🐝｜swarm-ui

[signal forge]

it worked, I just generated my first image

I let swarm install comfy as normal. cd comfyUI, rm -rf venv, pip install -r requirements.txt; then re-install the rocm pytorch. no issues

because I'm amd on linux, it doesnt make sense for me to have all these different container technologies within docker, like conda env and python venv. it becomes a mess to manage all those layers

[eternal aurora]

I asked at GitHub about the failure of tensorrt to build an sdxl engine which works ... but no reply!

[verbal lantern]

is stable sswarm faster then comfy?

[lucid bloom]

it is comfy in the back end

[cold flame]

they're the same. swarm is just a more user friendly frontend for comfy. it's still running comfy behind the scenes for the image generation

[nimble belfry]

There is a new dedicated Discord for SwarmUI #🐝｜swarm-ui message

[dim sonnet]

https://www.reddit.com/r/StableDiffusion/comments/1dlivre/swarmui_is_now_independent_from_stability/

[verbal lantern]

damn so i can do everything with swarm which is possible in comfy right?

[dim sonnet]

yes

[austere cypress]

Just migrated. Going to give it go. Thanks @dim sonnet

[thick flicker]

im having trouble with the img to img workflow in SwarmUI when generating my custom Pixart Sigma Img to Img workflow over to Swarm from ComfyUi

The regular Text to Img workflow for Pixart works perfect in swarm but for some reason INT IMG selection doesnt work

and the Load img node doesnt give me a option to load an img

it just shows up as a text selection

[dim sonnet]

swap it for a SwarmLoadImageB64 (or a SwarmInputImage if you want a clean paramater-managed workflow)

there's a special split thingo where LoadImage lets you specify filenames and SwarmLoadImageB64 lets you direct upload images to swarm ui

(lil messy ngl, needs work)

[main grotto]

all results i've had with swarm have a washed out colorscheme to them. no matter what settings or model i've tried. with or without image input. it still wont' give me a dark look

they tend to have darker tones in the previews until the last step

[eternal aurora]

I asked over on REDDIT: can I just change the name of the StableSwarmUI folder to SwarmUI - and then git pull?

[drifting helm]

nope, the .git folder still thinks its the old repo

[dim sonnet]

just follow migration guide https://github.com/mcmonkeyprojects/SwarmUI/discussions/2

[dim sonnet]

some models have better color range (eg CosXL or SD3 are good, vs SDXL and SD1 are limited), also boosting CFG Scale can increase color range (just be careful cause it does other things too). For older models you can also find Offset Noise LoRAs which will improve the color range on those models

[main grotto]

using the same models in easy diffusion and invokeai seem to be better for it using the same settings. i tried loras too. but never seen offset noise ones sepecifically i'll have to look into that. something about swarm makes me wanna use it. but i keep ending up back at the now getting outdated easy diffusion.(rabbit hole plugin probably being the biggest reason i'm guessing)

[potent panther]

They previews aren’t decoded with the VAE

They don’t have the same color range

[eternal aurora]

I've migrated but cannot generate

Can I Symlink the Models Folder? I'm not sure what the status of the !OfficialStableDiffusion and OfficialStableDiffusion Folders is? Must I re-create these folders in the Symlinked Models Folder?

Good work, much appreciated what with all the upheaval! 😄

[dim sonnet]

• you have not fully migrated, that is still Stable version, be sure to run an update after the migrate step
• you do have to select a model to generate lol
• you can set model folder path in Server Config (just be sure to restart swarm after), no need for symlinks
• if you have a folder with a ! prefix (?) that's... some form of weirdness on your PC not related to swarm, possibly some external program/antivirus/idk messing with files
• do note Swarm has its own dedicated discord now as linked in the pinned post here and in the swarm readme

[eternal aurora]

I Symlinked the dlbackend/comfy/comfyui/custom_nodes to my main comfyui/custom_nodes folder - PiXart-Sigma now works perfectly!

[verbal junco]

what is clip skip

in swarm

[dim sonnet]

type clip into the parameters filter (top left) and check displayed advanced

[pliant junco]

Does anyone have any tutorials or resources they recommend for sd3 w/ swarm ui?

[dim sonnet]

if you join the swarm discord (pinned here), there's a channel with a list of youtube videos about swarm. Also there's this reddit post https://www.reddit.com/r/StableDiffusion/comments/1de65iz/how_to_run_sd3medium_locally_right_now/

[lofty roost]

Sorry to jump in several days after your post, first big thanks for your work, and if you don't mind my asking, will the new SwarmUI, as you said it is supported by Comfy Org, also aim to become a multimodal interface like Comfy's plans for the future ?

[dim sonnet]

yeee

[lofty roost]

Wonderful news, thanks Alex - now the simple minds like me have hope, can both download Comfy workflows and keep the click-this check-that 🥳

[verbal junco]

how can one remove the masked area?

[open cairn]

does swarm support reading the metadata from automatic1111 produced images?

[dim sonnet]

max out creativity, optionally bump resettonorm too, and specify in prompt what you want to fill the area with

[dim sonnet]

yes, drag an image to the center area of swarm and it should parse it

[open cairn]

can you use upscaler models such as 4xultrasharp in swarm? if so how?

[dim sonnet]

yes, under the Refine/Upscale parameter group

put the model in (Swarm)/models/upscale_models and it will show up in the upscale method list

[urban maple]

how do i specialize the resolution for swarm ui?

it only says 1x or 2x which i hate

nvm exporting to comfy then importing back to swarm fixed it LMAO

[verbal junco]

In a1111 and foocus there's a tab for styles

That has gothic dark, cinematic etc

Is there a way to get those styles on swarm?

[urban maple]

yes i was talking about hi rez

they have like 8 styles

im using a cyber one

ok theres 7

[verbal junco]

I see I see

[urban maple]

Modern Dark, Modern White, Dark Dreams, Gravity Blue, Cyber Swarm, Punked, Eyesear White

[verbal junco]

Could I just import all the ones

I have?

[urban maple]

probably

im looking at them rn and it seems like it'd actually be easy to just make ur own

https://github.com/Stability-AI/StableSwarmUI/blob/master/src/wwwroot/css/themes/cyber_swarm.css

can just mess around with the numbers and color codes before u get something you like

its all .css files

[potent panther]

That's the biggest miscommunication I have ever set my eyes upon this year

[potent panther]

Yes they are called presets here and they use a somewhat standard format if you want to search for some

[potent panther]

He was asking about presets on Fooocus not the ui theme xD

[dim sonnet]

what

[dim sonnet]

that's the legacy repo, not the modern one https://github.com/mcmonkeyprojects/SwarmUI/discussions/2

[drifting helm]

btw @dim sonnet, the pinned has the old repo

[dim sonnet]

generally yes you can just import styles from other UIs to the presets tab

[dim sonnet]

i literally can't edit this message :'(

non-staff aren't allowed to linkdump

and editing a prior linkdump triggers automod

[drifting helm]

oh

[dim sonnet]

maybe if i deleted the discord link it'd work but the discord link is important

[drifting helm]

ask a staff to post the linkdump

[urban maple]

Ohhhh xD

He meant art styles I thought he meant ui themes...

[potent panther]

lol

[verbal junco]

how to go about

doing so?

ok nvm

its pretty easy

[verbal junco]

https://github.com/Bing-su/dddetailer

how to go about using this in swarm?

also why is every guide about a1111 when the SD discord doesnt even have a a1111 channel anymore lol

[silver spade]

anyone here knows how to use LoRA Section Confinement (

[signal forge]

Don't underestimate how many people still use auto, it's still the majority. You can install anything that works in comfy, I'm not aware of any limitations in that regard

[dim sonnet]

from the readme that looks about the same as https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Features/Prompt Syntax.md#automatic-segmentation-and-refining

[dim sonnet]

?

[silver spade]

sorry for the ignorance. but I am not very certain about how to use it.

[silver spade]

when I try to define the coordinates in the advanced part of the interface, it gives error. I don|t think I am supposed to use text there. what I wanted is to define one region to use a lora and another region to use other.

[silver spade]

Lora Section Confinement? how do I set the coordinates there?

[verbal junco]

when I use the 2x upscale

it adds extra fingers

to the image

[versed copper]

much less of that if you set the refiner strength/creativity lower - depending on what you use

[verbal junco]

control percentage?

and refiner upscale?

[potent panther]

that isn't Swarm's fault

try tweaking your refiner settings

avoid latent upscale

[dim sonnet]

the "upscale 2x" button is laziest bare minimum method possible, don't use it if you want good results lol

i intentionally hit it under the "More" menu in a recent dev commit

[verbal junco]

I see I see

I mean I'm assuming I should mess around with refiner settings then

[potent panther]

I thought you had removed it already 😂

[silver spade]

can we use IP adapter in swarm ui?

[dim sonnet]

yes, just drag an image to the prompt box

[silver spade]

You are sh*ting me... Now I have to go back home to try this!

[silver spade]

Mind blowing!!!

[thick flicker]

Tried it, worked flawlessly good job @dim sonnet

[light badge]

how can i do img2img ???

do i change the comfyui workflow

would it work ??

[dim sonnet]

Init Image parameter group, or drag an image to the center and click Edit Image

[verbal junco]

HiRes: 4xNMKD-Siax_200k with 15 Steps and 0.3 Denoise + 1.5 Upscale

so this in swarm is Refiner Steps, denoise is refiner control %, refiner upscale

[dim sonnet]

something like that yeah

depending on who wrote it that could also be 50 steps and 0.3 control%

(ie is it 0.3 denoise of 50 steps yields 15 final steps, or is it 0.3 denoise of 15 steps flat no nonsense)

[dim sonnet]

delayed reply but I finally realize what you mean, you dug into advanced/internal stuff and found the internal param by that name -- that's for <segment:> usage internally, not meant to be used directly, and doesn't support <region: currently, only <segment:

loras are also confined by <object: usage (though note that's handled separately internally not by the advanced internal param you found)

[dim sonnet]

• New Swarm Release version 0.9.1-Beta
  https://new.reddit.com/r/StableDiffusion/comments/1dy9rzr/swarmui_091beta_release/
  https://github.com/mcmonkeyprojects/SwarmUI/releases/tag/0.9.1-Beta

[silver spade]

anyone has any idea why some models end up producing images like this? matte and looking with wholes like it is eaten by some sort of insect?

[cold flame]

the person who uploaded the model most likely pruned the VAE. load the proper vae for the model type separately and you should get normal gens. if you need a link to download the correct VAE just let me know which SD version the model is and I can grab the link for you.

[silver spade]

that is fantastic, I loaded a VAE and worked like a charm, thanks

[cold flame]

Coke

[chilly bobcat]

any guides for getting started with making videos in swarmui? I am trying to make some videos to go along with my D&D campaign.

[verbal junco]

tried using faceid portrait unnorm

got this error

[dim sonnet]

oh, that's brand new in ipadapter -- i pushed a commit that adds support

[verbal junco]

ok

[silver spade]

hey Alex, would you guys accept requests for features? I have two: 1- to allow to export a collection as pair image + txt file so we can use good results for training loras and 2- integration with ollama and gpt for smart prompting.

[dim sonnet]

feature reqs go to https://github.com/mcmonkeyprojects/SwarmUI/issues

LLM prompt assistance stuff is planned, not yet built

[verbal junco]

damn

is there some alternate that doesnt make me download a1111

[dim sonnet]

what?

[verbal junco]

for blip

its an extension for a1111

right?

[open cairn]

Is it possible to acheive something similar to adetailer in swarmui?

[dim sonnet]

https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Features/Prompt Syntax.md#automatic-segmentation-and-refining

[open cairn]

do BREAK and AND work in swarmui, or do you need to use alternatives?

[cold dirge]

is their a way to install swarmui in to existing comfy location ?

[cold dirge]

nvm solved ti xD

[dim sonnet]

"BREAK" is <break> in Swarm syntax, "AND" uh... offhand i forget what that actually does

[open cairn]

Will the webui style syntax work at all in swarm, or do you always have to substitute the swamui way of doing things?

[dim sonnet]

some syntaxes are supported, others aren't - eg "BREAK" does not work written that way, you do need <break>

(some word:1.5) prompt weight syntax works fine (albeit weight values are interpretted a bit different since auto does renormalization and comfy/swarm don't)

[x:y:0.5] timestep syntax works fine

[downweight] doesn't, use (some word:0.9) instead

generally all syntax other than prompt weight in swarm you can just type a < and it'll pop up hints on what's available and how to use it

[open cairn]

Is alternate equivalent to the the use of | and || operators in webui? Particularly something like [thing | thing2 | thing3].

[dim sonnet]

yes

iirc the auto syntax for that works too

[open cairn]

Does swarmui use freeu or freeu2 by default?

[dim sonnet]

idk what freeu2 is and googling it just gets results about apple's free u2 album lol

freeu was never particularly impressive either way imo

[open cairn]

Its an improved version of the original freeu implementation outlined in the following paper. https://arxiv.org/abs/2309.11497v2 . I think there is a node available for it in comyui along with the original v1 implementation. My understanding is it is supposed to be more efficient while giving better results, although this is not something I have tested too much myself. Being able to choose which versions of freeu is used would be a nice feature to have in swarmUI for those who rely on it for their workflows.

[dim sonnet]

that's the freeu paper

the 2 part is the part i haven't heard of

oh, heck, there is a FreeU_V2 in comfy

... is that even an actual thing or is that just comfy did an experiment 9 months ago and published it

[open cairn]

I have only used the v2 implementation in webui

[dim sonnet]

post on the swarm issues page @ https://github.com/mcmonkeyprojects/SwarmUI/issues to track a feature request for that

[open cairn]

Done.

[potent panther]

new compromised account

[cold dirge]

🤔

[eternal aurora]

Just update current Swarm version; or a fresh install?

[dim sonnet]

if you're on SwarmUI, just update. If you're on the old stability version, need to update, then migrate, then update again - or reinstall

[eternal aurora]

Thanks

[haughty moon]

is there a way to define what the "Upscale 2x" button does? It seems to be using init image instead of refiner/upscale, takes longer, and introduces more change in the image than using the refiner tab to upscale

Also, when using the 'Reuse Parameters' button it resets the refiner settings to defaults if that data does not exist in the meta. I can see why disabling it would happen, but it's annoying to have to reset all those settings every time

[slender granite]

yooooooooooooooooooos ultimate upscale. Do not upscale latent. It suffers from bullsh*t

[potent panther]

it's a useless button

use refiner instead

[haughty moon]

I do use refiner, that's why i want to change the button

[potent panther]

Oh like having the upscale button use your refiner settings instrad?

Like a button called Refine

[haughty moon]

Yes, exactly

[gray geyser]

is swarm better then a1111 and comfyui ?

[dim sonnet]

Better than auto? Yeah
Better than comfy? Better interface, same internals. Swarm is built on top of comfy

[gray geyser]

i will try it tonight 👍

i hope it directly take civitai models and prompts to interface and save them as history.

since i am a non native speaker i prefer generated prompts.

[split sedge]

I'm new to Swarm, so bear with me, lol. So I just jumped from Forge to Swarm and I'm trying to figure out Hires Fix. I created an ESRGAN Folder in the Models folder and copied my ESRGAN_4x and others into it. But when I click on Upscale Method, none o fmy upscalers are there. Any ideas what I might be doing wrong?

[dim sonnet]

the folder should be named upscale_models

after setting things up, you can refresh the models list or restart swarm (whichever's easier for you) and it should load

[split sedge]

Thanks! I just tried that, but not seeing it in any of the drop-down menus in the Refine/Upscale area. I think I'm maybe just looking in the wrong place?

[dim sonnet]

that looks right, it should show up under Upscale Method

[split sedge]

I appreciate your help. I was thinking they should be there too. For whatever reason, they're just not popping up. I'll try to reinstall in the morning. Thank you again for your help!

[lean spade]

Does swarmUI support flux?

[lean spade]

When I specify the number of images, is that like Aut1111 batch size or the number of batches of size 1?

[dim sonnet]

yes

[dim sonnet]

there's a separate batch param if you want batching, but for the most part it's pretty redundant to bother

[lean spade]

I'm getting an error when trying to load flux dev model in Windows

Error loading model on backend 0 (ComfyUI Self-Starting): System.InvalidOperationException: ComfyUI execution error: ERROR: Could not detect model type of: H:\projects\AI\SD\Base\flux1-dev-fp8.safetensors
at StableSwarmUI.Builtin_ComfyUIBackend.ComfyUIAPIAbstractBackend.GetAllImagesForHistory(JToken output, CancellationToken interrupt) in F:\projects\AI\swarmAI\StableSwarmUI\src\BuiltinExtensions\ComfyUIBackend\ComfyUIAPIAbstractBackend.cs:line 451

[dim sonnet]

if you didn't use the official download of flux, some third party reups have the metadata janked, meaning you'll have to fix it yourself

should join the swarm discord btw, pinned in this channel

[split sedge]

I got everything working and it's glorious! Thanks again for your help. Sometimes a fresh install is better than trying to figure out what you screwed up, lol. Got Flux working on Swarm now too. Amazing.

[robust summit]

a boy and a girl , bench

[lean spade]

why does the standard install-windows.bat installs version 0.6.4, not the latst 0.9.1?

does it do that only for me?

[dim sonnet]

because you're in the StableSwarmUI repo not the modern https://github.com/mcmonkeyprojects/SwarmUI repo

[quasi marten]

Pinned message should probably be updated to new repo.

[quasi marten]

sigh... had issues with swarm in the past and put it aside.
Revisiting, and can't even get it to run. 😦
Brand new install.
[ComfyUI-0/STDERR] OSError: [WinError 126] The specified module could not be found. Error loading "C:\sd\SwarmUI\dlbackend\comfy\python_embeded\Lib\site-packages\torch\lib\fbgemm.dll" or one of its dependencies.

[quasi marten]

seems to be a known issue - nightly torch resolved for me.
more info: https://github.com/pytorch/pytorch/issues/131662

[gray geyser]

same issue with comfyui

[dim sonnet]

a fresh install of swarm today should not have that error

[gray geyser]

@dim sonneti never had issue in swarm ui , i had it with comfy, was it a general torch issue ?

i mean how come installing extensions , restarting and everything get messy.

[dim sonnet]

yes it's an upstream torch issue at the core, that comfy&swarm both have workarounds for

[dim sonnet]

because the pip landscape is a nightmare

in a better world we'd all leave python behind

[gray geyser]

i hope one day all these webui be standalone apps just like amuse 2.0 no venv no torch... no browser.

is it possible ?

[dim sonnet]

well there will always be stuff inside but yeah not having to look at / think about those, 100% I hope that too

"no browser" we can do that pretty easily by abusing Electron

Electron is just tiny embedded browser that hides that it's a browser. Which for example is how even Discord works, it's just electron

[dim sonnet]

Swarm itself is partially attempting this dream

Most things Just Work ™️

but... damn depending on python makes such a mess

[gray geyser]

is it a tough work from scratch if u leave python ?

[gray geyser]

for machine learning yeah

its tough to leave python

there's a C++ version of SDXL on github that seems cool but everything is much more work in C++

[dim sonnet]

I mean, https://github.com/dotnet/TorchSharp exists

literally pytorch but shoved into C#

very behind tho i think

[gray geyser]

thanks thats really cool

[silver spade]

hey, how can I load .sft files in swarm-ui?

[dim sonnet]

if you're using the correct current SwarmUI (not old Stability version) https://github.com/mcmonkeyprojects/SwarmUI?tab=readme-ov-file#installing-on-windows they work automatically

for flux info see also https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Model Support.md#black-forest-labs-flux1-models

[clear skiff]

does swarmui not support fp16 flux-dev? when i tried to use swarmui again recently it threw out a bunch of errors when i tried loading the regular fp16 stuff, works fine in comfy tho.

oh nvm, i see the option in the text u provided

[earnest sinew]

Hi! I'm using Swarm for the first time and I'm having a blast, but I'm having some trouble with the Segmentation feature

It detects the face correctly but regardless of what I write in the segment prompt (or nothing to repeat the main prompt), the segmented part in the final image is very blurry and low res

Does anyone know what could cause this?

[dim sonnet]

Swarm 0.9.2 Beta Release https://new.reddit.com/r/StableDiffusion/comments/1eosnug/swarmui_092_beta_release/ & https://github.com/mcmonkeyprojects/SwarmUI/releases/tag/0.9.2-Beta

[signal forge]

I keep getting an error about an unset dtype when I try to load the flux workflow do I need to be on a different git branch?

Seems Like it only sees sdxl and sd3

[dim sonnet]

make sure you're on https://github.com/mcmonkeyprojects/SwarmUI?tab=readme-ov-file#installing-on-windows and it's fully up to date

the old stability version is way too old to support flux naturally

[signal forge]

another git pull fixed it, thanks

my first flux gen 🥳

[silver spade]

will swarm-ui support lora for flux?

[dim sonnet]

it does, just not for nf4 atm

[mighty raptor]

everyone's looking for gguf support too

[dim sonnet]

swarm supports gguf

[signal forge]

can you load flux models in the generate tab? I know they work in the comfy workflow tab

[dim sonnet]

yes

[signal forge]

ok, they dont show up for me,do I need to be on a diff branch?

☰
StableSwarmUI v0.6.4.1 (2024-06-21 09:26:07)

[dim sonnet]

the Stable version is the old version, i quit stability a few months ago, the project is independent now https://github.com/mcmonkeyprojects/SwarmUI?tab=readme-ov-file#installing-on-windows

[signal forge]

thanks, I knew about that situation, but not that a new github was the result, but that does make sense. thanks

[dim sonnet]

that's the old repo, you want https://github.com/mcmonkeyprojects/SwarmUI?tab=readme-ov-file#installing-on-windows

[sudden skiff]

When I launch SwarmUI, I get this error:
https://cdn.discordapp.com/attachments/1002602742667280404/1279316536204726314/image.png?ex=66d3ffde&is=66d2ae5e&hm=44bab20eba93dd2b5e5f972a0c7bb3aa452b6180b04bbcff5c966b9b7af3c43a&
Here is my cmd log:
https://cdn.discordapp.com/attachments/1002602742667280404/1279316938064924745/image.png?ex=66d4003e&is=66d2aebe&hm=d3beb51d0b1772012041d849d788781284e7f16d8ab4e6051c4ece1773af64f1&

[dim sonnet]

yeah AMD on Windows sucks, it'll do that. you can ignore that one. you'll want to disable previews in server->backends to make it less crashy

ideally use WSL or Linux since those have better drivers

[sudden skiff]

All of the boxes are grayed out, I cannot check or uncheck them. How do I uncheck them? Also, I should change that extraarg to --use zluda, shouldn't I, or no? Not that I can edit the text boxes either atm

[dim sonnet]

click the edit button (the lil pencil at top right), then edit, then save

[sudden skiff]

Oh, Lord, I thought that was just for name changing.

Thank you.

[dim sonnet]

zluda does work better than directml if you can set that up

[sudden skiff]

I'm embarrassed.

Yeah, I already have comfyui on its own using zluda just fine

[open cairn]

Is there any additional information anywhere about what each of the refiner modes do? I am trying to figure out what each does, what settings are ideal for them, and what the use-case for each mode might be.

[dim sonnet]

do you mean the Refiner Method param, or something else?

[open cairn]

that is the one

[dim sonnet]

the method param is advanced and should usually be left on post-apply, step-swap is just the original technique suggested for the SDXL Refiner Model, and Noisy is a variant of that which arguably works better in some cases

the default is the one to use 99% of the time

the actual internal difference is post-apply runs one generation, then img2img it, vs step-swap will stop the first gen before it's done, and img2img at the identical step count where it left off with the other model, and noisy will do that with an extra step iirc

so eg for 20 steps at 0.2 refiner control, PostApply does 20 steps base then 4/20 steps refiner, vs StepSwap does 16/20 steps base then 4/20 steps refiner

[open cairn]

are step-swap and noisy preferable when using a different checkpoint for refining than the one for generation?

[dim sonnet]

no, again, those are almost never a good idea in practice

[verbal junco]

does flux work

on swarm?

[signal forge]

yes

on swarm, not stableswarm

[signal forge]

something with comfy's newest commits broke the swarm workflow graph from loading. I had to revert comfy to v0.2.2. I'm running linux

I was able to run comfy separately

[dim sonnet]

update swarm to latest - also join the swarm discord lol, i announced the patch for that issue at very close to the same time you posted here

[signal forge]

cool, thanks

[open cairn]

anyone else find that the model downloader will fail for some models for no reason with the error "appears to be a CivitAI link, but seems to not be valid. Please double-check the link."?

it will fail to read metadata for these same urls, but not others

[dim sonnet]

gimme an example link that doesn't work

[signal forge]

I dont know what others do for inpainting, but I've found the krita plugin is pretty fabulous. inpainting was the thing I missed most I think. and it has basic flux support (no controlnet yet it seems)

[amber rune]

I actually love Forge inpaint upload, I'm a Photoshop person and feels so much easier to mask stuff in PS and paste in Forge

Been trying Swarm as a step gap to comfyui but not easy

Took me a LOT of time just to figure how to have all of them using same models,clips,controller, etc files

[dim sonnet]

• SwarmUI 0.9.3 Beta Release https://github.com/mcmonkeyprojects/SwarmUI/releases/tag/0.9.3-Beta
  Reddit thread: https://new.reddit.com/r/StableDiffusion/comments/1g2n9ya/swarmui_093_beta_release/

[open cairn]

Is it possible to add samplers and schedulers to swarmui like you would with comfyui?

[dim sonnet]

yes

[mighty raptor]

@grand sun stable swarm stuff in this channel

[potent panther]

SwarmUI*

[mighty raptor]

was just letting him know what channel to find what he and i had been DMing about

[potent panther]

StableSwarm shouldn’t be used anymore

[mighty raptor]

really?

[potent panther]

What

[mighty raptor]

first of all, you weren't reading the conversation i had with makrok, so you don't know what swarm i was talking about. i pinged him here to let him know what channel to look in. and secondly, not sure why you're down on stable swarm, but also not sure why you assumed that's what i was talking about

[potent panther]

Oh god what are you on about

[dim sonnet]

"StableSwarm" is the old name of the project that moved to now be the independent "SwarmUI"

[open cairn]

Stopped having the issue with a later update, not sure which update fixed it for me.

[quasi marten]

Trying to use a gguf model (flux) but get this. Anyone know how to use these?

`12:33:05.747 [Error] [BackendHandler] Backend request #92 failed: No backends match the settings of the request given! Backends refused for the following reason(s):

• Request requires model 'flux/fluxdev/gguf/flux1-dev-Q5_K_S.gguf' but the backend does not have that model`

I also get errors with the nf4 version. (v1 and v2) 😦

[dim sonnet]

relevant docs here: https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Model Support.md#gguf-quantized-models

make sure you put gguf's in diffusion_models and not the normal model folder

[quasi marten]

I use custom specified folders for my model files. Is there a way to redefine the diffusion_models location within the settings?

[dim sonnet]

not currently no, that's a weird special case folder

it needs more thought on how to handle it well and/or cheating and hacking comfy to just load diffusion_models out of the normal model folder instead of separating them like that

so for now it's just an internally tracked special magic folder

[quasi marten]

Got it, thanks for the info. 🙂

How about the upscaler models folder? Same thing?

[dim sonnet]

that one's less special, though I don't think anybody's ever asked to be able to rename it lol. Can post a feature request @ https://github.com/mcmonkeyprojects/SwarmUI though idk why you'd need to rename that

[quasi marten]

I keep all my models (checkpoints, loras, upscalers, etc.) on a separate, larger drive, from where I run UIs. They all get configured to point to the same alternate location and share models.

[dim sonnet]

oh - just edit the ModelRoot setting then

you don't need to edit anything else

[quasi marten]

Fair, i could give that a try. 🙂 Will need to restructure the folders a bit, but should work.

[quasi marten]

Where are presets stored on the filesystem?

[dim sonnet]

user database -- if you want manual file wrangling for em, you can export/import in the UI

btw there's a dedicated swarm discord linked in the pin here

[quasi marten]

Huh.. turns out I was already in there. 😄

[eternal aurora]

https://huggingface.co/stabilityai/stable-diffusion-3.5-medium/tree/main

[barren moat]

Haven't seen you much in the Swarm server...

[eternal aurora]

Since Flux came along I've been more of a ComfyUI user - can never match up my personal workflows in ComfyUI as a backend in Swarm!

[open cairn]

Does swarm's download utiltiy only support civitai downloads, or can it also download from other places like tensor.art and huggingface?

[dim sonnet]

It has full support for Civitai, it technically supports HF but limited
the difference is civitai has a metadata API that swarm reads to load in the model's metadata automatically while downloading, whereas HF doesn't have a matching API, and worse usually has files without metadata of any form anyway.
Other platforms can be supported if a feature request is submitted to https://github.com/mcmonkeyprojects/SwarmUI/issues

[lavish breach]

Hey, Im having a possible issue, i installed swarm and im getting this stuff in the server log, and I also get a warning on the top, will this be fine or no?

[dim sonnet]

at a glance, I'm guessing you have an AMD GPU that you've previously made work with Zluda? You're gonna have to patch the comfy install at SwarmUI/dlbackend/comfy with zluda as well

[lavish breach]

Okay cool, i didnt know it required that at the moment, I appreciate it, Ill try that later today!!

[dim sonnet]

swarm has an AMD compatible install by default (with DirectML), but you look to have disabled that in the installer. Which is good if you want to use zluda tho, zluda's better anyway, just not particularly clear how to automate in an installer, especially after the whole AMD takedown notice thing

[lavish breach]

I see, yeah I had a bit of trouble trying to figure that out but im gonna try and add the comfy install into the folder!

[bronze shale]

for what it's worth I find it easier to install and run ComfyUI separately and connect SwarmUI to it later (I run two instances of ComfyUI, one for each GPU)

Make sure you follow the readme and install the requirements.txt stuff

[dim sonnet]

it is very heavily recommended that you absolutely do not do that

there's a lot of weird ways that can go wrong and/or have unexpected non-obvious side effects

if you have a pre-existing Comfy install, use ComfyUI Self-Starting for it

[bronze shale]

I found exactly the opposite. I'll try it again I guess.

[dim sonnet]

the issues that hit you with self-start are in your face the moment you set it up, and once you fix them it's good

the issues that hit you with api by url are hidden surprises for later

[bronze shale]

I have my comfy stuff backed up at least

[uncut prism]

Started trying out swarm but... I miss my Automatic1111 extensions

[plain cradle]

I have started and trying out swarm as well but I am getting way worse images using swarmui than with auto 1111.. I am not sure why I am doing wrong and would love to get some help ❤️

[dim sonnet]

What extension(s) specifically are you missing?

[dim sonnet]

That's pretty vague, would need more info. In theory as long as all your parameters are identical, you should get (near) identical outputs. (Not exactly identical, seeds and prompts are parsed very slightly differently).
So the two most likely problems are (1) you have a parameter difference somewhere you've missed or (2) you're just seeing different images and succumbing to human bias, it's different so must be wrong (I've seen a lot of people do this, but then if you gen a bunch and blind yourself to which came from which UI and look them over, you can't actually pick one as better than the other anymore)

[plain cradle]

I did a comparison with auto 1111 and SwarmUI and all same settings (I think) and the swarmui was worse

[dim sonnet]

My only reply to that is repeating what I said above

So the two most likely problems are (1) you have a parameter difference somewhere you've missed or (2) you're just seeing different images and succumbing to human bias, it's different so must be wrong (I've seen a lot of people do this, but then if you gen a bunch and blind yourself to which came from which UI and look them over, you can't actually pick one as better than the other anymore)

[plain cradle]

Could you recmmend some good negatives and settings?

[dim sonnet]

your results are bad because you're trying to use LoRAs that aren't right

the LoRAs menu at the bottom of the UI is the recommended way to set those up, if you must use prompt syntax, type it into the Swarm prompt box (don't just copypaste from auto) so that it will give you tab completions to get the names correct

side note: do remember this is a discord with a few hundred thousand people in it, and consider what prompts you want the general public to see you doing lol

[plain cradle]

I got not Lora's setup on my SwarmUi atm

This is with SwarmUI and Auto 1111

[hardy falcon]

any swarmui experience @mighty raptor

[mighty raptor]

a little. what is wrong?

[hardy falcon]

im getting All available backends failed to load the model 'D:\SwarmUI\Models\Stable-Diffusion/Flux\getphatFLUXReality_v20.safetensors'. error @mighty raptor

it works on sdxl models but not this flux model i just downloaded

[mighty raptor]

what flux model did you download and where'd you download it from?

[hardy falcon]

getphatfluxreality v20 from civitai

22 gb

[mighty raptor]

does it work on other flux models?

[hardy falcon]

this the only one i have but it was a clean download

and lot of ppl have success using it from civitai

[mighty raptor]

i don't know as i've not used that model. this is a question i think @dim sonnet probably needs to see

[hardy falcon]

@dim sonnet u there?

[kind marsh]

swarmui on online services https://github.com/gutris1/segsmaker

[jagged mortar]

Hi, new to swarm. I am trying to use control nets. I only have one control net downloaded, but my preprocessor looks like this. But anytime I try to use any of them I get an error like the one shown. Does anyone know how I download these pre processors so they actually work OR remove them from the list so Swarm stops lying to me?

[dim sonnet]

some of those preprocessors need models to work, and they're autodownloaded - this one failed its autodownload. Copy the path it has in the error, go open the folder it named, and delete the whole .cache folder, then restart and try again

for depth btw iirc MiDaS is the recommended default preprocessor for it

[jagged mortar]

Thanks I will give that a shot! And good to know, sadly I have the exact same error for midas as well.

[dim sonnet]

same basic solution should fix that too then -- but you'll want to figure out why downloads are failing if that happened repeatedly

you miight need to enable long paths according to a quick google search https://superuser.com/questions/1807770/how-to-enable-long-paths-on-windows-11-home

[jagged mortar]

Ill take a look at that when I have some more time. Sadly I deleted .cache, reset, and it just recreated .cache and printed the same error as before.

[jagged mortar]

Longpaths looks like it fixed it, thanks for that info, never heard of that before.

[dim sonnet]

pushed https://github.com/mcmonkeyprojects/SwarmUI/commit/bcced64eb8f75d61ba7ec9bfbfea8fb396475e1d to autodetect that and give a better error message

[true thistle]

I updated because I got a message saying there were some available. I updated. Now it's broken.

"No backends available."

Might be about time to switch to Invoke. Gotta generate something tonight and I have no idea what to do about Swarm now.

[true thistle]

My Invoke generations look like ass, not sure why. 😅 Guess I'd better find out what's wrong with Swarm...

2024-11-23 20:48:54.050 [Info] Self-Start ComfyUI-0 unexpectedly exited (if something failed, change setting LogLevel to Debug to see why!)
2024-11-23 20:48:54.050 [Debug] Status of ComfyUI-0 after process end is LOADING
2024-11-23 20:48:54.050 [Info] Self-Start ComfyUI-0 had errors before shutdown:
[ComfyUI-0/STDERR] Total VRAM 24564 MB, total RAM 98045 MB
[ComfyUI-0/STDERR] pytorch version: 2.3.1+cu121
[ComfyUI-0/STDERR] Set vram state to: NORMAL_VRAM
[ComfyUI-0/STDERR] Device: cuda:0 NVIDIA GeForce RTX 4090 : cudaMallocAsync
[ComfyUI-0/STDERR] Using pytorch cross attention
[ComfyUI-0/STDERR] [Prompt Server] web root: C:\SwarmUI\dlbackend\comfy\ComfyUI\web_custom_versions\Comfy-Org_ComfyUI_frontend\1.3.26
[ComfyUI-0/STDERR] Traceback (most recent call last):
[ComfyUI-0/STDERR] File "C:\SwarmUI\dlbackend\comfy\ComfyUI\main.py", line 225, in <module>
[ComfyUI-0/STDERR] load_extra_path_config(config_path)
[ComfyUI-0/STDERR] File "C:\SwarmUI\dlbackend\comfy\ComfyUI\main.py", line 190, in load_extra_path_config
[ComfyUI-0/STDERR] for y in conf[x].split("\n"):
[ComfyUI-0/STDERR] ^^^^^^^^^^^^^
[ComfyUI-0/STDERR] AttributeError: 'bool' object has no attribute 'split'
2024-11-23 20:48:55.047 [Debug] [Load ComfyUI Self-Starting #0] ComfyUI-0 self-start port 7822 loop ending (should now be alive)

Anyone know what's going wrong here? I get backend errors ever since applying the latest SwarmUI update.

[dim sonnet]

your comfy backend is outdated - go to server->backends->edit the comfy backend->set autoupdate to always or Forced->save

[next arch]

@dim sonnet interested in adding this #1188183242747158630 message to SwarmUI?
It's CFG inference from a distilled model. Samples are using a negative prompt on Flux Dev.

(sorry about the ping on the other server. I didn't know about this channel before)

[dim sonnet]

not sure what specifically you're asking? Flux-Distilled works in Swarm if that's what you mean. also ftr swarm's dedicated discord is linked in the readme https://github.com/mcmonkeyprojects/SwarmUI?tab=readme-ov-file#swarmui

[next arch]

that's not what I mean. Follow the link. On the left side of the samples is Flux Distilled. On the right side of the samples is inference using CFG and negative prompts, as you would infer from an undistilled model - but using the distilled model (Flux Dev).
Is that already in Swarm? If so I haven't found it yet.

thanks for the link. I'm there now if you prefer to answer there.

[dim sonnet]

Er yes sorry I meant flux-undistilled is supported yes

There's an actual model where they undid the distillation

To readd cfg

[next arch]

I know but the quality isn't very good. This is inference using the distilled model but using real CFG.

not another model

it's a bit of math and code

[dim sonnet]

Using cfg on standard flux dev isn't great

You can use eg dynamic thresholding to unbreak it

Few different "cfg fix" techniques out there

[next arch]

🤔 so Swarm can already do that and/or not interested to implement negative prompts for Flux Dev distilled?

[next arch]

I think it works great. much better realism with negative prompt than distilled with only positive prompt

really not sure how dynamic thresholding is related to this

I tried the un-distilled model. It's nowhere close to this in quality

[dim sonnet]

If you just want to use raw cfg and nothing more on flux, nothing is stopping you

[next arch]

I think we are talking past each other. raw CFG on Flux, i.e. just increasing the CFG factor from 1 in SwarmUI, obviously fails. That's not what I was proposing.

[dim sonnet]

Then what are you proposing?

It's not raw CFG, it's not a CFG fix technique, it's not an undistillation model. But it is CFG. What is it?

[next arch]

replied here: #1243185862234210389 message

[quasi marten]

Does Swarm support this LTX video thing?

[dim sonnet]

yes https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Model Support.md#lightricks-ltx-video

[barren moat]

If you need help setting it up I just did a video on it.

https://youtu.be/Zzt9kXMNd-M?si=BL6SPVyOJi-CseBX

Did y’all know there’s a dedicated swarm discord ran by the one and only mcmonkey? Now you do 🙂

[quasi marten]

i'll scope it out

[barren moat]

if you need help just HMU

[dim sonnet]

• SwarmUI 0.9.4 Release! https://github.com/mcmonkeyprojects/SwarmUI/releases/tag/0.9.4.0-Beta
  Reddit thread: https://new.reddit.com/r/StableDiffusion/comments/1h81y4c/swarmui_094_release/

[jaunty saddle]

oh shitt this is PEAK

[quasi marten]

Thanks Alex! Great to see the new release! 😄

[quasi marten]

is that 'frame interpolation' install really required?

[barren moat]

Not that I noticed

[quasi marten]

cool

got it running.. just got pretty butt results... need to find a good test prompt 🙂

well... i mean.. it made something. 😄

[barren moat]

it's not horrible

[quasi marten]

enough to know it works 😛

[jaunty saddle]

Why doesnt the lora get saved in the metadata?

im using comfyui workflow to generate tab tho

it should have the shrek lora but its not there in metadata

need to post to civit

[dim sonnet]

things added by a workflow and not listed in the generate tab don't go into metadata

[jaunty saddle]

does this mean ill have to generate inside comfyui itself?

[dim sonnet]

if you want exactly correct metadata either (A) use the generate tab or (B) use the comfy tab

using both together gets a lil wonky

[jaunty saddle]

im not sure if vpred is automatically enabled on swarmui generate tab

dats why i need comfyui sampling discrete

[dim sonnet]

Edit Metadata on the model -> set the prediction type

[jaunty saddle]

pretty sure newer noobaixl vpred has the state dict

[dim sonnet]

it should be automatic if marked properly for comfy yes

otherwise you can do it manually via the edit metadata menu

either way generate tab should work fine

[quasi marten]

So I've been using the "comfy" release of the flux dev model (flux dev fp8 with the text encoders built-in). And I use SwarmUI.
Is there now support for a faster workflow versus the original flux dev model? Anyone know?

[dim sonnet]

what do you mean a faster workflow

swarm supports both the "comfy version" and standard version of flux models ftr

[quasi marten]

I recall the nf4/gguf models allowing for faster generation, but they required special support. Thought maybe there may be some other model releases with optimizations since then.

[dim sonnet]

gguf is supported in Swarm. Nf4 is too technically but i don't recommend it.
those aren't faster in general, they use less vram which is faster if you're low on vram but slower if you have a high end card.

there's "hyper" flux which is like a turbo thingy, it lets you run flux dev in ~8 steps with a slight quality drop

[quasi marten]

Hmm, before I switched to Swarm they were definitely faster for me. NF4 was anyways. GGUF was slightly. 3060, 12GB VRAM.
Haven't tried the hyper flux thing. I'll take a look. I didn't much care for schnell.

[dim sonnet]

yeah 12gib vram is lowend for ai, so gguf would help yeah

flux is a 12b model, so fp8 needs 12 gigs for the model and a few more gigs for activations n context n all, so 16gib+ is flux's preferred space

[quasi marten]

ah... maybe when that 5000 series drops. 😉

[timid crest]

Hello there, any update regarding this feature https://github.com/mcmonkeyprojects/SwarmUI/pull/289, thanks

[vocal bay]

Controlnet union inpainting with sdxl seems to do nothing unless I set the controlnet strength super low (0.25 or something). How should I start looking for the source of the problem?
If I disable the controlnet, inpainting works but consistency is of course completely lost.

[vocal bay]

Got it workoing. SwarmUI has a bug in that it doesn't set "black pixel for xinsir cn" on in inpainting processor when controlnet union is used.

[dim sonnet]

that's less bug more feature request, that's not something I've even heard of before. Please post as a feature request to https://github.com/mcmonkeyprojects/SwarmUI/issues

[vocal bay]

Well, if you consider "inpainting is completely broken for SDXL models" not a bug, sure

[dim sonnet]

That is not true. Inpainting is broken for xinsir(?) controlnet-union inpainting sdxl models, which is a class of model i've never touched

inpainting is great on SDXL outside of that specific thing

it's "lack of feature" not "bug" because i've never even attempted to include support for that specific subclass of model

[dim sonnet]

• SwarmUI 0.9.5
  Release notes: https://github.com/mcmonkeyprojects/SwarmUI/releases/tag/0.9.5-Beta
  Reddit thread: https://www.reddit.com/r/StableDiffusion/comments/1ieh81r/swarmui_095_release/

[hardy sorrel]

good stuff

[quasi marten]

Docker support to sandbox - nice

[dim sonnet]

why did you copypaste an earlier message out of context

[gritty ether]

🤖 (most likely)

[teal ruin]

Hello~

Can anyone help me understand how to fix 'error code 128' when installing SD Auto1111?
Someone suggested using Swarm-UI but it was trying to run comfyui instead of auto1111...im so lost and confused

I've been trying to get SD running for the past 24 hours but nothing is working, i've tried everything I could think of from so many different forums

[turbid tusk]

Well swarm has a comfy ui backend while having a easy to use interface

[turbid tusk]

If you get an error. i recommend posting the full cmd into #🤝｜tech-support

[teal ruin]

Okay. I will try again later~
Thank you!

[dim sonnet]

guide to generating Wan videos faster in Swarm for anyone who wants that https://www.reddit.com/r/StableDiffusion/comments/1j0znur/run_wan_faster_highres_fix_in_2025/

[sonic dirge]

@dim sonnet hey sorry to bother you. i just saw your reddit article there that you posted about highres fix with Wan.
I dont use Swarm, but I was so interested in what you mentioned at Step 2: init image with the creativity value.
Since I am using ComfyUI and I think that swarm uses comfy in the background, this got me really interested as to what
is the equivalent of that technique in pure ComfyUI? Like what would be the workflow for that? Can you share a similar
workflow of that technique but for just ComfyUI? or is this technique only possible on swarm?

[dim sonnet]

that's just the way in swarmui you upload an image for use with image2video, the comfy equivalent is... use the reference image2video workflow

[sonic dirge]

sorry for late reply, was doing something. when you say reference image2video workflow, you mean from here:
https://comfyanonymous.github.io/ComfyUI_examples/wan/ ?

But isnt that just image to video and in your example you have basically like a video to video, cause you generate from step 1
and then step 2 input is that video?

can you quickly make a comfyui workflow and share it to showcase how would this work, cause i really dont get it

[dim sonnet]

OH I latched onto the wrong part of your initial question

The highres part is: load the first video, pass to an Upscale Image node, then run it through a sampler with whichever technique (image2video workflow recommended). Vae encode the 'image' (video) and pass that as the latent, instead of the latent generated by the i2v node

for non-swarm you need VHS nodes to load a video directly. Swarm has its own node for loading videos

also note: when in doubt, just set up a gen how you want in Swarm's Generate tab, then click the Comfy Workflow tab, and hit 'import', and it will give you the full comfy workflow for what you set up

swarm is designed among other things to help teach comfy

[sonic dirge]

kk il try ty 🙂

[limber zodiac]

I want to install SwarmUI on my Mac, but I happen to have various versions of Python installed, including v3.13. I'm not sure where these versions are installed. I read on the Github not to use v3.13, so how can I avoid using this version? Or put otherwise: how to force SwarmUI to use e.g. v3.10 (which I have on my Mac).

[dim sonnet]

if it's installed in /opt/homebrew/opt/python@3.10 (which it should be by default), it'll work automatically
Swarm favors 3.11 > 3.10 > 3.12 > system default, in that order

[limber zodiac]

Ah thanks! I will give it a go and report back.

[dim sonnet]

Swarm has a website now https://swarmui.net/ and a submission thread in the Swarm discord for images to add to the background scroll

[limber zodiac]

I installed SwarmUI on my Mac Studio Ultra M2 and the process was extremely smooth and succeeded without any problem! I'm very impressed by the generating speed and the user interface. Of course I still have many aspects to discover, but I'm sure it will replace all my other applications for AI, like ComfyUI, Automatic1111, Draw Things etc. etc. Well done!

[limber zodiac]

I'm still OK with SwarmUI, but not for the first time when using AI applications on a Silicon Mac I get an error related to mps. This one came up when choosing the 2x upscale option in the image editor. For me atm it is not a big deal. but still...

[dim sonnet]

wow, straight up internal pytorch error. ow

[limber zodiac]

I must be getting old, but how do I launch the server? (I had closed Safari on my Mac and had the idea that after starting Safari inputting "localhost:7801" would start SwarmUI, but Safari tells me it can't find the server...)

[limber zodiac]

I'm feeling a tiny bit younger 😉. SwarmUI starts from the terminal after a 'cd' to SwarmUI folder and using the 'launch-macos.sh' command. (This was already pointed out on the website, sorry).

[limber zodiac]

So not your cup of tea?

[dim sonnet]

I don't even have a mac personally, and either way that's internal pytorch ie an error in these guys https://github.com/pytorch/pytorch code

it's possible that eg a manual torch upgrade will fix it? if it's an already-solved bug. In which case see here https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Troubleshooting.md#i-need-to-install-something-with-pip re pip install and you want to install -U torch torchvision torchaudio

[limber zodiac]

OK. I will have a look at the pytorch github and if needed report the error. Thanks!

[solemn vessel]

#imagine dragon volando

imagina un gato con alas

lllllllllllllllllllllllllllllllllllllllll

imagina un dragón volando sobre un castillo

manzana

visita una fábrica de coca cola

[gritty ether]

That's not how it works .... #artisan-faq

[dim sonnet]

SwarmUI 0.9.6

Release notes: https://github.com/mcmonkeyprojects/SwarmUI/releases/tag/0.9.6-Beta
Reddit thread: https://www.reddit.com/r/StableDiffusion/comments/1jztcuu/swarmui_096_release/

[sonic dirge]

@dim sonnet hahahaha, im sorry i just love your Readme stuff, i was reading your HiDream readme and you have stuff like:
"HiDream uses a quad-textencoder of CLIP L, CLIP G, T5-XXL, and LLaMA-3.1-8B (this is unhinged I'm so sorry for your RAM size)" 🤣

i mean it is ridiculous indeed, according to what i saw from one reddit post, it seems the llama is doing all the heavy lifting anyway, so maybe
we can use only that, idk..

[fringe dragon]

is there any way to change the noise mode inside swarm ui ? I got the feeling it's using auto 1111 gpu noise instead of cpu or cpu/gpu + internal seeds. I'm saying this because i'm comparing gens with comfyui and the noise type, in comfyui when, in auto1111 gpu noise, the images composition changes a lot, and for cetain workflows, overbakes the image, same type of overbaking the image i'm noticing in swarm ui. Any other noise mode does not do this in comfy. So i was wondering if swarm was doing an auto1111 gpu noise style and can we change that ?

[dim sonnet]

(replied to above in swarm discord)

[random valley]

||What a healthy-looking eye.||

[dim sonnet]

what

[merry herald]

it's healthy

[analog coral]

when trying to do randomisation between multiple prompt, like this {blue|orange|red} haired, it doesn't work, do i need to download or activate something?

[final jackal]

https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Features/Prompt Syntax.md

[analog coral]

thanks!

[jovial ruin]

fghn

[tight kayak]

Hey guys, to mcmonkey’s delight I had finally added Swarm support to my discord bot relatively recently.

I’ve been developing this bot forever, it has lots of cool features. In terms of Swarm, probably the most enticing/unique feature is “Custom Commands”- allows the bot owner to very easily create slash commands to execute payloads / workflows with customized input fields (prompt, img/vid Attachments, booleans, ints, floats, Choices, etc etc).

https://github.com/altoiddealer/ad_discordbot

[pliant agate]

please dm me or tag me so i can receive notification

[dim sonnet]

multiple images yes, 8-15 probably not?

talking more like 1-4 as the reasonable range

yes you can run kontext locally in swarm just fine

swarm basic install here https://github.com/mcmonkeyprojects/SwarmUI?tab=readme-ov-file#installing-on-windows flux kontext specific info here https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Model Support.md#flux1-tools

[pliant agate]

but i don’t know exactly where to begin on comfyui, can you please give me a basic blueprint?

[dim sonnet]

Install SwarmUI, use the generate tab, I linked docs here #🐝｜swarm-ui message

[worthy zephyr]

so I'm redownloading swarm and i wanted to ask if maybe the issue could be this.
"Run the launch-windows.bat. It will open up in Browser. Click 1. Agree, 2. No, for Default install (not AMD). 3. Just Install." I always click no for default install, i wanna make sure I'm not confused.

I have an AMD

[final jackal]

It has been roughly 1 year since I started using SwarmUI and I never reinstalled, so...
I always go for 'custom', but I really don't know if it makes any difference for the backend, which is Comfy.
I believe you should use WSL2 bc amd implementation is better on Linux. Additionally install Zluda.
You're going to have a hard time using AMD and SwarmUI is just using ComfyUI as a backend. I recommend asking in their discord for help on drivers and dependencies.
sorry it was not possible to post an invite link here. You can jump from comfy org to their discord

[dim sonnet]

it's wsl2 =or= zluda rather

zluda is a hack to make amd work on windows

on linux there's good native drivers

that user already has a thread on the swarm discord

[dim sonnet]

SwarmUI 0.9.7 Release
GitHub release notes: https://github.com/mcmonkeyprojects/SwarmUI/releases/tag/0.9.7-Beta
Reddit post: https://www.reddit.com/r/StableDiffusion/comments/1mzsc62/swarmui_097_release/

[dim sonnet]

giant Z-Image grid made in swarm https://www.reddit.com/r/StableDiffusion/comments/1paj4pj/zimage_turbo_parameter_megagrid/

[quasi marten]

Love it... going to check out the grid. 🙂