisation, especially for cards like yours. I saw a very noticeable and positive difference in generation times. Might be worth a look and I would recommend the manual install route over the 1-click option.
There's no excuse for 1 hour for a generation. I have a decent laptop with 3060 gpu (6gb as it's the laptop version) and in comfyui or fooocus I usually have a 1000x1000 generated image in about 1 minute. Your app for some reason doesn't see your gpu, it would be the only explanation.
That doesn't sound right. Are you sure it isn't going with CPU?
Is this beneficial to someone who doesn't use comfy though? Can I build a workflow without messing with the spaghetti?
Its a UI for comfyui so no but with COmfyui you can just take any workflow drag + drop and use them so you can do that too if not wanted to mess with nodes
You could always find someone else's workflow.
So am I to understand you can drag & drop comfyui workflows into StableSwarm?
I have a separate models directory that I use for all my SD installs. I cannot seem to get an absolute path to work for modelroot
Does this manage different pytorch versions?
so does this have any downside compared to comfy? It seems like it's just building up on top of comfy and providing additional UI while keeping the entire comfy functionality?
please do some in depth dives and comparisons of soft inpainting and demofusion
You mean like the new soft inpainting in a1111 or swarm?
I am not able to launch local host on web browser nothing is showing 'This site can’t be reached'
Do I need to install Installing the NVIDIA Container Toolkit?
If yes please tell how to install in windows system?
does it work for the outpainting plz ?
Can you set the server paths to find all your a1111 folders rather than doubling up?
no its yet again same thing. or just use stable fusion forge. who use a1111. we use yet again something else LOL
@@JarppaGuru yeah I have been using forge. It's great. I'm not sold on swarm yet, but I am still playing around to give it a chance.
whats the swarm idea behind it? just a name? i got a machine with a 3090 and one with dual 1080... how to swarm them?
Multi gpu support. I don't have multi though so couldn't test that.
thank you for the info, but will it work on my AMD rx580 ??
so we need to install comfyui again !
No. There's an option to use existing ComfyUI
Only one comfyUI, you can map the comfyui installation. You can even use comfyUI that is running on another server. very flexible UI.
Anyone knows how does the API work? Can I run local models with their cloud GPU power or do I have to choose from their available models on the cloud?
Nice video, Does Swarm fully support multi gpu processing?
It should!
why not?
Hi, I'm new to Stable Diffusion and quite lost. If someone could help me, I'm getting (Some backends have erred on the server. Check the server logs for the details.)
I'm not sure but do I have to install anything else than Stable Swarm and copy the Model to Stable-Diffusion folder. Or is it necessary to have A1111 already installed?
Thank you for zour potential help :)
Is this working with AMD graphic cards? Buying a 7900 XTX was the big mistake.
Unfortunately I feel the same way. No idea why I bought that card instead of an Nvidia
ooof... i was gonna get that card, ended up going for 4080 super instead
Yh AMD GPUs only good with SD on linux and ROCm @@joefawcett2191
@@joefawcett2191 4090. what is amd gpu. cpu allready has gpu or its on mother board why we want buy new amd lol. why we need pay amd gpu when buy cpu lol
same vram usage as stable diff? because my pc cant generate properly because of my poor memory ram
Same as A111 ? Same as Forge ? Same as ComfyUI ? etc. These would have been meaningful questions. But "same as stable diffusion" doesn't mean anything. All those WebUIs are using the SAME Stable Diffusion models.
Virustotal doesn't help with an executable that downloads gigs of data..
Yes, you're right. Thanks for the feedback!
Followed you steps to the letter and got a what looks like just the frame for the UI but no buttons at all.
Has anyone tested this with multi-GPU? Do they need the full PCI-E slot, or would an old GPU miner setup work with the PCIE 1x adapters? I am happy to test it out if no one knows.
thank you for the info my dear
Is Stable Swarm faster than Forge? Because I've switched to Forge from A1111, using SDXL lightning 1024x1024 with my 2060 6GB takes less than 30 seconds. I'm so surprised how the Forge UI can handle low-medium specs hardware
can it handle the new IpAdapter plus? Since the creator recently completely changed it
If comfy can, then yes.
hot diggity beeez!
comfy UI does not like my four p40ties... The hotfix from github does not seem to work.
runtimeerror: cuda error: operation not supported.
Already tried to change the run nvidia bat file and included --disable-cuda-malloc...
Is it better than forge though?
Great Guide as always
I had to install git/dotnet first for the script to work. For whatever reason it gave errors when it tried to install git.
the only thing I'm curious about is something about the backend servers... are they sending data back to the Swarm people?
I think thats just Stability AI, no?
I wish there was an Option to just expose selected parameters, Inputifields etc. from the node graph to the simplified gui instead of everything. Does anyone know if this is planned?
You can ask the developer in my Discord, he runs a swarm channel there
Nicee
Is this one better than Auto11111?
its same diff method and layout. forge is faster
Does it support extensions? I NEED my ADetailer
Theres a node for that in comfy.
research running this locally
controlnet does not work.
noice
toitt
dose it have anything better than forge ? its faster ?
basically looks same using same code LOL so if you are forge master you do it thing faster on that. but this use a1111 so its slower than forge hihihi. even they same code lol
I can't see where he's clicking to get to the final windows he's showing.
Mac?
👍👍👍👍
🌟😊
Reminds me of stability matrix except it has no workflow editor.
It's actually available as a package on stability matrix.
@@stevehaney344 We can use workflows inside stability matrix?
3:28 can i take that as legal advice XD
"swarm" can i run this on several local computers and get all my gpu's working together ?
Yes!
Just found something out here. maybe i was looking at this in the wrong way, thinking "swarm" works together to generate ONE image, but it didnt work like that. the "swarm" only works when generating multiple images. and all my gpu's dont work together ec one works on their own image.. i was hoping this was able to give me a ton of compute power when generating ;) am i wrong or is there a way to maybe get all the gpu's working towards a single image ? that would be amazing.. hehe
@@sebastiankamphREALLY?! GET OUT! AWESOME ::::DDDD
@@aZiDtrip yah, it's mostly for video work or other kinds of workflows where you can do multiples, currently at least 😅
Bro the waifu universe is about to explode
It's been in beta for a couple of weeks
What do you call Bees that formed a rock band with Niro Knox? Wanna-Bees
Hah, I see what you did there
2:15 if.. then we support them? LOL. im hear thats click for you
How do you mean?
Finally Really dislike ComfyUI giving this a bash.
2:47 means nothing. it just not find those virus and malware that on database. its new can have new malware. its python lol. lotsa code nobody checcked just use someone made
aaa
no A1111? well, pass...
wait, they decided to call it StableSwarm? what about StableSwarm then?
2nd
very bad you did't tell the main thing how much raim or pc requirements?
Nvidia gpu. Rest is irrelevant
@@FlipFinderzGreat comparison lol
he did't mention pc requirement what problum with you?@@FlipFinderz
Anything from rtx 3060 6gb and above
The 3060 6gp is the minimum
Swarm your data.
JIF!!!!!
I fail to see how it's even a debate anymore.
The guy who came up with the actual bloody format and file extension said it is DJIF.
So...
😆
You should start the video by explaining what the hell StableSwarm is. Okay, so it's UI. UI for what? For Linux? Blender? Some sort of AI thing? Something to do with Stabele Diffusion? You have no created interest with me, but annoyance, and that is a really stupid way to a run a RUclips channel.
its trash
Kindly answer this "does it run cascade and playground?"