🐝｜swarm-ui

[hearty notch]

working with comfy editor? or need to disable it?

[dim sonnet]

if you use custom workflows, you gotta add LoraLoader nodes for any loras you want

[inner vault]

@dim sonnet ohhh for some reason when i try to select a VAE it only detects 2 out of 6 that i have ...

[dim sonnet]

the loras selector on the main generate tab is only for when not using custom workflows

[dim sonnet]

are you sure the other 4 are in the correct folder?

[inner vault]

yep

[dim sonnet]

oh

.pt files lol

i doesn't even check for those atm

fixed

[inner vault]

ohh i see so itll be fixed right ?

xD

already

@dim sonnet yo what happened xD i updated and now this happened

w8

ohhh sorry

sorry

it was my fault i had another instance open and didnt notice

[dim sonnet]

lol you have two copies open

yep

[inner vault]

xD

yo btw in automatic 1111 there is this extension that lets u zoom in on picture when you want to inpaint with finer details or have a better look you know its actually really usefull

[dim sonnet]

you can click an image to view it fullscreen

there's no inpaint ui yet

[inner vault]

full screen is not same tho .. like when you zoom in you zoom till you start seeing pixels ... am just throwing the idea out there if you need it u know

aslo do you think that this UI will be compatible with the extentions that are already available in 1111 or this is not possible

[cursive badger]

How do I use the lora? The lora:name did not worked. And if I use a cutom UI workflow, can I use Loras with normal weights?

Also your UI (because of multi GPUs) would be so useful to test Loras block weghts . That would be what I would most use it for https://github.com/hako-mikan/sd-webui-lora-block-weight

[dim sonnet]

CTRL+ mouse wheel up to zoom on any webpage ever?

[dim sonnet]

that's not possible no

[dim sonnet]

lora support in the main UI is brand new and i'm currently wrapping up weighting support

if you use a custom comfy workflow, just add em via nodes

not the prettiest UI in the world but it works

[inner vault]

doesnt work proprly + in inpainting you cant do that cuz you want the brush to scale with u

[dim sonnet]

oh shit you right, can't zoom in properly in the UI lol

and inpainting UI doesn't exist yet

[inner vault]

well i figured ... but it ll be nice if its extensions friendly

[dim sonnet]

it is extension friendly

just, auto webui extensions are built for auto webui

[cursive badger]

It worked from a custom workflow. Just wish lora block weights would be a think for comfyui, but could not find it anywhere

[inner vault]

i know i know am just saying .... btw comfy ui already have inpainting in it

[dim sonnet]

i'm aware

[inner vault]

ofc you are .. you are the best dev a company could have

😉 keep the good work

also noticed something else xD when you open the pic in full screen there is no close button xD you ll have to click in the edge of screen to leave

[dim sonnet]

yes

[inner vault]

is it on purpose ?

[dim sonnet]

yes

i had a convenient close button in a prior version but nobody liked it

[inner vault]

idk man like i had to randomly click to figure that out ... cuz its ltrly in the far edge of the screen

i think it should be when you click on the gray area also when you click ESC

[dim sonnet]

oh, huh, escape key doesn't work

fixed, escape key works now

[inner vault]

great

also is there a way to control the upscale i tried to upscale x2 and it gave an entire diffrent image

[dim sonnet]

yeah that's kinda just a quick n dirty placeholder to get a little bit of functional in before there's a proper image editing UI

you can do slightly more controlled upscaling if you do it in the Refiner settings

but a full proper thing is pending the dedicated image editing interface

[inner vault]

will there be a tiled upscaler ?

[hearty notch]

i am planning to try t his in few as I am running a grid render https://github.com/ssitu/ComfyUI_UltimateSDUpscale

you wanna try and share feedback?

[inner vault]

bro it works perfectly even better and faster than the 1 in A1111 . also you can stack it like i did ... tho i d love if you can go higher than x4 times and also if it can show the resolution you gonna end up with

[hearty notch]

yes, I just tired it too

very nice

but I can not copy to generate tab anymore after the most recent update

[inner vault]

wdym i dont understand

[hearty notch]

@dim sonnet looks like the last update is not allowing use this workflow in generate tab anymore

[hearty notch]

when you click on use this workflow in generate tab nothing is happening

[inner vault]

idk it works for me

[hearty notch]

not for me

[inner vault]

try updating again

cuz every min there is an update

[hearty notch]

then do not pull the recent one, looks like there is a bug, hope that @dim sonnet will look into that

can you share your json workflow?

[inner vault]

@hearty notch btw look at the result from the ultimate upscaler xd amazing

[hearty notch]

i know, really amazing

[inner vault]

wait xD xD i just noticed th tiles xD

[hearty notch]

??

you are not using refiner?

try with refiner hahah

it's even way more better

[inner vault]

wait can you sent your wroklow ?

send*

[hearty notch]

ummm, looks like my workflow is is the issue

hold one I will share it

[dim sonnet]

when in doubt, first thing to try is update and CTRL+F5

yeah if there's a workflow-specific issue, gimme workflow

[hearty notch]

this is the offical one from Comfy,

[dim sonnet]

works fine

[inner vault]

nah i meant like send me the .json file

[hearty notch]

do you mind if I DM it?

[dim sonnet]

uh, sure

[inner vault]

@hearty notch can you send me also i wanna se how you used the refiner

[hearty notch]

load this png into your comfy tab

it will load the nods

as the one I created is broken now for some reason

[inner vault]

i did but there is no upscalers!

[hearty notch]

yeah as I was sharing all WF with upscaler that i build is broken and loadding anymore

[hearty notch]

but you can add it the same way you did with this WF

[inner vault]

i just add it in the end ?

[hearty notch]

yeah

before the save image nod

[dim sonnet]

got tanuki's bug replicated & fixed

[inner vault]

great

[hearty notch]

let me test it hahah

[inner vault]

damn the details

tho how can i fix the tiling effect ?

[hearty notch]

do not use only the stock settings

play around with tile padding a bit and find what suites your taste

snapshot your upscaler nod and share here

let me see your settings

[hearty notch]

yeah as I guessed, 32 for tile padding is not enough

which GPU you are using?

[inner vault]

3060ti

am trying to play with diffrent settings

[hearty notch]

VRAM?

[inner vault]

ohh xD it has 8GB

[hearty notch]

12?

ah shit!

[inner vault]

xD

bro wth xD

when i turned of seam-fix-mode its fixed xD

isnt it supposed to be the 1 fixing it ?

[hearty notch]

increase the tile padding also

[inner vault]

i did now its 128

[hearty notch]

and do not go berzerk on the upscale

keep it like X2 fro now

[hearty notch]

yeah thats not bad

@dim sonnet what is the sampler/ schedule you recommend for this UI?

for best results

[inner vault]

bri idk even with 128 i still can notice some tiny tiling

should i go even more ?

[hearty notch]

donly get too greedy

[hearty notch]

yeah, try more and see,

usually 128 always is my sweet spot

[inner vault]

bro its bothering me xD

its so noticeable for me specially in the sky

[hearty notch]

128 is working with me tbh

[inner vault]

fuck it xD am going double or nothing xD

tell me you dont see it

[dim sonnet]

everyone has differing claims. Euler x Normal is the official default reference, but DPM++ 2M x Karras is probably better

[hearty notch]

also keep all your res native to SDXL, 1024

[hearty notch]

this is what I am testing all night today, this is the same one?

[dim sonnet]

no that is not

[hearty notch]

ah ok let me try

[hearty notch]

oh god no!

[inner vault]

bro no diffrence between 128 and 256 🙃 wth

[inner vault]

the horror

[hearty notch]

hahah i dont know, it's ok for me here

[inner vault]

screen shot for me plz

the node

[hearty notch]

there is a bit of tiling for me, but meh I do not mind it

try searching another tile nod and let me know

https://github.com/BlenderNeko/ComfyUI_TiledKSampler

try this and let me know

[inner vault]

yo i just increased mask blur to 32 and there is no more tiling at all it even added details ... i think yours had minor tiling cuz it was 1024 res but try also putting 16-32 mask blur

@hearty notch

[inner vault]

yo can sm1 send where can i find VAEs for SDXL ? cuz SD vae doesnt seem to work on it

[dim sonnet]

you don't really need an alternate VAE, the builtin one is fine

[barren moat]

I managed to break it somehow. I changed the IP in the server settings, shut down the console to restart and now it's throwing these errors... Is there a config file somewhere I can edit?

[toxic shale]

so about that url XD

[toxic shale]

http://localhost:7801
should look like this afterwards

[worldly oxide]

Some Questions for StableSwarm UI 🙏🏾

1. If you already have ComfyUI installed, in detail how do I link ComfyUI directory folder with Swarm UI?
2. Does Swarm UI understand every custom node or is it only a small set of nodes and more support to come?
3. Can I have multiple folders and sub folders for saving output with the seed number and model in the file on the images like I can do easily in Automatic 1111?
4. Does Swarm UI have an Image Gallery?

[hearty notch]

did you install swarm? I am sure all your qustions will be answered if you do install it

also, read the documentation posted on GitHub

[worldly oxide]

Thanks for trying to help, but the answers are not there!

[hearty notch]

1. it will install comfyui when you use the setup guide, and there is an advanced option that will use a pre-setup backend, either CUI or A1111 I did not try it but you can also edit and create these once you set it
   2.swarm is yousing comfy as a backend, so basically anything CUI can do swarm can do
   3./4.SSUI have a good out of the box gallery I am not sure how deep you want to customize it, but personally out of the box is a good option for me

you can always install and try to see if this is a good for you or not

[toxic shale]

1.) install without included comfy. once its setup, link existing comfy via port as I screenshotted above
2.) so far, it's worked with all custom nodes. probably depends on how it was coded and if it used the existing functions

...and tanuki answered while iw as typing XD checks out

[worldly oxide]

Hey @hearty notch @toxic shale thanks for the tips, the mistake I was making was having my dir set to *E:\Stable Diffusion* instead of *E:\Stable_Diffusion* because Swarm UI doesn't like gaps so this fixed the problem, all up and running now!

[sudden parcel]

That last point is actually very interesting to me. Because I had trouble with it recognizing my folders and had no idea what might cause it. And I'm pretty sure that this is the reason.

[toxic shale]

@dim sonnet I think it may be worth investing in a path variable encoder & decoder that you use all around XD

sanitize em urls

[dim sonnet]

yes, StableSwarmUI/Data/Settings.fds

you can also launch with --host localhost as is documented at https://github.com/Stability-AI/StableSwarmUI/blob/master/docs/Command Line Arguments.md

[dim sonnet]

re 3 yes it's very customizable, see also docs about custom path formats @ https://github.com/Stability-AI/StableSwarmUI/blob/master/docs/User Settings.md#path-format

[dim sonnet]

no listen you don't understand how much sillier the problem is

some inputs require quotes around paths with spaces

other inputs forbid quotes around paths

like not in my code, but in windows and etc

so if spaces misbehave it's usually because something somewhere has quotes where quotes are forbidden or vice versa

also sometimes quotes get doubled up because some systems require quotes but don't parse the quotes

Last I tried spaces worked for a comfy launch dir on windows but i'll re-test

yep, re-tested, launches fine

I haven't tested what if comfy with spaces installed on a secondary drive yet tho so lemme test that

nope even E:\ with a space works

oo

omg if swarm's filepath has spaces, then comfy can't load (because quotes are needed in the thing that passes the config around)

[dim sonnet]

fixed, spaces in a path should work now

[barren moat]

it won't boot back up to fix it

[dim sonnet]

I think caith thought you mixed up a backend instead of your core server settings

[barren moat]

ahhh ok

Not sure what happened

[dim sonnet]

check the console window

[barren moat]

error: The source specified has already been added to the list of available package sources. Provide a unique source.

Usage: dotnet nuget add source [arguments] [options]

Arguments:
PackageSourcePath Path to the package source.

Options:
-n|--name Name of the source.
-u|--username Username to be used when connecting to an authenticated source.
-p|--password Password to be used when connecting to an authenticated source.
--store-password-in-clear-text Enables storing portable package source credentials by disabling password encryption.
--valid-authentication-types Comma-separated list of valid authentication types for this source. Set this to basic if the server advertises NTLM or Negotiate and your credentials must be sent using the Basic mechanism, for instance when using a PAT with on-premises Azure DevOps Server. Other valid values include negotiate, kerberos, ntlm, and digest, but these values are unlikely to be useful.
--configfile The NuGet configuration file. If specified, only the settings from this file will be used. If not specified, the hierarchy of configuration files from the current directory will be used. For more information, see https://docs.microsoft.com/nuget/consume-packages/configuring-nuget-behavior.
-h|--help Show help information

MSBuild version 17.6.8+c70978d4d for .NET
Determining projects to restore...
All projects are up-to-date for restore.
StableSwarmUI -> D:\StableSwarmUI\src\bin\live_release\StableSwarmUI.dll

Build succeeded.
0 Warning(s)
0 Error(s)

Time Elapsed 00:00:00.52
09:33:45.864 [Init] === StableSwarmUI Starting ===
09:33:45.895 [Init] Prepping extension: StableSwarmUI.Builtin_StabilityAPIExtension.StabilityAPIExtension...
09:33:45.901 [Init] Prepping extension: StableSwarmUI.Builtin_ScorersExtension.ScorersExtension...
09:33:45.901 [Init] Prepping extension: StableSwarmUI.Builtin_GridGeneratorExtension.GridGeneratorExtension...
09:33:45.901 [Init] Prepping extension: StableSwarmUI.Builtin_DynamicThresholding.DynamicThresholdingExtension...
09:33:45.902 [Init] Prepping extension: StableSwarmUI.Builtin_ComfyUIBackend.ComfyUIBackendExtension...
09:33:45.902 [Init] Prepping extension: StableSwarmUI.Builtin_AutoWebUIExtension.AutoWebUIBackendExtension...
09:33:45.920 [Init] Parsing command line...
09:33:45.920 [Init] Loading settings file...
09:33:45.926 [Init] Re-saving settings file...
09:33:45.932 [Init] Applying command line settings...
09:33:45.942 [Init] Prepping options...
09:33:46.015 [Init] Loading models list...
09:33:46.034 [Init] Loading backends...
09:33:46.035 [Init] Loading backends from file...
09:33:46.040 [Init] Initializing backend #0 - ComfyUI Self-Starting...
09:33:46.040 [Init] Prepping API...
09:33:46.041 [Init] Prepping webserver...
09:33:46.047 [Init] Self-Start ComfyUI on port 7820 is loading...
09:33:46.138 [Init] Scan for web extensions...
09:33:46.139 [Init] Readying extensions for launch...
09:33:46.139 [Init] Launching server...
09:33:46.139 [Init] Starting webserver on http://localhost:7801
09:33:46.158 [Init] Program is running.
09:33:46.664 [Init] Launch web browser...
09:33:46.860 [Info] Creating new admin session for 127.0.0.1
Traceback (most recent call last):
File "D:\StableSwarmUI\dlbackend\comfy\ComfyUI\main.py", line 143, in <module>
load_extra_path_config(config_path)
File "D:\StableSwarmUI\dlbackend\comfy\ComfyUI\main.py", line 109, in load_extra_path_config
with open(yaml_path, 'r') as stream:
^^^^^^^^^^^^^^^^^^^^
OSError: [Errno 22] Invalid argument: '"D:\StableSwarmUI/Data/comfy-auto-model.yaml"'
09:33:48.383 [Info] Self-Start ComfyUI on port 7820 exited.

[dim sonnet]

i'm going to cry

[barren moat]

I'm sorry

[dim sonnet]

I just added quotes there to fix somebody else's bug with spaces in the line

and you have an error that the quotes aren't being stripped properly

why does windowsgod hate us

okay it now only adds quotes if there's a space

maybe that will suffice??

[inner sand]

Just installed Stableswarm-ui on two machines (home and work). Both proceeded flawlessly. Extremely impressed by the speed and frugal VRAM use, although I do have a 3090 with 24GB VRAM (home). REALLY appreciate the advance this represents as I started out (like most) with A111 and a ton of extensions before migrating to ComfyAI - before the leap forward in functionality that Stableswarm represents. Congrats! Much appreciated!

[barren moat]

ok I'll update and try it

Looks like we're all good!! Thank you @dim sonnet !

[narrow blade]

got this error

i dont see any error in the logs

[barren moat]

try to update and re-launch it

might be the same error I was getting

Alex just made a fix and it's live now

[dim sonnet]

that looks like you just opened the workflow editor tab before comfy was loaded, so just hit Try Again and you should be good?

[narrow blade]

no, tried clicking on the button

but same issue

[dim sonnet]

Oh. Then, uh, yeah try what Jay said, update and relaunch

oh and CTRL+F5 on the page to be safe

[narrow blade]

wait

i closed it

and restarted and deleted one of these

which server should I select?

[dim sonnet]

if you used the default comfy installer, you want ComfyUI Self-Starting and set StartScript to dlbackend/ComfyUI/main.py

[toxic shale]

would it be possible to get optional (via tick box) vertical text for left column on the infinite grid?
would makes exporting images so much simpler if a wall of text wasn't exported on save image of current view XD

also, during grid generation, individual images often dont render for absolutely no reason. rerunning script though, from scratch, will results in some other random image not generating

no pattern to it really

I feel like it's usually one image out of like 20 where this occurs

[narrow blade]

will use it\

[narrow blade]

just startscript?

[dim sonnet]

ye

[narrow blade]

oh wait

working

[narrow blade]

@cunning holly how to use x/y/z on swarm?

[dim sonnet]

Tools at the bottom

[narrow blade]

👍

i am not getting a dropdown to select the models/loras

how can I get that?

[dim sonnet]

huh?

if you're adding a model axis, click the Fill button on the right

it's a text field not a multiselect field (I should probably update it to have a multiselect inputter thingy as an option, definitely not the main input given how many complaints the multiselect in auto webui got when they added that)

wouldn't entirely make sense for loras too, since for loras you can do lora1,lora2 || lora1 || lora2 wherein the first value in the axis uses both loras at once

[narrow blade]

i mean i dont have a dropdown menu to select loras/checkpoints in the menu(same in x/y/z too)

[inner vault]

@dim sonnet yo yo ... so why is it when i click on an embedding to use nothing happens ?

[dim sonnet]

yes that's not fully implemented yet

embedding tab is a glorified file browser only atm

[inner vault]

ohh ok i see

also noticed that the embedding doesnt read the thumbnails i already have for them

or wait ... my cuz A1111 adds "preview" to the pic name maybe that is the reason

cuz in the loras it only says the name

like this

[dim sonnet]

oh boy, more filename variants to account for

[dim sonnet]

added support for recognizing that

[inner vault]

xD happy to help xD

@dim sonnet you said that refiner should work with SD models and it does in the default generate tab settings but when i try doing the same thing in nodes it gives me this results here ... can you check my workflow and tell me what am i doing wrong ?

[dim sonnet]

without even looking i'd bet VAE issues

to use two different types of model, they have different VAEs, so you need to decode+encode before sending over

[inner vault]

so between woth samplers i do it ?

right ?

[dim sonnet]

ye

[inner vault]

tho which vae do i use the SD model or the rifiner ?

[candid hornet]

Hi I would like to play with swarm, but I have a question. Is confy in swarm fully functional? Could I uninstall my confy?

[inner vault]

no bro its part of it if you dont have the comfy you gonna have to install it

but if you have it than keep it

[candid hornet]

Ok so I won't have 2 instance of it

[inner vault]

yeah tho be careful to not download it again in Swarms ui installation cuz it is going to give you the option

[dim sonnet]

btw for the record the way to use embeddings anyway is just type embedding:filenamehere.pt into the prompt input

[dim sonnet]

it is fully functional yes.
You can either keep your existing install, or use the one that the swarm installer generates for you, either is fine

[dim sonnet]

the first model decodes after it's done sampling, and then the second model's VAE re-encodes and is used to decode at the end

[inner vault]

bro i know i know its just easier to do it from tab that is why i asked xD

[inner vault]

yeah i figured that out tho thanks

@dim sonnet yo... so in the default settings of the template i use the SD/SDXL sampler is set return_left_over_noise activated and in the refiner the add_noise is set to desactivated (like in the pic)... but i noticed that when you using SD models for some reason that adds some weird artifacts . so i did the opposite where i set off the first one and turned on the second ( like in the 2nd pic) ... it seemed to work fine , but is it the right way ?

[dim sonnet]

add_noise on, return_leftover off is to my knowledge just the proper way to do it

people have been playing with it and finding other ways of doing it sometimes does neat things but like

generally you don't want noise left at the end

[acoustic briar]

Question: Does SwarmUI have an option like ComfUI's "automatially queue prompt when queue size hits 0" option, which allows me to generate continually?

[inner vault]

yeah and also that what gave best results for me too

yo anyone knows how to create something like a high-res fix in comfy UI

[dim sonnet]

oo, no, but that's a good idea to add

[dim sonnet]

same refiner-like process but toss an upscale in the middle

the generate tab can automatically emit that as well, there's an upscale option under refine

[inner vault]

i know i know but i want to create A1111 like experience in nodes where you have all the same shit and you have the option to use them if you know what i mean ... tho id love if there was a way to kind skip nodes kind of make the inactive while still keeping the current without any effect on the output ... tho now thinking about it that will only work with nodes that have same type of input and output

[inner vault]

so i just put in between the decoder and encoder from earlier ?

[dim sonnet]

ye

[hearty notch]

@dim sonnet i have 2 machines at home, the one I am testing SSUI with, and another one with a 3060, is there anything can be done to connect the second machine and use the extra power from that 3060 with my current machine ?

[inner vault]

i think when they said multi GPU they meant GPUs in same machine connected with an SLI connecter BUT am not sure so dont take my word for it

[hearty notch]

now that will be the dream hahaha, but I am not sure if SLI is still a thing, but thanks for the note

[dim sonnet]

nope

[inner vault]

i mean current high end GPUs like 3090s have smthn similar

[dim sonnet]

yep literally just boot up comfyui on the other machine and add it as a backend

[inner vault]

xD what xD really

[dim sonnet]

ye, the additional GPUs can be anything anywhere

[inner vault]

how is that even working

xD

[dim sonnet]

can even use google colab or rented servers or whatever you want

[inner vault]

ok that is actually impressive

[hearty notch]

we really need some proper documentation, your UI is backed with tons of featuers that no one even knows about hahah

[hearty notch]

this is why am asking

[inner vault]

didnt know that was a possible thing

[hearty notch]

pool VRAM no one manage to crack it, or SLI as you called it

[inner vault]

wait a sec you can combine Vram right ?

[hearty notch]

can you share how? we are not very well-versed with these settings .. haha just the general public

[hearty notch]

no

this is what I am saying

you can not pool VRAM in one machine

hence why I am setting up my old 3060 in another machine and now I am asking how to harness this extra juice with my current 4090

[inner vault]

so just the rendering power ?

[hearty notch]

I am waiting on Alex to share more info

[inner vault]

@dim sonnet enlighten us ||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​|| _ _ _ _ _
https://cdn.discordapp.com/emojis/1004939086546604102.gif?size=64&quality=lossless

[hearty notch]

https://cdn.discordapp.com/emojis/1004939086546604102.gif?size=64&quality=lossless

[barren moat]

Ok so we need two separate machines?

[hearty notch]

still waiting on some extra info from alex,

and soon I will sell my kidnees to pay the power bills

[dim sonnet]

there i wrote a doc about it https://github.com/Stability-AI/StableSwarmUI/blob/master/docs/Using More GPUs.md

[hearty notch]

bless your code!

[dim sonnet]

however you get multiple GPUs, you can use em, it's designed to be very generic to bend to different setups easily

[barren moat]

Awesome!! I’m going to checkout out that doc now

[hearty notch]

so it will use a separate backend on each gpu regardless if its oner machine or multi?

ah ok I get it, it will split the gens

sweet!!!

[inner vault]

@dim sonnet yo alex ... do think this kind of tech that utilizes 2 gpus at once can be used in other programs like 3D render blender for example or maybe games altho that would result in a massive massive delay

[inner sand]

Exactly how is that done? I have two GPU's in my home computer - an RTX3090 with 24GB VRAM and an RTX 2080 Super with another 8GB I have an open frame MB with the potential for 6 GPU's. I tried comma delineating the GPU's but that didn't work. So what is the secret...multiple instances of the same program??

[hearty notch]

you need to create 1 backend per gpu

[hearty notch]

I am following the steps, but I can not find the backend on my main machine

do we need to ad the port to the address? and what is the exact format? like this?

also do we need --listen on the main machine only? or secondary?

[toxic shale]

http is needed

[hearty notch]

not working also

[toxic shale]

is how it looks for me

[hearty notch]

ah you are using 2 machines?

can you show the settings for both?

[toxic shale]

is on same machine. haven't tried on different machines

[hearty notch]

well I am trying to run 2 backend in one machine

[inner sand]

So...then if I do a batch file of say 4 images, two are processed by each GPU using their respective back ends? I presume (but do not know, so I ask) all four images can be saved to the same output folder w/o conflict?

[hearty notch]

the confusion is how to setup both properly, the wiki shared form @dim sonnet when I followed it did not workout, I guess I am issing something

[hearty notch]

#🐝｜swarm-ui message just read this

the UI is what using the backend so imgs will be saved where UI is controlling

not the backend

[inner sand]

Sue thing. I love to learn! I bet not too many 70 year-olds like me here! LOL!

[hearty notch]

we are all still learning this UI

[inner sand]

Heck, I'm still learning Discord!

[dim sonnet]

A similar concept has been used in games before (as variant of SLI) but it doesn't really go well because of the raw timing precision needed for games

for 3D rendering, it absolutely is used

Do you think movie studios render CGI on one machine? Nah they got entire server farms full of GPUs to render it out

[dim sonnet]

on the secondary machine, --listen tells comfy that it should open itself to remote connections

the 0.10 on the page was an example, you need to find your actual lan ip

[hearty notch]

I have that

still not working

[dim sonnet]

can you open it in a web browser?

[hearty notch]

no

[inner sand]

Nice read. Short, sweet and to the point. I get it now!

[dim sonnet]

well, get that working first

[hearty notch]

how?

[dim sonnet]

get the network connection working, then add the actual swarm part in

that sure is a question lol

It's either (A) you started it wrong, (B) your firewall is blocking it, or (C) your LAN address is wrong

A can be checked with a screenshot of the comfy console, C i could tell you pretty quick if it's right or not if you tell me what ip and where you got it

[hearty notch]

all ABC are not wrong, I am using the 2md machine in remot desktop as we speak, I allowed all firewall, I got the IP from cmd..ipconfig

i even changed the port on the 2d machine

[dim sonnet]

Well... if nothing was wrong, it'd be working, wouldn't it?

it ain't working, so something is wrong

[hearty notch]

this is the 2nd machine

[dim sonnet]

OH

i see

you need comfy on the second machine, not swarm

[hearty notch]

whatis the name of the app as it may appear on firewall? so I can double check

[dim sonnet]

you can use the comfy in swarm, but that puts it on port 7820 by default

it'd probably show up as just python tbh

good ol' unhelpfully vague python

[hearty notch]

can you elaborate on this part? 😄

[dim sonnet]

on your main machine pop open (whateverip):7820

in a web browser

it'll probably work

[hearty notch]

ah there is one for comfy and one for swarm now I see

hold on

[dim sonnet]

weirdly enough i haven't built a swarmui backend for swarmui yet

[hearty notch]

but now both comfy are on 7820

[dim sonnet]

though i do need to do that for an autopropagator eventually

[hearty notch]

we can change that?

[dim sonnet]

ports are per-machine, so it's not a problem

[hearty notch]

so now both are running

and swarm will automatically distribute gens into both servers? because nothing running on backend no 2

[dim sonnet]

ye

as long as you're generating more than 1 thing at once it should distribute them automatically

[hearty notch]

well nice thing I can even acess from my phone haha nice nice,

will be an amazing option to have a phone app for swarm that will run these local backend

not sure if you are thinking about such option

[hearty notch]

where gens will be saved? on main machine? or local machine?

ah, so I need to have lora on both machine also

[dim sonnet]

mobile support is on the TODO list ye

it's technically possible to open in mobile now, it just doesn't format cleanly for mobile atm

... well i haven't actually looked at all...

[hearty notch]

yeah you can run the comfy on the mobile browser,

[dim sonnet]

oh god i tried it's bad lol

yeah definitely need a dedicated mobile format lol

[dim sonnet]

on main machine

wherever Swarm is running

[hearty notch]

I recon that, I got one error when I load a lora, then I remember tthat we beed to have on both machines too lol

it's funny tho to see 3060 spitting 1.8 it/sec while the 4090 is firing up 7+it/sec XD

are you planning like some proper youtube deep dive and learning how to use Swarm?

[dim sonnet]

owo! https://www.youtube.com/watch?v=330z7P_m7-c

[hearty notch]

i never liked his videos, just reading what finds on his discord, and then reads what he is doing on the screen, not too much helpful information

lol

[dim sonnet]

i'm just happy to see videos about swarm

[hearty notch]

sebastain did one also, he called it the super secret project

https://www.youtube.com/@sedetweiler

hope to see Scott doing his magic on Swarm

very informative, and explains properly how things works,

btw, I think it will be super awesome if we can have a comfy nod that is reflecting how your UI is using refiner, I am sure there will be community magic on pushing the limits of the refiner upscale which still having some issues when trying to get a proper details on the upscaled images, unless you are planning a hi-res fix kind of an option

[dim sonnet]

the Refiner options includes a direct equivalent to high-res fix

and you can do it all with nodes, the main tab just auto-generates a workflow for you

[inner sand]

Sbastian Kemph did a much better video.

[untold merlin]

hmmm...im playing around with stableswarmui and loaded a custom workflow (e.g. sytans)...for some reason i does not recognize the refiner model node as it seems. i can configure the base model, base pass even refiner pass, but there is no section for refiner model?! did i miss some configuration or is that a known issue?

[candid hornet]

SO I am trying this again, I am on windows 10, I have downloaded and run install-windows.bat it created the StableSwarmUI folder but not the desktop file. If I tried to run launch-windows.bat I get a download a .NET SDK witch I already done. Can anyone help?

[bright fractal]

I dont have a desktop shortcut either but was able to run the launch.bat. just now looks like it is installing

[candid hornet]

Ok I reinstall a different version of .net and now I have a better result but I am stuck at Creating new admin session for ::1

[nova coyote]

Got this problem when install StableSwarm..
Anyone can help?

[candid hornet]

I am not an expert but it look like you don't have .NET instaled

[nova coyote]

NET framework?
If .NET, i already install it

[candid hornet]

As I said I am new here and you can see that I a still strugling to install my instance

[bright fractal]

what version of .net did you install?

[nova coyote]

No problem, mate. Thx for the attention

[nova coyote]

7.0.306. Newest i think

[candid hornet]

Ok so I was able to install swarm now I need to connect it to my confy installation.

I went to Server/Backends/ComfyUI Self-Starting and set the path to my main.py as request but I get an error

I reinstall swarm whit local comfy installation and I have the same error

[soft cliff]

Hi

[bright fractal]

Stupid question, should one have Python installed?

[hearty notch]

@dim sonnet getting this error all of a sudden after everything is working fine

[dim sonnet]

All parameters in a custom workflow should be recognized, at worst it just won't map them to the correct default param ids but that's fine

if it doesn't, then ... well, first, update and CTRL+F5 in case it was recently fixed already, then if still not working, gimme the exact workflow / screenshots / anything you got to help diagnose what went wrong

[dim sonnet]

that message indicates your browser has connected properly, so... check browser window?

[candid hornet]

I have the problem in the browser window.

[dim sonnet]

er... that looks like files are missing or in the wrong location?

[candid hornet]

If I click on the comfy tab, I have a connection error

[dim sonnet]

you don't need python installed for swarm, and if you use the default comfy backend installer, that comes with its own copy of python

[dim sonnet]

er... Aside from the error message, what's actually breaking? That error is generic and might mean nothing on its own

[dim sonnet]

what error?

[hearty notch]

oh thanks for reply, looks like closing the UI with one applied Comfy workflow on the gen tab and a difrent one on Comfy edit tab will error out, it was sorted by just reseting all tabs

[hearty notch]

ctrl+f5

[dim sonnet]

check the console window, should have an error message in it

[hearty notch]

seems to sort out many things

[candid hornet]

I have tried whit the default path dlbackend/ComfyUI/main.py 'I have both my personal installation and I have install compfy whit swarm, in both I have the same error

this is the message in the console

and as you can see I do have the main.py

16:53:49.379 [Warning] ComfyUI start script is '', which looks wrong - did you forget to append 'main.py' on the end?

[dim sonnet]

that error does not match the screenshot above 0.o

your screenshot shows you have a startscript, but that error says it's empty?

[hoary kelp]

It work for me, I need to start my ComfyUI firt and Later Start the Swarm, my problem is that I can get the Comfy parameter in the I try to move with the Use this Workflow in general tab

[untold merlin]

Thats the example workflow im using. And generating images using this workflow works. The only thing i dont understand is, why there isnt a section for the refiner model node like there is for the base model. i dont need it right now, as i can pick the refiner model in the comfy workflow directly, and apart from that i dont have any other refiner model i would switch to, but from my understanding their should be a mapping section for this node?!

[dim sonnet]

you're using a self-start backend there - as in, it starts itself, not you start it. You shouldn't be starting comfy, let swarm do it. If you want it to run it yourself first, use the API By URL version of the backend

[dim sonnet]

OH. Your "Base Model (Node 10)" got mapped as a secondary model, meaning your refiner got mapped as the main model

[hoary kelp]

Copy That, Thanks @dim sonnet

[dim sonnet]

you can workaround that pretty easily by just deleting the refiner model node and replacing it (ie so it gets listed with a higher ID and thus is understood to be the secondary)

[hoary kelp]

@dim sonnet I let SwarmUI start the Comfy, but I getting this error now, Thank you for your help in a Sunday!!

17:14:22.890 [Init] Launch web browser...
17:14:23.957 [Error] [WebAPI] Error handling API request '/API/ListBackends': Request input has unknown session id
17:14:23.975 [Info] Creating new admin session for ::1
17:14:27.302 [Info] Creating new admin session for ::1
17:14:51.253 [Init] Self-Start ComfyUI on port 7820 started.
17:15:41.647 [Error] Internal error processing T2I request: System.ArgumentNullException: Value cannot be null. (Parameter 'source')
at System.Linq.ThrowHelper.ThrowArgumentNullException(ExceptionArgument argument)
at System.Linq.Enumerable.Cast[TResult](IEnumerable source)
at StableSwarmUI.Builtin_StabilityAPIExtension.StabilityAPIBackend.Generate(T2IParamInput user_input) in C:\Media\AI\SwarmUI\StableSwarmUI\src\BuiltinExtensions\StabilityAPI\StabilityAPIBackend.cs:line 154
at StableSwarmUI.Text2Image.T2IEngine.CreateImageTask(T2IParamInput user_input, GenClaim claim, Action1 output, Action1 setError, Boolean isWS, Single backendTimeoutMin, Action`2 saveImages) in C:\Media\AI\SwarmUI\StableSwarmUI\src\Text2Image\T2IEngine.cs:line 126

I'm using the Stability API

with their SAPI safetensor

[hoary kelp]

Whats Up Boddy good to see you here!!!

[dim sonnet]

er... that seems to indicate the StabilityAPI did not return a valid image after generating

response["artifacts"] was null

[untold merlin]

Aye, yeah that worked. good to know that the id of the node is used to identify the refiner model node! Ty.

[candid hornet]

Weather or not it match, it is what I got.

[hoary kelp]

@dim sonnet Dang

[dim sonnet]

if you update and launch with --loglevel debug it will now output more data about StabilityAPI errors, which we can hopefully use to identify what went wrong there

[hoary kelp]

@dim sonnet Very good!! on it!!

[candid hornet]

I event restart every and got the same error

[dim sonnet]

oh btw, you can also make primitives and title them SwarmUI: Model and SwarmUI: RefinerModel and they will automatically use the matching builtin swarmui parameters

[dim sonnet]

Could you please post

• The full complete content of the console log after a restart
• CTRL+F5 the webpage, then take a screenshot of the backends list

[hearty notch]

@dim sonnet there is a bug when using this nod to feed all samplers it will always error out when in the generation tab

[candid hornet]

Full console log after a restart

[hearty notch]

need to go back to Comfy tab and play with it

otherwise the seed will not change ]

[dim sonnet]

there's no errors at all there

(other than nuget but ignore that)

[candid hornet]

Crt+F5 the webpage and this is the scren grab

[dim sonnet]

set seed to -1 in the UI (or click the 🎲 icon) and it will randomize

[hearty notch]

i did

[dim sonnet]

I need to set it to recognize that if you have randomize it means use -1

[hearty notch]

it's not randomzing

this is why I am sharing this,

it will just keeping the same seed

[dim sonnet]

In the logs it launched without error. Somethings going weird... do you maybe have two copies of swarm open or something?

Or... can you post your StableSwarmUI/Data/Backends.fds file?

[candid hornet]

I do have a automatic1111 open at the same time

[dim sonnet]

that shouldn't be the issue

(though might be wise to close that while debugging to reduce variables)

[dim sonnet]

can you post the workflow file?

[hearty notch]

and when I try to control from the gen tab it will error out like this: [Error] Internal error processing T2I request: System.Exception: ComfyUI errored at StableSwarmUI.Builtin_ComfyUIBackend.ComfyUIAPIAbstractBackend.AwaitJob(String workflow, CancellationToken interrupt) in D:\StableSwarmUI\src\BuiltinExtensions\ComfyUIBackend\ComfyUIAPIAbstractBackend.cs:line 126 at StableSwarmUI.Builtin_ComfyUIBackend.ComfyUIAPIAbstractBackend.Generate(T2IParamInput user_input) in D:\StableSwarmUI\src\BuiltinExtensions\ComfyUIBackend\ComfyUIAPIAbstractBackend.cs:line 209 at StableSwarmUI.Text2Image.T2IEngine.CreateImageTask(T2IParamInput user_input, GenClaim claim, Action1 output, Action1 setError, Boolean isWS, Single backendTimeoutMin, Action2 saveImages) in D:\StableSwarmUI\src\Text2Image\T2IEngine.cs:line 173`

[dim sonnet]

nothing in there looks off. I'm confused

[candid hornet]

ok I closed swarm and restart and now this is my console log

[dim sonnet]

Aha! You do have multiple copies of swarm open!

[candid hornet]

And automatic1111 is now closed

no

[dim sonnet]

yes, that's what that error is

there's another copy of swarm running when that was launched

[candid hornet]

1 sec

[dim sonnet]

heck

[candid hornet]

ok yes, I am now back on the original error

no error in the console but still the backend error

[hoary kelp]

@candid hornet Yes I'm getting that error too

[hearty notch]

[Error] ComfyUI error: { "error": { "type": "prompt_outputs_failed_validation", "message": "Prompt outputs failed validation", "details": "", "extra_info": {} }, "node_errors": { "54": { "errors": [ { "type": "value_smaller_than_min", "message": "Value -1 smaller than min of 0", "details": "seed", "extra_info": { "input_name": "seed", "input_config": [ "INT", { "default": 0, "min": 0, "max": 18446744073709551615 } ], "received_value": -1 } } ], "dependent_outputs": [ "9" ], "class_type": "KSampler" }, "63": { "errors": [ { "type": "value_smaller_than_min", "message": "Value -1 smaller than min of 0", "details": "noise_seed", "extra_info": { "input_name": "noise_seed", "input_config": [ "INT", { "default": 0, "min": 0, "max": 18446744073709551615 } ], "received_value": -1 } } ], "dependent_outputs": [ "9" ], "class_type": "KSamplerAdvanced" } } }

this is the other part of the seed issue

[dim sonnet]

fixed the mishandling of -1

[dim sonnet]

what error?

[hearty notch]

will check it

[dim sonnet]

wtf

[dim sonnet]

Can you launch swarm with --loglevel debug? and get the console output from that

open a command prompt in the folder and launch-windows.bat --loglevel debug

[hoary kelp]

@dim sonnet This error, Look like he fixed

[dim sonnet]

that logs doesn't have any error in it (other than the nuget one at the top which you can ignore)

[hoary kelp]

ok

[candid hornet]

ok do you need the console log or a file?

[dim sonnet]

console log

[dim sonnet]

it recognizes this now

[dim sonnet]

h- huh?

There's just... no output? wtf?

[candid hornet]

I would realy like to be able to help you on that

[dim sonnet]

omg wait i have a theory hol up

[dim sonnet]

okay my theory wasn't right, but I added more debugging to hopefully identify what's going wrong - can you run update.bat then launch with --loglevel debug again and post the new log?

[hoary kelp]

@dim sonnet This is the error with the stability AI using debug, I going to update to try again in case you fix something for @candid hornet

  "id": "1ce3018d9f487ec72863b2275ca154ce",
  "message": "can't decode JSON request body: malformed JSON",
  "name": "bad_request"
}
18:16:31.251 [Error] Internal error processing T2I request: System.ArgumentNullException: Value cannot be null. (Parameter 'source')
   at System.Linq.ThrowHelper.ThrowArgumentNullException(ExceptionArgument argument)
   at System.Linq.Enumerable.Cast[TResult](IEnumerable source)
   at StableSwarmUI.Builtin_StabilityAPIExtension.StabilityAPIBackend.Generate(T2IParamInput user_input) in C:\Media\AI\SwarmUI\StableSwarmUI\src\BuiltinExtensions\StabilityAPI\StabilityAPIBackend.cs:line 168
   at StableSwarmUI.Text2Image.T2IEngine.CreateImageTask(T2IParamInput user_input, GenClaim claim, Action`1 output, Action`1 setError, Boolean isWS, Single backendTimeoutMin, Action`2 saveImages) in C:\Media\AI\SwarmUI\StableSwarmUI\src\Text2Image\T2IEngine.cs:line 126
18:16:31.256 [Debug] Refused to generate image for local: Something went wrong while generating images.```

[dim sonnet]

... huh

[dim sonnet]

i added more debug, so update now and try again

if you get the error again it will debug output the raw input it sent to sapi, so if there's actually malformed json there it will be visible

(shouldn't be possible, that error makes no sense, but idk)

[hoary kelp]

Thanks @dim sonnet Let's Do it

[hoary kelp]

@dim sonnet This is the error with the new debug, but strage things is I made another test disabiling The "Custom ComfyUI Workflow" and he render was done, I also switch the model to a SD1.5, may is the SDXL 1 SAPI this is the error console:

[dim sonnet]

OH yeah you can't use custom comfy workflows with stabilityapi

that's what broke it

it'll now warn if you trigger sapi without any prompt

++ it will now also force comfy workflows to be routed exclusively to comfyui and give a clearer error if it can't do that

[hoary kelp]

@dim sonnet So what it's the right way to use ComfyUI with the API? or is not posible!

[dim sonnet]

StabilityAPI is limited only to basic generations rn. If you want custom comfy workflows, you have to use a comfy backend.

[hoary kelp]

@dim sonnet You mean switch ComfyUI by API URL instead ComfyUI self Starting? or working apart with ComfyUI standalone without the API?

[dim sonnet]

ComfyUI by API URL has nothing at all to do with stabilityapi

stabilityapi is dreamstudio

comfyui is comfyui

"API" is just a word that means an interface between applications

StabilityAPI is a way to interface with Stability's servers which are not running comfyui they're running a simpler software

[dim sonnet]

omg what path=F:\Programes\ComfyUI_portable\ComfyUI, yielded=F:\Programes\ComfyUI_portable\ComfyUI\launchtools\

it just

decided to add an extra folder there

and broke itself

why

WAIT

NO

OMG

[hoary kelp]

@dim sonnet thanks

[dim sonnet]

you have your SwarmUI installed in the "ComfyUI" folder

So that's the root issue there - you copied swarm into a comfyui folder

they need to either (A) be separated or (B) have comfyui as a subfolder of swarm, not the reverse

[candid hornet]

Ok

Actual no, my swarm is not in a comfy folder

[dim sonnet]

... oh my god wait this is a spaces-in-filepath issue isn't it

a

alrigh hol up

[candid hornet]

If you think it is, I will remove all space

[dim sonnet]

ideally spaces shouldn't explode like this, just ... batch is cursed.
I pushed an update that should fix it

[candid hornet]

Since I mostly work in windows I use them, but I know the history behind the problem.

Thanks

[candid hornet]

It is now working. Thanks

[dim sonnet]

:D yay

[toxic shale]

@dim sonnet

see how it goes from 2/6/2
to 2/7/2 and casually skips one image?

no error in comfy nor in swarmui

while it happens rarely, it's kinda annoying to have a huge grid and have just one image stand out for not being completed >_<

[dim sonnet]

it's not an error, it's a side effect of the async backend splitty thing - it doesn't strictly enforce order, so it can sometimes have one image end up delayed til later for a bit

if it's delayed significantly tho there's a problem

i did make some changes not too long ago to make it maintain order a bit more on single backends (it used to be chaotically random)

might be a way to improve it further on that

pushed a commit that might make it a bit better

[nova coyote]

Already installed 😁

[untold merlin]

Do we have a list of all built in parameters? Or more specific...in my custom workflow, there are separate prompts for positive (clip_g, clip_l). Using the title "SwarmUI: Prompt" maps one of them to the built in prompt parameter. Is there a way to map the prompt for _l?

[dim sonnet]

there isn't a built in L parameter - you can see the list of built in parameters, with, uh, your eyes?

lol idk

like

these things?

these are the parameters

anything that's in there by default, is a built-in parameter

[untold merlin]

yeah okay, i thought there might be more hidden ones 😄

[dim sonnet]

there technically are, but, well, if they're hidden you can't use them

(the one exception is Model is handled weird cause it's on the persistent bottom bar)

the other hidden ones are like - StabilityAPI Engine, is only visible when you have the stabilityapi backend enabled

or variation seed is only enabled if you use an auto webui backend

[untold merlin]

alright, understood, ty

[untold merlin]

hm...the mapping works for base prompt, but for some reason not for parameters like "Seed, Model, RefinerModel" or did i use wrong titles?

[dim sonnet]

oh, oops, i see why, fixed - update and it should work as intended now

[untold merlin]

The Seed parameter works now, but the Model nodes still dont

[dim sonnet]

oh, the title has to be on a primitve, not the LoadCheckpoint node itself

[untold merlin]

Aaah...okay yeah now that works, ty

[soft cliff]

@dim sonnet Sir hi, I could not install any of these custom nodes. I don't see any error in termal. Do you have any idea?https://github.com/daxthin/facedetailer https://github.com/Gourieff/comfyui-reactor-node https://github.com/imb101/ComfyUI-FaceSwap

[soft cliff]

When I treid again this one and I get this error

[bright fractal]

Sorry if this has been asked before but:

1. can I install StableSwarm on my potato PC with 4GB vram (8GB MEMORY), then..
2. Use StableSwarm Installer to install comfyui (even tho I dont have a Nvidea Card, just a normal Intel Graphics Card)
3. Use one or two colab sessions from here https://github.com/comfyanonymous/ComfyUI/blob/master/notebooks/comfyui_colab.ipynb

Oh and 4. I have my models on my Google Drive. I have the desktop app but the drive is set to be like a virtual drive so there is no physical folders or files. Will this be a problem if I point stable swarm to it for my models?

[dim sonnet]

Yes, though it might be easier to just do swarm directly on colab https://colab.research.google.com/github/Stability-AI/StableSwarmUI/blob/master/colab/colab-notebook.ipynb
if you do swarm in local and comfy remote that'll work fine, only limitation to note is right now swarm assumes you have the same models in local and remote (that's on the todolist to fix)

[dim sonnet]

:(

fine yeah okay i gotta make sure it knows to prioritize jobs that have been waiting for longer

[dim sonnet]

this seems like an issue with insightface which is presumably a dependency of one of these custom nodes? I'm not really sure what to do about that, probably would have to ask whoever made the node

[toxic shale]

happens when I run my 600+ image grid

I have to restart it every 150 images

[bright fractal]

Dude I was thinking the same thing! Rather just run it on colab! Ok I am playing with some ideas how I am going to do this. Your colab installs in the colab itself where comfy's install on to drive so everytime you start frsh everything is there again. I am thinking if yours could do the same i can just remove the default model folders on google drive and replace it with a shotcut folder where my models are saved on my drive. Hope I am making sense

Thanks for hopping on here and helping us pleabeans out like this!

[dim sonnet]

every bit of the colab's code is just right there on the one screen, I imagine you can just swap /content/ for whatever the colab path into drive is and you're golden?

should probably add a variable for that

[bright fractal]

Ok so the first cell seem to work, here are the changes

Install apt dependencies

!apt install dotnet-sdk-7.0 git

Install Clouldflared (not on apt)

!wget https://github.com/cloudflare/cloudflared/releases/latest/download/cloudflared-linux-amd64.deb
!dpkg -i cloudflared-linux-amd64.deb

Download StableSwarmUI

!git clone https://github.com/Stability-AI/StableSwarmUI

Download ComfyUI backend

%cd /content/drive/MyDrive/StableSwarmUI
!mkdir /content/drive/MyDrive/StableSwarmUI
%cd /content/drive/MyDrive/StableSwarmUI
!git clone https://github.com/comfyanonymous/ComfyUI
%cd /content/drive/MyDrive/ComfyUI

Setup ComfyUI

!pip install -r requirements.txt

but running the cloudflare i get

/content/drive/MyDrive/StableSwarmUI
bash: ./launch-linux.sh: No such file or directory

[hearty notch]

Hi @dim sonnet I forgot to ask you about the nVidia Drivers, for some reason people used to recommend using the old driver i think 531 as the new one is having some issues, at least with a1111 and vlad, do you reommend also using old drivers? or just up to date?

[bright fractal]

Ok so i got StableSwarm now on MyDrive! And it opens up, so thats good. I did the following under:

Download StableSwarmUI

%cd /content/drive/MyDrive
!git clone https://github.com/Stability-AI/StableSwarmUI

Giving me a server error, but will tackle that tomorrow

[hearty notch]

why you are using the manual setup??

the bat file setup is working fine?

[dim sonnet]

i've heard about that but i'm not sure on the details. If you already have a driver that works, just don't change I guess.

[dim sonnet]

he's mucking with the colab

[dim sonnet]

okay i'm actually looking over the code now and i'm very confused how that happens

because it does already prioritize jobs that have been waiting

[median forum]

@dim sonnet can i get help on installing swarm ui, whenever i run install-windows.bat it runs and creates the folder but doesnt create a shortcut and doesnt open any server. i run it again and it tells me that its already installed

[hearty notch]

Click launch windows!

Install for install

And Launch for running the server

[scenic oasis]

How come stability.ai gave up on making an open source discord bot repo?

[dim sonnet]

seems like a topic for the discord bot channels rather than the swarmui channel?

[hearty notch]

btw, after spending 45 min building the nice comfy workflow,

it's just not showing anymore

i can still see it in the gen tab

i can gen imgs with it

but nothing is showing in the comfy edit tab

and when I pull imgs inside the tab also not showing the workflow

@dim sonnet the comfy WF is saved in the images generated with Swarm? right?

[bright fractal]

lol , definetely mucking aboutOk so StableSwarm on my drive and but when i run cloudfare I get the below

16:38:38.323 [Init] Prepping options...
16:38:38.488 [Init] Loading models list...
16:38:38.492 [Init] Loading backends...
16:38:38.494 [Init] Loading backends from file...
16:38:38.507 [Init] Initializing backend #0 - ComfyUI Self-Starting...
16:38:38.511 [Init] Prepping API...
16:38:38.516 [Init] Prepping webserver...
16:38:38.517 [Error] Failed init of ComfyUI with script target '/content/StableSwarmUI/dlbackend/ComfyUI/main.py' because that file does not exist. Please verify your start script location.

I have change the paths under /content/drive/MyDrive/StableSwarmUI/colab/colab_backends.fds but no luck

Also getting more errors further down from cloudflare

ERR error="stream 125 canceled by remote with error code 0" cfRay=7f3934757610ebfe-SEA event=1 originService=http://localhost:7801
ERR Request failed error="stream 125 canceled by remote with error code 0" connIndex=0 dest= THE TUNNEL

[dim sonnet]

it's not atm, needs to be but atm it's not tracked

if you refresh the comfy tab should fix itself usually

and it retains a history of your last workflow

[hearty notch]

not working, just lost the time buildinmg the WF 😦

[ruby glacier]

ControlNet does not work with the acceleration 😦 ------> NotImplementedError: The operator 'aten::upsample_linear1d.out' is not currently implemented for the MPS device. If you want this op to be added in priority during the prototype phase of this feature, please comment on https://github.com/pytorch/pytorch/issues/77764. As a temporary fix, you can set the environment variable PYTORCH_ENABLE_MPS_FALLBACK=1 to use the CPU as a fallback for this op. WARNING: this will be slower than running natively on MPS.

[hearty notch]

what is this? for SwarmUI??

[ruby glacier]

ComfyUI(backend)...

[bright fractal]

sigh, give up... I cant find the file to change the path, the path Is created. Might try again after getting a degree in in computer programing

lol

[bright fractal]

@dim sonnet DUDE, just saw your update on the colab! Thank you , going to try it out now!

[dim sonnet]

heh, was waiting to make sure it perfectly worked before i pinged you, but as of now it does indeed work

[hearty notch]

@dim sonnet ERROR:aiohttp.server:Error handling request Traceback (most recent call last): File "D:\StableSwarmUI\dlbackend\comfy\python_embeded\Lib\site-packages\aiohttp\web_protocol.py", line 433, in _handle_request resp = await request_handler(request) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:\StableSwarmUI\dlbackend\comfy\python_embeded\Lib\site-packages\aiohttp\web_app.py", line 504, in _handle resp = await handler(request) ^^^^^^^^^^^^^^^^^^^^^^ File "D:\StableSwarmUI\dlbackend\comfy\python_embeded\Lib\site-packages\aiohttp\web_middlewares.py", line 117, in impl return await handler(request) ^^^^^^^^^^^^^^^^^^^^^^ File "D:\StableSwarmUI\dlbackend\comfy\ComfyUI\server.py", line 42, in cache_control response: web.Response = await handler(request) ^^^^^^^^^^^^^^^^^^^^^^ File "D:\StableSwarmUI\dlbackend\comfy\ComfyUI\server.py", line 428, in post_prompt json_data = await request.json() ^^^^^^^^^^^^^^^^^^^^ File "D:\StableSwarmUI\dlbackend\comfy\python_embeded\Lib\site-packages\aiohttp\web_request.py", line 671, in json return loads(body) ^^^^^^^^^^^ File "json\__init__.py", line 346, in loads File "json\decoder.py", line 337, in decode File "json\decoder.py", line 355, in raw_decode json.decoder.JSONDecodeError: Expecting value: line 1 column 1 (char 0)

after ther last update, nothing is working in the CUI editor

i tired everything,

not working anymore

[dim sonnet]

:'(

my code cleanup broke thing

i fixed now

[hearty notch]

let me update and check

[bright fractal]

Ok let try lets try agian, and again thank you for this ! I am getting the following error when I add the path /content/drive/MyDrive/ComfyUI/main.py

Final error (4) while initializing backend #0 - ComfyUI Self-Starting, giving up: System.ComponentModel.Win32Exception (13): An error occurred trying to start process '/content/drive/MyDrive/StableSwarmUI/./launchtools/generic-launcher.sh' with working directory '/content/drive/MyDrive/StableSwarmUI'. Permission denied
at System.Diagnostics.Process.ForkAndExecProcess(ProcessStartInfo startInfo, String resolvedFilename, String[] argv, String[] envp, String cwd, Boolean setCredentials, UInt32 userId, UInt32 groupId, UInt32[] groups, Int32& stdinFd, Int32& stdoutFd, Int32& stderrFd, Boolean usesTerminal, Boolean throwOnNoExec)
at System.Diagnostics.Process.StartCore(ProcessStartInfo startInfo)
at StableSwarmUI.Backends.NetworkBackendUtils.DoSelfStart(String startScript, String nameSimple, Int32 gpuId, String extraArgs, Action1 reviseStatus, Func2 initInternal, Action2 takeOutput, Func1 getStatus) in /content/drive/MyDrive/StableSwarmUI/src/Backends/NetworkBackendUtils.cs:line 152
at StableSwarmUI.Builtin_ComfyUIBackend.ComfyUISelfStartBackend.Init() in /content/drive/MyDrive/StableSwarmUI/src/BuiltinExtensions/ComfyUIBackend/ComfyUISelfStartBackend.cs:line 87

Maybe I must just install the backend folder agian like the previous colab ?

[dim sonnet]

yeah you started it up before i got it fully working

you're ~1 commit behind

i think if you just kill the colab runtime and open it again it should work

[bright fractal]

omg! hahaha okay wait here I go again ...wish me luck !

[bright fractal]

right , so I added the path
/content/drive/MyDrive/ComfyUI/main.py
to self start and get
20:52:32.598 [Init] Self-Start ComfyUI on port 7820 is loading...
20:52:43.486 [Info] Self-Start ComfyUI on port 7820 exited.
20:53:15.678 [Init] Initializing backend #0 - ComfyUI Self-Starting...
20:53:15.680 [Init] Self-Start ComfyUI on port 7821 is loading...
20:53:17.627 [Info] Self-Start ComfyUI on port 7821 exited.

[wheat sonnet]

Yes, there are a lot of pytorch operation that do not have MPS versions.
Just do as it says and export the environment variable before running SwarmUI, it will just drop back to the CPU while doing that operation using the GPU for all the rest.
So many people see that message and think it means all of torch will run on the CPU if they set it.

[bright fractal]

Ok so still mucking about with the colab. I get it working if. I do a total fresh reinstall of StableSwarmUI onto my Google MyDrive and I choose Local Install. I can even replace the ComfyUI and main.py by deleting the ComfyUI Folder in dlbckend on MyDrive and replacing it with a shortcut to my own ComfyUI folder with custom nodes and manager which is also on MyDrive.

Howver I am unable to replicate this again if I do a second google runtime?

/content/drive/MyDrive
fatal: destination path 'StableSwarmUI' already exists and is not an empty directory.

I get a bunch of ERR Request failed error="stream 197 canceled by remote with error code 0" connIndex=0 dest

[dim sonnet]

cc @ruby glacier i pushed an update that just tosses that env var to the mac launch script. It's a fallback so I don't think it'll have any negative side effects?

[wheat sonnet]

No it shouldn't have any bad effects and will stop some other breakages too, IIRC some of the schedulers need it too, UniPC for example.

[dim sonnet]

nice!

why is it not default on then lol

[wheat sonnet]

You'll need to ask the PyTorch folks that one 😆 I've got a feeling they really don't like supporting anything thats isn't CUDA

[acoustic frigate]

did comfyUI progress bars go away in the CLI? i'm able to generate but don't see progress anymore until it's done

[dim sonnet]

haven't made any changes in that regard recently

whether that shows is dependent on settings tho (loglevel, self-start vs apiurl backend, ...)

i need to get that visible in-UI tho

[soft cliff]

Sir hi, I can not install https://github.com/Gourieff/comfyui-reactor-node this custom comfyui. I believe issue occur cause insightface issue. I I've been contacted developer but I just wanna ask you too. Maybe you know the answer. @dim sonnet

[bright fractal]

@dim sonnet I got the colab working! Back and Forth with Claude the whole day but I did it! I used the following:

import os
os.chdir("/content/drive/MyDrive/StableSwarmUI")

with open("settings.yaml", "w") as f:
f.write("backendPaths:\n ComfyUI: /content/drive/MyDrive/StableSwarmUI/dlbackend/ComfyUI/main.py")

I just need help loading my models. seeing that I did a "Local" install I cant point to /content/drive/MyDrive/ can I?

[dim sonnet]

uh... none of that does anything

AI-generated nonsense code

you can change the model path in the server -> serverconfig tab

[untold merlin]

@dim sonnet tryin out the grid generator (love it so far). I have one question: Is there a way to to generate multiple images for a specified grid param? For example...i have 5 loras. i want a grid with each column (or row, doesnt matter) representing a specified lora. in each row (lets say 3), i want a random seed image. So i can compare 3 images against 5 different loras?

Right now i can achieve this by specifying the seed param providing 3 different seeds by hand, which works. But it would be more comfortable to set it to somthing like "use 3 random seeds", because then i can hit the Generate button again to get another grid with new images without altering the seed param every time by hand

btw..whenever i use the grid generator with the same grid parameter (for example 3 different samplers) multiple times, i always get the image previews of the very first generation even though seed is set to -1. On the generated html grid there are the correct new generated images though. seems to be a preview issue in the ui only

[inland condor]

Hey there, love this program, I am able to use 2 computers one with a 3090, and the other with a 4080 and its works really well and fast. The only issue I have is with faces, are you planning on adding "restore faces" in the future? or is there a way to add this to the workflow now?
Thanks again for the great work.

[bright fractal]

Yeah you right, I spoke to soon. Ah well

[toxic shale]

@dim sonnet lora support plz ❤️
explaining how to set up lora support via custom workflow and how to make comparison grids is haaaaard

#🔧｜finetune message
I keep wanting to write a guide on how to test the best epoch via stableswarm... but oh god that would be a long post, and I can't even post images inside the text on discord

[untold merlin]

lora support is already there?! or what exactly are u referring to? 🤔

oh in custom workflows ... nvm

[toxic shale]

yeah. and I wouldn't mind if that was all there was to it - but the thing is we can't just run a "no lora" + checkpoint 1 + checkpoint 2
which leads to complex setups to properly compare multiple checkpoints. especially once you start training clip as well

it's not that it's not working for me specifically - its that its hard to explain to a newbie who just managed to get lora training working. dont want to cause information overload

[untold merlin]

ye i see

[lunar canyon]

what is stable swarm?

[soft cliff]

It is like mixing auto1111 and comfyUi

[dim sonnet]

there is lora support. Have you still not updated?

is a lil tab at the bottom

[toxic shale]

weeeeeeeeeeeeeeee

time to update! ❤️

[dim sonnet]

you can do -1, -1, -1

[untold merlin]

haha nice...didnt think of that one 😂

[dim sonnet]

uhh i know there's a custom comfy node for it that you could use. Would be nice to be integrated and a nice easy checkbox tho
https://github.com/Stability-AI/StableSwarmUI/issues/44

[dim sonnet]

oo caching gets you if you overwrite directly in place

[untold merlin]

What would be the recommended alternative? Changing the grid output folder name on every new invocation?

because when i uncheck "overwrite existing files" then nothing happens, because no param has changed since last invocation

[dim sonnet]

yeah changing the folder

[worldly oxide]

How do I fix this problem?

23:41:38.827 [Warning] [BackendHandler] backend #0 failed to load model rundiffusionXL_beta.safetensors
23:41:38.828 [Warning] [BackendHandler] All backends failed to load the model! Cannot generate anything.
23:41:38.829 [Error] Backend handler loop error: System.InvalidOperationException: All available backends failed to load the model.
at StableSwarmUI.Backends.BackendHandler.LoadHighestPressureNow(List1 possible, List1 available, Action ReleasePressure, CancellationToken cancel) in E:\Stable_Diffusion\StableSwarmUI\src\Backends\BackendHandler.cs:line 751
at StableSwarmUI.Backends.BackendHandler.T2IBackendRequest.TryFind() in E:\Stable_Diffusion\StableSwarmUI\src\Backends\BackendHandler.cs:line 585
at StableSwarmUI.Backends.BackendHandler.RequestHandlingLoop() in E:\Stable_Diffusion\StableSwarmUI\src\Backends\BackendHandler.cs:line 674

[dim sonnet]

it looks like you're using backends that aren't linked to your models folder properly?

[hearty notch]

my new trained model, used with SSUI, I love the WF!

[hearty notch]

@dim sonnet hi .. some of gens now are failing, also, the metadate is now showing anymore like before when you select imgs from the sidebar, and after gen an imge and you click on it to enlarge it will show that placeholder img now, and you need to click on it from the side bar before you can review

you can see that only 7 out of 10 imgs made it, not sure why this is happening after the last few updates

[soft cliff]

Same

[dim sonnet]

oh, ye, metadata was broken the live previews

the ones not showing up, i'm not sure about

will do some testing later

[toxic shale]

is there an option to disable the placeholder, and just leave the last image up until the new one is generated?

[jade juniper]

Install goes to the point of opening the StableSWarmUI installer page and only loads a portion of it. Once the install page opens, it goes as far as is shown in the image and then stops. Never finishes loading. Unable to scroll to end. Any ideas? Tried different browsers. Currently using Vivaldi. Anitvirus off and all allowed in Vivaldi.

Time Elapsed 00:00:00.82
21:18:39.630 [Init] === StableSwarmUI v0.5.3.0 Starting ===
21:18:39.668 [Init] Prepping extension: StableSwarmUI.Builtin_StabilityAPIExtension.StabilityAPIExtension...
21:18:39.675 [Init] Prepping extension: StableSwarmUI.Builtin_ScorersExtension.ScorersExtension...
21:18:39.676 [Init] Prepping extension: StableSwarmUI.Builtin_GridGeneratorExtension.GridGeneratorExtension...
21:18:39.676 [Init] Prepping extension: StableSwarmUI.Builtin_DynamicThresholding.DynamicThresholdingExtension...
21:18:39.677 [Init] Prepping extension: StableSwarmUI.Builtin_ComfyUIBackend.ComfyUIBackendExtension...
21:18:39.677 [Init] Prepping extension: StableSwarmUI.Builtin_AutoWebUIExtension.AutoWebUIBackendExtension...
21:18:39.698 [Init] Parsing command line...
21:18:39.698 [Init] Loading settings file...
21:18:39.706 [Init] Re-saving settings file...
21:18:39.715 [Init] Applying command line settings...
21:18:39.727 [Init] Prepping options...
21:18:39.797 [Init] Loading models list...
21:18:39.799 [Init] Loading backends...
21:18:39.800 [Init] Loading backends from file...
21:18:39.801 [Init] Prepping API...
21:18:39.802 [Init] Prepping webserver...
21:18:39.910 [Init] Scan for web extensions...
21:18:39.912 [Init] Readying extensions for launch...
21:18:39.912 [Init] Launching server...
21:18:39.913 [Init] Starting webserver on
21:18:39.935 [Init] Program is running.
21:18:40.450 [Init] Launch web browser to install page...
21:18:40.930 [Info] Creating new admin session for ::1

[prisma iron]

You could always post a hyperlink extension to a google drive folder or something of the like on a cloud for various images in reference to the guide. Keep it short, succinct and to the point while providing a method of reaching source material for those who would read it. Not everything needs to be done right within Discord. if you can blend between third party platforms multiple channels are actually quite effective in getting information across efficiently.

The biggest thing with guides is usually reader retention which is keeping things as short as possible while trying to retain information size in relation to the project at hand. So condensing it through hyperlinks and other various platforms such as clouds and so on make it much easier for readers to digest because it doesn't come across as an essay but rather a short guide that's separated based upon which step you are on.

The size is similar but the pieces are broken apart separately for optimal retention from the reader's perspective.

The illusion of brevity alters the perception of the reader and therefore makes the guide more useful to the general user.

[prisma iron]

You can also use spoiler boxes to the same effect within Discord as dropdown text menus to condense the guide size and allow people to extend/retract the information necessary to them at any given time.

There are a lot of various tricks of the trade when it comes to writing and marketing to a particular focus group that assist in upping systemic retention of your target audience.

[spiral valley]

In it's current release does StableswarmUI support custom nodes in ComfyUI. Is .\dlbackend\comfy\ComfyUI\custom_nodes the place to put them and if I put Manager in there will is manage to install other custom nodes and models without breaking stuff?
I was kinda expecting to find a "custom_nodes" directory in the structure where Models and Output are.

[dim sonnet]

oop, i hate webtech sometimes, is fixed

if you update or just reinstall the install from scratch it should be good

[dim sonnet]

yep, should all just work

swarm doesn't currently have a custom comfy node management system, but anything you put in comfy itself inherently works

[spiral valley]

Cool. Having a bit of an eperiment day with SDXL and decided to go all in and look at moving away from Auto1111 as well.

[dim sonnet]

i fixed, behaves how you expect it to now

the placeholder & live preview show up on the batch view as intended rather than all up in yo face

[dim sonnet]

metadata fixed

i'm not sure about the gen failures

[hearty notch]

that was from a nod i was using, looks like it will trigger a fail randomly

[toxic shale]

thank you! ❤️ you're the best!

[prisma iron]

That's what I hear too

[inner vault]

@dim sonnet yo alex is the embedding option still not working ?

also when i Ctrl+arrow (up/down) it doesnt do anything like in A1111 and comfy where it raises the value of the word before it or the shaded sentence exmpl: (word:1.7)

also an idea for the trigger words what if every time you load a lora or a checkpoint or what ever ... if it have trigger words it shows them to you floating or in a list and you can click on what ever you need of them and it will add it to the prompt maybe also a drag and drop to positive or negative prompt

#1011228667659178055

[wispy marsh]

Okay, this is a mystery to me, @Alex (mcmonkey), maybe you can tell me how to fix it? If I use my CompyUi backend through the tab in Stable Swarm Gui it doesn't have all the custom nodes, that I have in the same CompyUi Running at the same time on the direct port. Also, If I save a workflow, or nested nodes or or template, in one of them it isn't available in the other. The version that is not on the GUI has more flexibility, but I would really prefer to just use the one on the GUI. I don't think it is the same problem as when the manager wasn't showing up because I have implemented that fix. Any help would be awesome!

[dim sonnet]

It's still just a listing. You can use em anyway by just typing iirc embedding:NAME_HERE into the prompt box, all it'll really do once clickable is automatically insert that (the only complexity is determining which prompt box to use, as some embeds are eg meant as negative embeds)

[dim sonnet]

oh yeah gotta add that

[dim sonnet]

already on the feature plan list in the readme

[dim sonnet]

It should be all the same, unless there was an error? Check CTRL+SHIFT+I web console for error messages

[wispy marsh]

That gave me a TON of errors!

logging.js:315 [mtb] - Aborting the loading of mtb's imageFeed in favor of pysssss.ImageFeed
console.<computed> @ logging.js:315
logging.js:315 [NestedNodeBuilder] Nested node setup
logging.js:315 [🐍 pysssss] Skipping LinkRenderMode as core extension found And this is the log from the other

[dim sonnet]

oo

hol up

[wispy marsh]

So evidently there is something different between the two. . . Perhaps there is a setting that I could change, there was also something from chrome that said something about blocking stuff, and it had a link but it wasn't helpful.

Sorry that I'm such a pain in the ass, I just like to use a bunch of custom nodes. . .

[dim sonnet]

@wispy marsh the majority of errors are from one single extension, so I just made a PR to it to fix em: https://github.com/pythongosssss/ComfyUI-Custom-Scripts/pull/39

for the rest, similar fix is needed

not particularly sure what some of those like 'mtb' even are

but you can show the links in that PR to the dev of any extension that has those errors and they should be able to fix it very quickly

[wispy marsh]

So it's basically the same problem that you found in the manager extention? Right

[dim sonnet]

yep

it's easily fixed, the hardest part for me to fix it myself is literally just cloning the extension and figuring out how to test it and make sure i don't accidentally muck anything up

[wispy marsh]

Ha Ha, that's why you have me. . .any way you are much too busy trying to get the GUI and all the "smarmy stuff," working perfectly. I'm hoping that someday soon this will be in beta and ultimately released.

MTB is a developer that makes amazing custom nodes, particularly for video. he and I have had a lot of conversations about other things. His biggest challenge has been getting some of his nodes to install properly on any any of the comfyui versions, like portable, that have an embedded environment, rather than a venu. I will let him know about this problem,

[inner vault]

you can make it kinda drag and drop on the positive or negative

[dim sonnet]

if he has any issues getting things to work, tell him to pop in here, comfy and i can help

[wispy marsh]

meanwhile I am modifying all these files myself, quick question: on some of the paths you prepend with "../../..," and others with "../../../.." will either work or is it a case by case thing?

[dim sonnet]

that's all specific to that one weird extension

most things should be ../../

it's only that one extension that had multiple sublayers like that

[sand eagle]

Hey, people. I've seen that this is some kind of common issue, but couldn't find a solution and people say to not worry, but meanwhile when I try to load a present it's broken. It still works - creating random images and stuff 🙂 Should I be worried, or have I done something that I shouldn't. Will appreciate any help. Thanks

[turbid veldt]

can stableswarm be used to create a model-hosting API service? I want to design image-generation pipelines in ComfyUI and then have those hosted on some GPU-servers and have a front-end (client-app) that multiple users can call into

or would someone recommend a good multi-user model-hosting solution? Horde-AI?

[wispy marsh]

I have sent the dev an email if you want to put in a Pull his address is: https://github.com/melMass/comfy_mtb

[dim sonnet]

that message looks like sde_gpu sampler being silly with rounding? Doesn't seem like something that should matter

if something's actually broken please specify what actually broke

[dim sonnet]

Yes it can, though bear in mind it's in alpha status right now so it's not perfect nor perfectly documented currently

[sand eagle]

(from what I can recall now) I can load a workflow, but it doesn't load fully - had this workflow from the guide which I've followed to install the SS UI and it's not showing the prompt window anymore, and there were no errors thrown in the cmd prompt which I can see why it do that, but will check it in few sec to see if there are any errors.
This is the workflow

After loading, Before loading it

[dim sonnet]

seems to load fine?

only weird part is refiner & base are out of order

[sand eagle]

no errors - but no prompt window 🙂 so how to generate image if there is no prompt idk 😄

[dim sonnet]

your workflow has 6 different prompts in it, so scroll down:

[sand eagle]

discard this

yeah

[dim sonnet]

or better, name the main prompt primitive SwarmUI: Prompt

which forces it to be the main prompt input

more compatible that way

[sand eagle]

don't think I need that many dials and gadgets, but hey someone made it, just maybe i'll have to move this prompt window in front, instead in the middle of everything, sorry for bothering you, and keep what you're doing

just tested with standard prompt which this workflow came and it throw this error - will reload everything and will se if it do it again
20:29:06.109 [Init] Self-Start ComfyUI on port 7820 started.
F:\Stable-Diffusion-Swarm\StableSwarmUI\dlbackend\comfy\python_embeded\Lib\site-packages\torch_utils.py:830: UserWarning: TypedStorage is deprecated. It will be removed in the future and UntypedStorage will be the only storage class. This should only matter to you if you are using storages directly. To access UntypedStorage directly, use tensor.untyped_storage() instead of tensor.storage()
return self.fget.get(instance, owner)()
100%|██████████████████████████████████████████████████████████████████████████████████| 25/25 [02:23<00:00, 5.75s/it]
20:45:45.378 [Error] Invalid/null/empty image data from ComfyUI server for 'ComfyUI_temp_ajiiz_00001_.png', under {
"images": [
{
"filename": "ComfyUI_temp_ajiiz_00001_.png",
"subfolder": "",
"type": "temp"
}
]
}

it doesn't save the image as it tries to save it with some weird name 🙂

[dim sonnet]

oh, uh, the preview nodes get mishandled a bit. I should probably set it up to silently remove them internally

Done

it handles preview/save intelligently now

[sand eagle]

did you update something? 🙂

@dim sonnet It still cannot save the image, for some reason, but at least I can see what it generated

[dim sonnet]

yes, when i said "Done" above I pushed a new commit to fix that, so if you double-click update.bat or you open a command line and run git pull that'll load the new changes, then restart swarm -- you may also need to CTRL+F5 to ensure your browser doesn't cache it

[sand eagle]

Invalid data: ComfyUI errored: {
"error": {
"type": "prompt_outputs_failed_validation",
"message": "Prompt outputs failed validation",
"details": "",
"extra_info": {}
},
"node_errors": {
"192": {
"errors": [
{
"type": "value_not_in_list",
"message": "Value not in list",
"details": "model_name: '4x_NMKD-Siax_200k.pth' not in []",
"extra_info": {
"input_name": "model_name",
"input_config": [
[]
],
"received_value": "4x_NMKD-Siax_200k.pth"
}
}
],
"dependent_outputs": [
"190",
"9"
],
"class_type": "UpscaleModelLoader"
},
"11": {
"errors": [
{
"type": "value_not_in_list",

[dim sonnet]

"type": "value_not_in_list", "message": "Value not in list", "details": "model_name: '4x_NMKD-Siax_200k.pth' not in []",
you have an upscaler model in the workflow, but you don't have any upscaler models in your comfy

[sand eagle]

oh so I need to get one?

strange it's off

[dim sonnet]

off in the sidebar, for a custom workflow, means use the workflow's value

mostly just a protection from breaking workflow stuff

you'd have to remove it from the workflow if you don't want it

(or, like, not use the workflow lol)

[sand eagle]

the last part I think works better, as without this garbage workflow everything works nice

[eternal aurora]

What has Stable Swarm got over ComfyUI at all?

[hearty notch]

@dim sonnet hi, there is an odd bug with presets, nothing is working there anymore, if i click on edit or create new, nothing is working

[wispy marsh]

Stable Swarm is a front end for ComfyUi, The GUI is there for people that are afraid to use CompfyUi. But, the HUGE advantage is how it leverages your GPU's and memory. It takes advantage of the resources that are available. So, to answer your question directly: It is faster!

[dim sonnet]

Also has a lot of power tools (like the grid generator) and convenience tools (like presets), so in addition to literally generating faster as AIMusicExperiment said, it enables you to experiment and adjust things faster

and it's just got a really really nice UI for a lot of things

I'm biased but imo but the models/loras/images/etc. browser UI is the best of any UI out there

[dim sonnet]

... I hate javascript sometimes. Tiny typo went undetected and was the cause of everything. Fixed.

[split swan]

does comfyui allow you to use different models for different blocks? same with loras

so you might have an anatomy based model, a concept one, a color one, a prompt recognizing one

and a scenic one that decides what goes where etc

or is it something that can just be added in stableswarm itself? @dim sonnet

[dim sonnet]

that'd be complicated to do well - you'd have to either generate parts separately and then composite them together (I've seen it done before), or like... have each model take turns swapping out for each step with different areas masked off as their region

It's theoretically possible either way, but I'm not sure offhand how to do it in current UIs

(if you can find a comfy workflow that does it you can just slap it into swarm of course)

[split swan]

well you can judge what different models does when and if consistent, you know what steps it kicks in

[dim sonnet]

it ties into the idea of having an LLM control generation - part of that idea is the LLM should select what models to use and assign content to regions

[split swan]

by getting an image of each step and running them through that extension that can check what it modelled when

you can just use a rating extension and train it to rate concept, design, anatomy etc

[dim sonnet]

custom workflows can be saved, loaded, and gridded now

[candid hornet]

I was wondering, I have connect swarm to my comfy by the self-Starting type, why don't I have access to my modules?

I was hoping that I would simply have access the my comfy by the interface if not what is the point?

[dim sonnet]

you should have access to everything, barring some extensions have compatibility issues (that are easily fixed just need to identify which ones + report to them)

if things aren't working, gimme specific details so we can fix it

[candid hornet]

Starting with extension manager, It is not in my swarm.

@dim sonnet

[dim sonnet]

do you mean https://github.com/ltdrdata/ComfyUI-Manager? If so, that was fixed 2 weeks ago: https://github.com/ltdrdata/ComfyUI-Manager/pull/43 so just update everything

[candid hornet]

I have and still not there. I have run update-windows.bat then launch-windows.bat

[dim sonnet]

update comfy manager not swarm

swarm doesn't update your comfy extensions for you

[candid hornet]

I updated Comfy but I don't know how to update my extention?

[dim sonnet]

you'd have to go into the folder for the extension and git pull

[dim sonnet]

controlnet input option owo

[candid hornet]

Thanks

[narrow blade]

Does StableSwarm support dual GPU

[dim sonnet]

yes

[narrow blade]

That's great

[narrow blade]

And it's compatible without an NVlink?

[dim sonnet]

yes

doesn't even have to be in the same computer

[narrow blade]

Wdym?

How can we connect 2 gpus from different sources

[dim sonnet]

'internet

lol

if you have a backend (eg comfyui) running on the other computer, you can connect to that as a backend for swarm, and it will automatically split half your image generations to that other backend

[wispy marsh]

"AIMusicExperiment sounds so formal. . . Ha Ha, feel free to call me Ava.

[wispy marsh]

That's the whole point!

[wispy marsh]

You just piqued my interest, can you have two different Comfyui installs as backends? And can you API them across a LAN?

[dim sonnet]

yes and yes

[wispy marsh]

You are doing the lord's work sir! I hereby nominate you for an academy award!

[wispy marsh]

Perhaps you can solve my "problem de jour," It may well be related to that "relative path" thing that you identified before, but it is different. This is about "nested nodes." If I nest nodes on the GUI interface, they don't show up on the Local interface and the other way round as well. It is kinda boggling my mind. . .Any thoughts? Also unrelated, but if you have any relationship with the popular developers, would you ask them to test there stuff on "portable compyui," (like you use for swarm) to make sure that there nodes will install. So many assume that there is a VENU. Embedded Python is both a blessing and a curse.

[dim sonnet]

er... i'm not sure what you mean

[wispy marsh]

which part?

[toxic shale]

freshly updated XD it still dies after about 150 images

[toxic shale]

also thank you for making all this ❤️ I don't think I can ever go back to default comfy or auto1111 after having this XD

[spiral fossil]

hi everyone, can i have my "regular" comfyUI live inside my swarm?

i hate the idea of having 2 on the same computer

[inner sand]

Yes! One instance only is required.

[spiral fossil]

that's cool 🙂

so i'm just moving my regular one to comfy sub folder in swarm, right?

not that i was able to generate one image with frontend yet- it says i need to pick a model for some reason

[inner sand]

No need to do that. Look in the server tab and you will see how to point StableSwarm to your Comfy back-end.

[dim sonnet]

... hwhy!

why it do dat

it shouldn't even be possible for it to queue more than twice as many as you can handle at once

and it shouldn't be possible for it to get too far out of order

can you give more info about what's going on

is it, like, a model change that's failing to happen maybe? (ie from a grid with multiple models)

[dim sonnet]

yes, pick a model in the Models tab at the bottom of that screenshot there

and yeah the default installer can install a comfy backend inside of swarm for you

or you can setup your own via the Self-Start Backend option

[dim sonnet]

• I pushed a commit that will log more clearly if there's a model related issue

[dim sonnet]

anyway so you can slap images to the prompt box to do ReVision now

[split swan]

is it even possible to take and img into img2img and mask out say a few buildings in it that you want, then add a second image where the mask from the other one is visible and you can pick what to add where as well, then it sort of runs those 2 together without a prompt or any strong indicators. Before it runs it through the prompt and loras? So maybe take post apocalyptic buildings from one, a long bridge freeway from another, and an ocean underneath from a third

or just running img combos, not sure how MJ did that

[static fulcrum]

im very confused on the intalltion process

i gitclones the repo

and then i ran the launch and it didnt work..got a few errors

i tried the Readme installation method

but that was the exact same thign

[dim sonnet]

well... first of all, yeah, follow the readme lol

second, if you get errors while following how the readme says to install, show me what errors you got

[static fulcrum]

yeah i did, same issue eitherway

[dim sonnet]

that looks like you didn't run the install.bat?

[static fulcrum]

i did though

install_windows.bat

[dim sonnet]

the first thing install.bat does is download+install .NET 7

if for some reason it's failing? you can also do it manually: https://dotnet.microsoft.com/en-us/download/dotnet/7.0

or https://dotnet.microsoft.com/en-us/download/dotnet/thank-you/sdk-7.0.400-windows-x64-installer

[static fulcrum]

well i installed it manually

but still

[dim sonnet]

can you, uh, delete all swarmui files and redownload the install.bat

idek how you got project file does not exist but that sounds like something got mixed up while you're trying to make it all work

so, just, nuke and restart

you have the correct .NET now so it should be good from there

[static fulcrum]

i did that though, unless it installed files elsewhere... but i did a couple of times now

and its the exact samething

[dim sonnet]

... wtf? Can you take a screenshot of the stableswarmui folder

[dim sonnet]

if you double-click the first of the two launch-windows files, does that work? Or same error?

[static fulcrum]

it gives me the first error screenshot i sent

the shortcut on the desktop gives me the second error screenshot

[dim sonnet]

... wh... huh??

[static fulcrum]

i am in the same huh moment also

[dim sonnet]

that makes even less sense, you should not get the wrong .net versionerror again

[static fulcrum]

Sooo?

[dim sonnet]

Okay, new plan:

• Click into the address bar at the top of the folder, so that it gives you a text editor in the address
• Type cmd and hit enter
• this will open a manual command line
• in this command line, type launch-windows.bat and hit enter
• if it doesn't just work, screenshot the window

that removes basically every possible variable and guarantees all relevant info should be visible. And gives us a way to debug more from there

[static fulcrum]

isnt this just the same error as the first screenshot?

[dim sonnet]

That is. Wtf.

is something going wrong in the .NET 7 installer? 0.0

can you type: dotnet --version in the command line and screenshot what that gives?

and where dotnet

[static fulcrum]

im starting to believe it just dont want to work because it doesnt like me directly

[dim sonnet]

... are you sure you installed https://dotnet.microsoft.com/en-us/download/dotnet/thank-you/sdk-7.0.400-windows-x64-installer

exactly that file too

[static fulcrum]

yeah but ill re-do it

Ooo Ooo now its working

its doing something

Ooo it worked the StableSwarUi installer popped up i can design it

Thank you MrMCMonkey

[dim sonnet]

:D yay

[static fulcrum]

can i install comfy ui manager extensions to this?

id like to kinda design my own ui for my comfy workflow technically

[dim sonnet]

yes

even if you use the default comfyui backend installer, it's still a full comfyui instance, and can be used and modded in all the same ways

you can also just link to your pre-existing install of comfy if you have one already set up how you want

[static fulcrum]

how can i do that? or was i supposed to do that during the installation process?

[dim sonnet]

just click Server at the top of the UI, that will show your backends list

[static fulcrum]

ahhh i see

[dim sonnet]

you can edit or delete the backend that's there and/or add new ones

[static fulcrum]

cool

[dim sonnet]

you'd want either a ComfyUI self-start pointing to your comfy path, or a ComfyUI API By URL if you want to launch it yourself

if you do that make sure to go to Server Configuration and edit the ModelRoot path to match where your comfy models are

[static fulcrum]

the stableswarm desktop shortcut though still doesnt work

[dim sonnet]

error message or just silently fails?

if error, which error

[static fulcrum]

the same error

[dim sonnet]

which

[static fulcrum]

this one

[dim sonnet]

wait omg

i bet i know why

cause you're in an E: drive and windows is dumb

okay doubleclick update-windows, let it update, then try the desktop shortcut again

I think I fixed it

[static fulcrum]

yessir that did fix it thanks

[dim sonnet]

:D woo

[static fulcrum]

i have a node that works in comfy..but isnt working in the stableswarm ui

[static fulcrum]

@cunning holly is subgraph support coming soon to the master branch?

@dim sonnet im not sure how to make my comfy workflow work on the Ui?

its not using the workflow techincally for the generate tab

[regal imp]

In StableSwarm, when running ComfyUI, using the Load Image node results in a "400 - Bad Request" when trying to load any image other than example.png--even after complete re-install. No errors or diagnostic info in cmd. Running locally on Win 10 Pro with RTX 3090 FE. Anyone?

[balmy orchid]

Let's say I've done a 1 dimensional grid to evaluate several samplers. Is there a way we could use the arrow keys (or any key at all) to jump between images when in "fullscreen" viewer mode? That way I don't have to take my grid images into another app when I'm A/B deciding between two samplers or something. Having to close, and then click the next image to get fullscreen again breaks my flow.

[neon atlas]

using this workflow in generate tab button not sending workflow to generate tab after last update. Anybody please?

[static fulcrum]

Same here^

[spiral fossil]

when using swarm from lan, does it use the model and gpu on the server side or the client side?

[dim sonnet]

er... make sure you're fully up to date and hit CTRL+F5, that was a bug last week that got fixed

[dim sonnet]

hit CTRL+SHIFT+I and look for an error message in console?

[dim sonnet]

it uses wherever the backends are installed. It never uses your client gpu, unless you also install a backend there

[dim sonnet]

is it bugging like hespinop? If so, same thing - CTRL+SHIFT+I and look for an error message

if not that, then you might just not know you need to hit this button:

[dim sonnet]

that, uh, is already a thing? You should be able to hit arrow keys to move around in the fullscreen view of the grid viewer

[neon atlas]

happens with all workflows I have

[static fulcrum]

Nothing would happen, I hit that button it takes me straight me straight to the generate tab

But it doesn’t use the workflow

[neon atlas]

same here

[static fulcrum]

Plus it won’t let me technically save a workflow

[dim sonnet]

oo okay i might know why

okay, update (click update.bat or run git pull) it should work now @neon atlas @static fulcrum

there was a bug specifically related to new handling of ReVision parsing

[neon atlas]

you got it working! testing now

thanks

got this

16:34:11.117 [Warning] [BackendHandler] backend #2 failed to load model comunity_modle/dreamsmbbxlmixV10SDXL1_v10.safetensors
16:34:11.118 [Warning] [BackendHandler] All backends failed to load the model! Cannot generate anything.

[dim sonnet]

do you maybe not have the model folders linked properly?

if you're using a self-start comfy backend, you want the model folder in your Server->ServerConfiguration to be the main folder, and it'll automatically link comfy to it

[static fulcrum]

Yeah the model kept loading for a long time and then it didn’t generate anything

I have my models folders linked correctly cuz I can technically generate in the comfy workflow page

[neon atlas]

cannot get swarmui to work, but confyui works fine inside in the editor.

19:16:22.336 [Warning] [BackendHandler] backend #2 failed to load model sdXL_v10RefinerVAEFix.safetensors
19:16:22.337 [Warning] [BackendHandler] All backends failed to load the model! Cannot generate anything.
19:16:22.345 [Error] Backend handler loop error: System.InvalidOperationException: All available backends failed to load the model.
at StableSwarmUI.Backends.BackendHandler.LoadHighestPressureNow(List1 possible, List1 available, Action ReleasePressure, CancellationToken cancel) in F:\sworm\StableSwarmUI\src\Backends\BackendHandler.cs:line 734
at StableSwarmUI.Backends.BackendHandler.T2IBackendRequest.TryFind() in F:\sworm\StableSwarmUI\src\Backends\BackendHandler.cs:line 595
at StableSwarmUI.Backends.BackendHandler.RequestHandlingLoop() in F:\sworm\StableSwarmUI\src\Backends\BackendHandler.cs:line 694

now its the refiner sdXL_v10RefinerVAEFix.safetensors

[dim sonnet]

can you screenshot it working in the comfyui page, and then the error with the same model on the generate page?

[dim sonnet]

that's really hard to read the model names

but it does look like a proper match

why are backends refusing to load it then

you've only got the one, so no issues there

uh

it doesn't have any fail reason

oh i don't log error reasons properly for that oops

[dim sonnet]

okay if you update it will now actually log the reason for why it failed

[neon atlas]

thanks