SwarmUI Master Tutorial - Use Stable Diffusion 3.5 Large and FLUX model with Amazing Performance and more

Hi! I often use your method of inpainting with segment:yolo in swarmui. However, since the last update of swarm I get an error in swarm when including the yolo9c segementation: ComfyUI errored: { "error": { "type": "invalid_prompt", "message": "Cannot execute because node SwarmMaskBlur does not exist.", "details": "Node ID '#102'", "extra_info": {} }, "node_errors": {} } segment:yolo-face_yolov9c.pt-1,0.7,0.5 Edit: A fresh install of swarmui seems to have solved the issue...

The Flux links on the "28 November 2024 Update" are dead, are these still needed?

anyone having problems installing Swarm on runpod today? It worked like 2 days ago but been trying to start a new pod all day and the backend wont load. Have tried several times with different templates and different graphics cards and same issue with backend not loading each time? 20:06:47.372 [Init] Self-Start ComfyUI-0 on port 7821 is loading... 20:06:50.331 [Warning] [ComfyUI-0/STDERR] Traceback (most recent call last): 20:06:50.331 [Warning] [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/main.py", line 136, in 20:06:50.331 [Warning] [ComfyUI-0/STDERR] import execution 20:06:50.332 [Warning] [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/execution.py", line 13, in 20:06:50.332 [Warning] [ComfyUI-0/STDERR] import nodes 20:06:50.332 [Warning] [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/nodes.py", line 22, in 20:06:50.332 [Warning] [ComfyUI-0/STDERR] import comfy.diffusers_load 20:06:50.332 [Warning] [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/comfy/diffusers_load.py", line 3, in 20:06:50.332 [Warning] [ComfyUI-0/STDERR] import comfy.sd 20:06:50.332 [Warning] [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/comfy/sd.py", line 6, in 20:06:50.333 [Warning] [ComfyUI-0/STDERR] from comfy import model_management 20:06:50.333 [Warning] [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/comfy/model_management.py", line 168, in 20:06:50.333 [Warning] [ComfyUI-0/STDERR] total_vram = get_total_memory(get_torch_device()) / (1024 * 1024) 20:06:50.333 [Warning] [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/comfy/model_management.py", line 131, in get_torch_device 20:06:50.333 [Warning] [ComfyUI-0/STDERR] return torch.device(torch.cuda.current_device()) 20:06:50.333 [Warning] [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/venv/lib/python3.10/site-packages/torch/cuda/__init__.py", line 971, in current_device 20:06:50.333 [Warning] [ComfyUI-0/STDERR] _lazy_init() 20:06:50.334 [Warning] [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/venv/lib/python3.10/site-packages/torch/cuda/__init__.py", line 319, in _lazy_init 20:06:50.334 [Warning] [ComfyUI-0/STDERR] torch._C._cuda_init() 20:06:50.334 [Warning] [ComfyUI-0/STDERR] RuntimeError: CUDA unknown error - this may be due to an incorrectly set up environment, e.g. changing env variable CUDA_VISIBLE_DEVICES after program start. Setting the available devices to be zero. 20:06:50.626 [Info] Self-Start ComfyUI-0 unexpectedly exited (if something failed, change setting `LogLevel` to `Debug` to see why!)

Hello! I am new at all this and I'm running in a bit of an issue: I am trying to run SwarmUI_Kaggle_v13.ipynb as per instructed in this video: https://www.youtube.com/watch?v=XFUZof6Skkw At the 27:50 mark, it is instructed to ''go to the link you got from step 3 and click visit site'' However, I see in the comment section that step 3 is no longer required as Kaggle now supports Gradio live share. I do not know what Gradio live share is (I am a total beginner), is there any way someone can explain to me how to replace step 3 with Gradio live share so that I can open SwarmUI on Kaggle? Thanks and have a good day!

Hello Mr. Furkan.. I want to join your discord but it says the invitation has expired. how can I join. I am your monthly member.

Great are these tutorials, thank you very much. I have a problem with other checkpoints Flux downloaded from Civitai. The ones you provided work normally, the others don't want to. Can you give some hints which flux checkpoints will be correct in SwarmUI? Sorry for my English, this is an automatic translation, I don't know English :(

i played now a lot with the inpainting and redux. what i am not able to accomplish is inserting an object. i generated a picture and modified to my liking with inpainting. now i want to insert a picture of an object with inpainting. and the inserted object picture should mach the style of the base picture and also be positioned in a certain place.

Thank you Furkan for this fantastic FLUX Inpaint/Outpaint video. Just one question. Your Flux Guidance Scale is up to 30, but mine is max. 10. I am using 0.9.3.2. They changed this? Whats the difference?

Seems like the lora links are broken even on hugging face! :(

Is there a way you could share SwarmUI Presets? Like save all the settings you have where you generate this cool images and share it with us so we can load it in SwarmUI?

hi how do i use my modules ? do i need to download again ? i have stabled dufission it has modules and webui has same modules im running out of space

Any good reasons to crop images instead of using buckets? My training images have all sort of aspect ratios with relevant information I would like to keep.

I thought SwarmUI would be great...but it's not. Namely, i cant tell it where i want the generations to be placed(SAVED). Also, i have THOUSANDS of loras/embeddings/models that have spaces in them, and they wont load and crashes. I MEAN WOW... Why do i have to rename thousands of files when ever other UI can use them just fine no matter how they are named. PLUS, when it was loading the folders i placed to maunally point to them, the program was pulling from and compairing my database to one from Civitai!!! thats a serious issue when i didn't tell it to do this. I DONT want my database compromised and it looks like that this UI was doing it. WHY was it verifying my models/loras/embeddings?

hi, have followed instructions to load swarm on several runpods in the past but today I'm getting this error after the restart. 08:50:09.786 [Init] Starting webserver on http://0.0.0.0:7801 08:50:09.817 [Init] Swarm is up to date! You have version 0.9.2.3, and 0.9.2-Beta is the latest. 08:50:09.825 [Init] Program is running. 08:50:10.547 [Init] Self-Start ComfyUI-0 on port 7821 is loading... 08:50:12.437 [Warning] [ComfyUI-0/STDERR] Traceback (most recent call last): 08:50:12.437 [Warning] [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/main.py", line 90, in 08:50:12.437 [Warning] [ComfyUI-0/STDERR] import execution 08:50:12.437 [Warning] [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/execution.py", line 13, in 08:50:12.437 [Warning] [ComfyUI-0/STDERR] import nodes 08:50:12.437 [Warning] [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/nodes.py", line 21, in 08:50:12.438 [Warning] [ComfyUI-0/STDERR] import comfy.diffusers_load 08:50:12.438 [Warning] [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/comfy/diffusers_load.py", line 3, in 08:50:12.438 [Warning] [ComfyUI-0/STDERR] import comfy.sd 08:50:12.438 [Warning] [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/comfy/sd.py", line 5, in 08:50:12.438 [Warning] [ComfyUI-0/STDERR] from comfy import model_management 08:50:12.438 [Warning] [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/comfy/model_management.py", line 143, in 08:50:12.438 [Warning] [ComfyUI-0/STDERR] total_vram = get_total_memory(get_torch_device()) / (1024 * 1024) 08:50:12.438 [Warning] [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/comfy/model_management.py", line 112, in get_torch_device 08:50:12.438 [Warning] [ComfyUI-0/STDERR] return torch.device(torch.cuda.current_device()) 08:50:12.438 [Warning] [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/venv/lib/python3.10/site-packages/torch/cuda/__init__.py", line 878, in current_device 08:50:12.438 [Warning] [ComfyUI-0/STDERR] _lazy_init() 08:50:12.439 [Warning] [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/venv/lib/python3.10/site-packages/torch/cuda/__init__.py", line 314, in _lazy_init 08:50:12.439 [Warning] [ComfyUI-0/STDERR] torch._C._cuda_init() 08:50:12.439 [Warning] [ComfyUI-0/STDERR] RuntimeError: CUDA driver initialization failed, you might not have a CUDA gpu. 08:50:12.645 [Info] Self-Start ComfyUI-0 unexpectedly exited (if something failed, change setting `LogLevel` to `Debug` to see why!) 08:50:12.645 [Info] Self-Start ComfyUI-0 had errors before shutdown: [ComfyUI-0/STDERR] Traceback (most recent call last): [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/main.py", line 90, in [ComfyUI-0/STDERR] import execution [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/execution.py", line 13, in [ComfyUI-0/STDERR] import nodes [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/nodes.py", line 21, in [ComfyUI-0/STDERR] import comfy.diffusers_load [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/comfy/diffusers_load.py", line 3, in [ComfyUI-0/STDERR] import comfy.sd [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/comfy/sd.py", line 5, in [ComfyUI-0/STDERR] from comfy import model_management [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/comfy/model_management.py", line 143, in [ComfyUI-0/STDERR] total_vram = get_total_memory(get_torch_device()) / (1024 * 1024) [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/comfy/model_management.py", line 112, in get_torch_device [ComfyUI-0/STDERR] return torch.device(torch.cuda.current_device()) [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/venv/lib/python3.10/site-packages/torch/cuda/__init__.py", line 878, in current_device [ComfyUI-0/STDERR] _lazy_init() [ComfyUI-0/STDERR] File "/workspace/SwarmUI/dlbackend/ComfyUI/venv/lib/python3.10/site-packages/torch/cuda/__init__.py", line 314, in _lazy_init [ComfyUI-0/STDERR] torch._C._cuda_init() [ComfyUI-0/STDERR] RuntimeError: CUDA driver initialization failed, you might not have a CUDA gpu.

Yea I reinstalled and it works fine now, thanks! By the way, I manually deleted the 'SwarmUI' folder before installing again using your command lines. Was that the correct way? I mean was there a 'proper' way to uninstall SwarmUI first? Thx

Hi, everything was working for me in Runpod using Swarm with Comfy based on your instructions, until today suddenly ComfyUI is not showing any nodes and the 'Manager' button is missing from the floating buttons part. Please help, thanks a lot. (I have been using the Pytorch 2.1 template in Runpod, should I use 2.4 or 2.5 ? )

Hi, I am using RunPod with SwarmUI and from there ComfyUI, please can you help me install ComfyUI manager?

Hello, I get an error below as install swarm_ui yordam giris noktası, D:\swarm_ui\SwarmUl\dibackend\comfy\python_embeded\Lib\site-pa kages\torchaudio\lib\libtorchaudio.pyd dinamik baglantı kitaplıgında bulunamadı.

strange problem. I downloaded your caddy notebook. I did everything according to the instructions. Please help who can. But SwarmUI does not generate on kagli with any model. Attempts were with models in the temporary folder, and the model loaded during installation. The process hangs on loading the model and that's it. == SwarmUI logs 2024-09-02 16:37 == 16:37:34.184 [Info] User local requested 1 image with model 'OfficialStableDiffusion/sd_xl_base_1.0.safetensors'... In debug logs there is constantly waiting for server [Debug] [Load ComfyUI Self-Starting #0] ComfyUI-0 port 7821 waiting for server...

Bye. I joined your patreon yesterday and plan to follow many of your wonderful guides. I am starting with this one to use flux with SwarmUI on kaggle (I already use flux with forge on lightining.ai and with lora trained models on replicate). i am not an expert so i need a little help on a couple of things: I tried yesterday and managed to get it all started on kaggle. Today however I noticed that I have to reinstall everything every time because obviously the temp folder is deleted. That's fine. But I thought there was no need to reinstall swarmUI on kaggle/working, because it stays. But when I try to start it up it tells me I don't have permissions. So I reinstalled SwarmUI as well. But there's a problem: the backends don't start. They just stay loading. Can the permissions issue be resolved? And why do the backends remain loading? Edit: I have this error when start SwarmUI: 22:04:53.478 [Warning] [ComfyUI-0/STDERR] Traceback (most recent call last): 22:04:53.479 [Warning] [ComfyUI-0/STDERR] File "/kaggle/working/SwarmUI/dlbackend/ComfyUI/main.py", line 87, in 22:04:53.479 [Warning] [ComfyUI-0/STDERR] import comfy.utils 22:04:53.479 [Warning] [ComfyUI-0/STDERR] File "/kaggle/working/SwarmUI/dlbackend/ComfyUI/comfy/utils.py", line 20, in 22:04:53.480 [Warning] [ComfyUI-0/STDERR] import torch 22:04:53.480 [Warning] [ComfyUI-0/STDERR] File "/kaggle/working/SwarmUI/dlbackend/ComfyUI/venv/lib/python3.10/site-packages/torch/__init__.py", line 290, in 22:04:53.480 [Warning] [ComfyUI-0/STDERR] from torch._C import * # noqa: F403 22:04:53.480 [Warning] [ComfyUI-0/STDERR] ImportError: /kaggle/working/SwarmUI/dlbackend/ComfyUI/venv/lib/python3.10/site-packages/torch/lib/../../nvidia/cusparse/lib/libcusparse.so.12: undefined symbol: __nvJitLinkComplete_12_4, version libnvJitLink.so.12 ps. Sorry, I'm italian I'm don't speak english very well. Deepl is my translator ;)

First of all, thanks for all the great tutorials. I've just tried to get this running on Runpod. Everything works up until I have to connect the http connection. It doesn't appear as a button on my pod. And opening 0.0.0.0: 7801 from my browser also doesn't connect. I'm on a Mac.

Do you think kohya will support training on either SD3 or the flux models? Currently my training stopped at SDXL.

This tutorial is the best one I've seen and watched so far on SD! Wow - just so many details. I have also discovered the incredible power of in comfyui. Nothing could get close to Adetailer until I discovered this feature. I am now generating incredible high quality pictures using this feature. Adetailer was the one item left that was keeping me on A1111 and now I think I am a convert to SwarmUI. The features you describe are so easy to use. Thanks for all your effort on this video as it must have taken quite a long time to put togehter!

FYI Civitai banned SD3 because of their licensing..."as written the creator license seemed to imply that Stability had edit and takedown power over not just any resource trained on SD3 but also on any resources that used its outputs in their datasets." useful to know if you are posting images created with it or training on it.

Any plans for comparing SD3 with other models (SDXL, SD1.5) for dreambooth training?

Please, if you can post a Kaggle or colab notebook it would be great, as I don't have a GPU intensive machine....

Thanks for the styles , but it is lacking in details too much, I tried to 2x upscale , still coming out very bland. Its fast ,i should accept it. But not worth as of now. WE need loras/checkpoint to make it good. SDXL Still wins

Is it possible to have swarmUI running on one machine, say a Mac and using the resources (GPUs and CPU) on another machine (say Ubuntu)?

also looking for the option to point the model folder to look in the A1111 folder for checkpoints instead of the root. anyone find that? EDIT: ANSWER is at 14:24 in the video he put out... SERVER>SERVER CONFIGURATION>PATHS

I am excited for this tutorial! A super one-two punch, as I've haven't used the SwarmUI yet either.

I see the SD3 devs recommend and released a sample ComfyUI workflow. They also have a few different text_encoders options. Are you planning to do any experimenting with their flow and the various text encoder options? You do such a great job evaluating and finding ideal settings so I was just curious :)

This is really cool. Looking forward to the tutorial. Thanks.

Benefits/drawbacks?