💬｜general-chat

[fervent thunder]

I've been looking at making a new model but not loras

[ruby parcel]

whoa

thats gona be big hahahaha

[fervent thunder]

lol

gonna start small

[wooden ivy]

A glowing portal to another dimension,星空背景，科幻风格，次元之门，门外是城市夜景，门内是奇幻世界，4K高清，cyberpunk style, glowing particles, futuristic, --v 5 --ar 1:1

[atomic mortar]

#artisan-faq 👍

[abstract pivot]

Hi does anyone know if its possible to manually add Samplers to Forge? I like the 'Gradient estimate' sampler that is in Reforge but it does not come with Forge, is it possible somehow to copy into the folder or install it somehow into Forge? Thanks

[tribal jungle]

hi

[warm hull]

Could i send you a dm please

[untold pewter]

sure man

[balmy salmon]

Hi! sry guys I´m having trouble to mimic the configuration of flux Dev of Civitai in my Forge Console, anyone knows wich sample methods and schedulle type they use? I´m using same Loras and a flux1-dev-fp8.safetensors checkpoint . I´m breaking my horns trying to get the right configuration, but i keep having weird results.... thx!!!

[desert dagger]

you'll probably get faster response if you post in #🤝｜tech-support

[balmy salmon]

thank u darling!

[thorny light]

hi

[solid basalt]

does anyone know the current best way to use embeddings in comfyui? all the informatoin i find online seems to be from a year ago

[tall wave]

hey, whatst the newest versio nof stabl diff? i havent touched it in a months and worndering is it upto date?

[burnt copper]

Quick question I was always told that illustrious were not be able to read SD 1.5 Lora’s however today I tried it and it did read it. So was that incorrect?

[fervent thunder]

sometimes SD 1.5 and SDXL mix a bit

but they don't mix properly

there is a thing called X-adapter that allows them to mix a lot better

its not perfect though

[autumn apex]

Hey all, anyone got a good guide on setting up a video or animation base AI maybe through A1111?

[rancid crater]

What's a good way to prompt sad eyebrows, like the opposite of v-shaped. Like this / \

[deep horizon]

there was an extension i used to have where it would auto suggest ppromt stuff as i typed it, does anyoje know what its called? got new ppc and lsot all my stuff

[sharp plaza]

I can't get a1111 to work for more complex things.. comfy is easier but all the nodes and extra stuff you have to install grosses me out cuz I don't want all this junk on my pc

is there like a download pack for common tools

[untold pewter]

i dont know if you can use video gen models on A1111, you need to use comfy

[untold pewter]

for comfy, you have to download custom nodes if your workflow needs it. Nothing you can about that 😦 . You can check if pinokio have something you looking for

[bronze quiver]

Hello, I don't understand how to use this discord to generate a picture with stable diffusion?

[atomic mortar]

#artisan-faq is the only way

We mostly run it locally on our pv

[ruby parcel]

Hmmm

Hihi

I’m running n playing with SDXL jusggernaut

I prompted this
“ a guy in sweatpants, with a grey dragon behind him, with film grain effect, he is touching his hair with one hand”

Die die doesnt want to give me dragons lol

Hmmmm

It doesn’t get the concept?

[atomic mortar]

I think its mostly the prompt but what resolution are you trying to make it in? I find concepts such as those work better in a wider format

[ruby parcel]

Ah it’s in 900 width 700 height

[atomic mortar]

Try to use aspect ratios
1:1 1024x1024
4:3 1152x864
3:2 1248x832
16:9 1344x768

[ruby parcel]

Hmmm if I switch out sweatpants to medieval armor

The dragon comes out

[atomic mortar]

You can also swap them to get portraits

Try doing a (grey dragon:1.3) with your previous prompt

[ruby parcel]

oh wow

once i use bigger reso

like 1344 768

it finally understands to plop a dragon in behind

hmmm

why?

LOL

i'm confused

[atomic mortar]

Such images are useually in wider pictures lol

So its trained like that

[ruby parcel]

ghmmmm

i see.

so it doesn't break down the context of

a dragon

a man

behind

it's more like looking at the words as a whole and see which pics in it's memory comes up

[atomic mortar]

Have you tried looking at the example images on civitai for that model? It will show you how others prompt with that specific model

[ruby parcel]

I se

Hmmm

Means it’s more to playing with the prompt words

To see if I can induce that pattern out

[tribal crown]

what is the best img2video at the moment?

for comfyui and uncensored?

[autumn apex]

How about animated images?

[ruby parcel]

i ended up

asking chatgpt

to give me a prompt

[untold pewter]

Try LTX img2vid

[ruby parcel]

A highly realistic image of a handsome man with chiseled facial features, wearing (matte:1.4) grey sweatpants. He stands confidently with his hands running through his hair, giving a relaxed yet strong presence. Behind him, a majestic silver dragon with shimmering metallic scales looms, its eyes glowing with intensity. The background is subtly atmospheric, with soft lighting that enhances the dramatic effect. The scene feels cinematic, detailed, and ultra-realistic, capturing every fine texture and expression.

dang highly descriptive

i didnt know it needs to be this

[untold pewter]

Nah A1111 is too simple for that..
Comfy is your best bet

[ruby parcel]

my originla prompt was a guy in sweatpants, with a grey dragon behind him, with film grain effect, he is touching his hair with one hand

[autumn apex]

Thats not true. because I've seen it done. I'd like to be able to do it myself too.

[untold pewter]

It's doable but it's not the best way

[autumn apex]

Is there an easy way to setup Comfy?

[atomic mortar]

Oh no

Neural prompts

[granite thorn]

still cant, did i need to use code to make it for low vram?

[warm junco]

Come to #🤝｜tech-support with screenshots of your issue

[random reef]

Did Deepseek buy Ai.com? It hasn't been confirmed. The owner of AI.com initially redirected it to Chatgpt, then to Grok and now to Deepseek.

They are flirting with these companies to try and draw attention to the name. It is currently evaluated at USD $50 million, i think they are trying to push for more..

[ruby parcel]

I'm trying to run GGUF Flex.1 alpha q4, on forge webui. hmm i got the t5, clip and ae in. and hit run

i encountered this error

must be Tensor, not NoneType

anybody know what this is?

[atomic mortar]

A post for #🤝｜tech-support

[ruby parcel]

okie

[tribal crown]

the one from civitai is good?

[obsidian venture]

hi, new here. Hoping for a suggestion.

Have not done AI work in over 6 months and things change so faaaaaast 😆 wanting to get back, and in a couple of months build a 5090 VR/AI rig. In the meantime, I have a Mac M3 to get my feet wet with ComfyUI - Which I've never used before but really need to get started with

For now I have fun learning AI generative arts doing comedy sketches for social media, and would need an image generator that can take an image and prompt and produce midjourney level results. Everything image-2-image completely corrupts the original person's face, or commercial options flag my prompts as inappropriate, when there is literally no porn, guns, or anything of the sorts.

Looking to do this sort of stuff

https://www.youtube.com/shorts/EWAGzKCMhuE

Midjourney is fine, but its not trained on my little country's political or famous people.

What would you suggest to achieve results as the one above? Generate image-to-image (conserving the person) and then text-to-video to animate? What web services do you recommend for this, before I build the PC?

Thank in advance

[untold pewter]

https://huggingface.co/Lightricks/LTX-Video/tree/main

i downloaded from here

[untold pewter]

you can use comfy portable... just download and extract the file and you are pretty much done

[autumn apex]

got a link for me? I apprecaite the help.

[gritty lava]

ya I think comfy is just another type of ui

not like, a whole diff stable diffusion xd

[desert dagger]

that's exactly what it is. ComfyUI is a user interface in which you load your models and write code by importing nodes (which are programming functions) in order to create an AI (an ai is all of your models and code running together) that will generate images or videos for you

[gritty lava]

:3

[silk dew]

where did all the diff art channels go

did they remove them except for anime?

[atomic mortar]

probably. server is hella slow nowadays so its trimmed down a little i think

[fallow veldt]

I had the idea or impression that we would constantly be seeing new developments or models, but everything seems rather stagnant. Could they be working on other things or in different ways?

[atomic mortar]

wdym? theres a lot of improvements made every day but not everyhing is showed/public

[fallow veldt]

Yes I guess there are, but not like when we went from SD 1.5 to SD XL and all the developements of ControlNet, etc, and then flux and SD3, for generating locally

[atomic mortar]

It takes a LOT of money to train new types of models

And theres LTX, hunyuan and mochi for video generation

And luminous 2 was released not too long ago (uses a LLM to prompt images)

But reforge/a1111 cant run these

Luckily, im on swarm

[dusty trellis]

Any thoughts on this MidJourney opinion piece? https://youtu.be/TVHnguyd06g

[atomic mortar]

Sorry but I'm not watching a 10 minute video. However midjourney was great but the ability to run it local will always win

Energy bill up by 40 or spend 4x that on tokens to run it online

[untold pewter]

Gg

[fervent thunder]

Had a look at the last year of their non-SD3 releases, there's been a pretty steady stream:

3D stuff:
TripoSR, Stable Fast 3D, Stable Video 4D, Stable Point Aware 3D

Audio:
Stable Audio Open

Agents:
stable-assistant, stable-artisan

in particular TripoSR has nearly 100 arxiv citations at this point

[steel crown]

Hello everyone, brand new to discord and stable diffusion. I want to learn how to use stable diffusion to design artwork and for my business. i am looking to take my logo and turn it into a shield like Capt. America. I have been trying for about a week with little success. Using chatgpt for instructions and youtube for training.

[fervent thunder]

I'd recommend looking at lokr lycoris
if you can get it working then it helps a lot because it reduces the low-rank constraint that loras have
if that is too tricky then maybe koyha lora?

[steel crown]

yeah I have downloaded and tried several loras and using controlnet but I am not even close to getting anything close. Thanks for the info neon

[fervent thunder]

no problem
these are the notes I made about lora settings if that helps

alpha = grid search from alpha = 0.5*rank to 2*rank
learning rate = grid search from 1e-6 to 1e-3
weight decay = grid search from 0 to 1e-4, possibly up to 1e-2
gradient clipping = grid search from 1 - 2, possibly up to 5
optimiser = adamW
schedule = cosine
warmup = 2.5% to 10% of steps```

[atomic mortar]

I'm not sure if he's using loras at all since he's new to this

[steel crown]

everything helps, especially when you are brand new to stuff. I have a spice company that I want to to promote.

Euro, I have downloaded someloras but I thinki will need to train my own because the ones I have used arent giving me what i need

[atomic mortar]

Tough most that i can tell you is that chat gpt is horrible for advice and if you want to make a lora you need about 50-150 images on that concept

Sometimes a partial photshop with img2img it better

[steel crown]

i just need to watch more tutorials, I do have gimp I was using before for static images

[atomic mortar]

Most i do nowadays doesn't have tutorials. Mainly trail and error

[steel crown]

for me the videos help me understand what the settings do.

trying to teach this old dog some new tricks

[atomic mortar]

Hmm i get you but unfortunately there is some knowledge gaps/lots of outdates stuff floating around due the field advancing quickly

And honestly im not too interested in making tutorial videos as that would bring whole nother headache

Oh and some are behind paywalls too

[steel crown]

I get you there. but I am guessing places like this is better because everyone works together....to a point

[atomic mortar]

If you get stuck on specific problems like erros absolutely

But workflows? Sorry i prefer to keep most of it private

I love to help people to get started on this stuff but the rest is a creative process

[steel crown]

like you said ....trial and error.I keep a notepad nearby to keep notes on what i do soi dont make the same mistakes

[fervent thunder]

sharing workflows doesn't work that well because
you then have the issue that the workflow asks them to install 10 different things, and also their versions of stuff doesn't match

the closest thing I have seen to something easily shareable/reproducible is replicate cog but its not that popular, and is not for "live" workflows

[steel crown]

seems like I will need to spend time taking parts and pieces from everywhere and see what sticks.

may get to a point where i say I am too old for this stuff LMAO

[fervent thunder]

lol its tricky yeah

[gritty lava]

Imma ask again in shere hope

does anyone have, stable diffusion 1.7.0 or at least 1.7 something

and don't mind sharing that in a zip folder, u can exclude the models and Lora's and webui-user.bat

i just really really want the old version

:V

maybe u have it on an old ssd, just sitting about, or, maybe in a folder u forgot about when u was trying to figure out how to install it ;^:

;-;

yusu? ;+; enzyzoo? ;_;

[gusty rose]

Hi All, I'm using Flux with Forge UI. Does Forge have UI for training LORA?

[broken smelt]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/tree/cf2772fab0af5573da775e7437e6acdca424f26e
this is the 1.7.0 repository

im assuming you meant automatic1111 1.7.0 instead of SD

https://github.com/AUTOMATIC1111/stable-diffusion-webui/releases is where youll find the other releases if you wanted to and from there you can view and/or download the files as they were when they were released

[gritty lava]

would u mind trying it urself, for me, it seems when i downloaded from here https://github.com/lshqqytiger/stable-diffusion-webui-amdgpu/tree/v1.7.0-amd
it just downloades 1.10 anyway

;+:

:(.

i tried typing this into cmd in the root of stable diff
git checkout cfa6e40e
that "worked" but despite saying its 1.7.0, and not throwing errors, and opening the web Ui, it would not at all move from zero percent when u click generate ;-;

i found this method on a YT video, and sum1 said worked 9 months ago..

sure def didnt work for me

:v

is there a way to stop it from downloading the latest verison?

like, if i download automatic1111 1.7.0, i wan 1.7.0, not 1.10

:v

[gleaming spoke]

Hello

[thorny light]

HI

[fervent thunder]

https://github.com/stepfun-ai/Step-Video-T2V

30B model released

[ruby parcel]

hi

quick question.

in comfyui is there an equivalent workflow to forge where

I generate and preview, and see if it's nice. if it is nice, i right cclick "send2inpaint" or "send2image" then i continue working there

thinking if i can have one workflow just for generation purpose then preview image is ok, i will "send" that image(without saving and fiddling with saving pic and finding the pic file again), to another workflow to iterate repeatedly /inpaint

[gritty lava]

dangggggggggggggggggggggggggggggg

there we go '

bru 1.7.0 is THE bomb

do ppl still say that

now we makin some gud art

stable, 1.7.0, to me, is kinda like starting an engine, that hasnt ran in like 10 years

its rough at start, but once u get it doing stuff, it seems to make much better art as it get "warmed up" for say

like how an engine's rings pop outm and loosen up, and seals expand again

[fervent thunder]

impact pack preview bridge maybe

[ruby parcel]

aight. let me have a look at that

I JUST LEARNT I CAN DRAGA AND DROP IMAGES

T.T

omg

the world has changed!

[gritty lava]

xdd

drag and drop where doe

[umbral lotus]

A larger prompt-window might be handy 🙏

[atomic mortar]

In what UI? Theres like multiple

[atomic mortar]

Comfy probably

[umbral lotus]

Oh really? I didn't know that 🤔 I'm working on https://stablediffusionweb.com/app/image-generator

[atomic mortar]

Your using a online service? Well theres a civitai, tensorart etc

But locally there's comfyUI, A1111(outdated), forge webui, swarmUI etc

And some more

[gritty lava]

u should like

do it all local

[umbral lotus]

so you're sending me to the competition? 😁

[gritty lava]

cuz free

[atomic mortar]

You wanted a bigger promptbox lmao

But if you have a decent gpu you can run it locally

[gritty lava]

ya

plus u can make ur own CSS

make wut ever size shape box u wan

[umbral lotus]

i'm not there yet. New to the whole ai-generation-thing and i must say the results of the online generators, incuding stable-diffusion's don't dissappoint. Only minor issues with ui but that's no dealbreaker

[atomic mortar]

Online versions are ok but its less control & censorship here n there if you care about that

[gritty lava]

yas

u run it locallly, the sky is the limit

or.. ur screen rather

xd

[umbral lotus]

yeah but for now i can work with that. Later on ill check local options

[gritty lava]

if u have an nivida gpu, wont take u like 4 mins

[umbral lotus]

i mean what's the investment on the hardware?

[gritty lava]

to install it local

i say, for good results/performace, nivida gpu with 8GB gpu memory

32GB physical ram

and like 200GBs for models

xd

[umbral lotus]

ah i already have lot's more 🤗

[gritty lava]

oh nice :3

i def recommend local then

[umbral lotus]

right. thanks for the tip 🙏 i'll keep it in mind

[gritty lava]

ofc <3.

[fervent thunder]

not sure what this site is

think it might be a scam site

[–]Herr_Drosselmeyer 10 points 6 months ago 

That site is scummy and if you can believe it, they used to be worse. The site used to outright claim to be affiliated with Stability AI. The company is a shell company in the UK with a Chinese guy listed as CEO. I wouldn't even trust their privacy policy. Also, be wary of potential "tacit" renewal of your subscription.```

[umbral lotus]

scam-site? 🤔 i generate pics there

[gritty lava]

might jus be me

but givin the current population of earth

id say nu

however

i think, that sites that make u buy credits or tokens

to allow u to use something u can get for free

is so wrong

like, sure, there are some ppl who dont have nothing but a cheap nugget laptop, but, if u gonna pay so much for tokens on some random site to use stable, just save for a pc that can handle it

takes like 5 mins to set up stable, if ur not like me and tried chasing after an old version

i had an HP notebook pc that handled stable jus fine

costd liike 470 and had an AMD 5500U cpu and 8 gb ram

[umbral lotus]

i get what you say. I'm relatively new to the hole thing, so to be able to check out different generators and how they process models, lora's and prompts online is for me a quite convenient way to get acquainted with the technology before spending a lot of money ahead of knowing what i'm spending it on or for.

[atomic mortar]

Free? It uses computing power and electricity bruh

But the best free one imo is civitAI with the daily tokens and the fact you can use loras etc

[gritty lava]

xd

[gritty lava]

fair points :3

respectable

[umbral lotus]

i didn't say it's free. Each month i spent a good amount on subscriptions but i'm okay with that. Once i've reached a point where i know exactly what i want and need i'll start thinking about investing in new hardware and what generator(s) i'll work with. Like i said: for now it's convenient, and when the people at SeaArt, CivitAi, etc. extract some ideas from my feed while i'm at it the better. It's all for the greater good 😊

[lavish dirge]

bro

can someone add me rq

i need help with something

[atomic mortar]

with what? i got some time

[fresh ruin]

When training a lora of my own style, should i use the upscaled or original versions of my images?

[fervent thunder]

Hello

[gritty lava]

oh neat, same name as me

too bad you're not a borb

like me

u-u

[chilly gate]

👋

[gritty lava]

anyone ever get the feeling that it matters on what order you put words in the prompts

for me, I feel like it works better if you put the general stuff first, like what ur person is, what their doing, then I do cosmetics after

[atomic mortar]

i mean it does

[noble hull]

he

[untold pewter]

kinda does

[steel fable]

anyone can tell me how to create 2 differente char without mix please?

[atomic mortar]

luck and regional prompting

there was a extension that helped but yeah

was

[steel fable]

There is any video about how to use?

[atomic mortar]

i mean was as in doesnt exist anymore but you could look up the UI you use + regional prompting

on youtube

[steel fable]

U mean i use automatic1111+regional prompting right?

[atomic mortar]

i dont know. do you use Automatic 1111?

then probably so

[steel fable]

Yes

Okay, thanks i ll search here

[desert dagger]

you use a model that has the comprehension necessary to understand not to mix them. that's sd 3.5 large or flux dev/pro

[fervent thunder]

Anyone ever have issues with the xformers and getting it to utilize your gpu rather than defaulting to cpu mode because it can’t load xformers.

I ran into this issue with the dependencies and i end up going in circles installing different variations

ye I just delete xformers

this requires a certain pytorch version at least

not sure what the lowest version is that works that way but 2.4 is ok for sure

Oh okay thanks!

[desert dagger]

i just ignore the warning messages about xformers and don't worry about it

[fervent thunder]

Yeah, I’ve tested it that way and it works. I just have a decent gpu and wanted to be able to use it so I figured I had to get the xformers to work to do that

[desert dagger]

what would you be using it for inside comfy?

[fervent thunder]

I’m using automatic1111 which is probably my first mistake LOL

I just tried to pick up where I left off last year and it’s been super annoying

I had tried comfy before but I was intimidated by the UI

[narrow nova]

hello

[warm junco]

Make a clean reinstall and just add --xformers to the webui-user.bat commandline_args= line,
Then it should work

If not come to #🤝｜tech-support

[fervent thunder]

Thanks! I’m gonna try this when I’m out of work 😊

I think that’s actually what I had to do last time to get it to work as well

[gritty lava]

i knew it ùvú

smort borb

[gritty lava]

feels so gud to be back on automatic1111 1.7.0

like, ik it's just a webui version but seriously somehow, 1.10 just doesn't work like it should..it takes literally like 2 to 3 mins for it to even load weights, and then it don't even seem to follow my prompts at all, and stuff comes out looking baked a lot of the time, dunno wut it is, but 1.10 was ruining the hobby u-u

1.7.0 tho, starting it takes 1min, and stuff comes out looking great and follows the prompts very well

I think it might be that, originally stable was made for use in nivida stuff, and amd is like the after thought

and maybe 1.10 is more on nividia side than 1.7 was or sum, that's all I can imagine is the issue for me

it also doesn't help that I'm using like last gen hardware u-u xd
amd 5600g igpu

[sick hornet]

Whats the best way to outpaint

[steel fable]

How do you write something specific on an image?

[gritty lava]

u open

Twisted Brush Studio Pro

grab the artist pen, and chu write

export as png

#🏞｜general-with-images

xd

[steel fable]

Thats a problem, since almost all my models are pony 😦

[atomic mortar]

regional prompting can help but its not perfect

[rough flume]

Can I render this kind of image using Stable Diffusion?

Can't upload image. Going to post in other room

[coral fractal]

is civitia loading insanely slow and being buggy for anybody else or am i just having a wifi screwup?

[tough loom]

oopbuy？

[gritty lava]

um uh

waffles

[desert dagger]

you have to use the #🏞｜general-with-images channel to upload images

[knotty heart]

can any one tell me how can I finetune SD model on my own images ? what are the steps and what hardware requirements should I need

[atomic mortar]

Youd want to make a lora. I recommend koyha_ss but its not very easy if your not used to this

I recommend the civitAI online lora trainer but that has a slight price tag

and as always folks. scam/spam

[knotty heart]

thanks

[atomic mortar]

But generally if you want to make a character 50-100 images can work, a style like 75-150 iirc

[knotty heart]

I have close to 30,000 images

[atomic mortar]

Youd not want to use all of them as it will take a LONG time

[knotty heart]

so I have to rent gpus ?

[quartz swan]

no youd just not want to use 30k images

[atomic mortar]

Yeah, use like 400

[quartz swan]

more does not always equal better 🙏 use 400ish

pick out the best ones, preferably

[knotty heart]

I want to use SD model for inpainting patterns and text on cloths , as of currently it is not working correctly so, I need to finetune it

[atomic mortar]

Ah text, you might have stuggles with that

[quartz swan]

might?

💀

[knotty heart]

No, I will upload a pattern and with the model it will paint the pattern on the cloth , but i am not getting exact pattern

[atomic mortar]

Well yeah, thats stable diffusion

It generates something similar

[quartz swan]

i dont think 'exact' is in SDs vocab rn

[atomic mortar]

In Photoshop it is

[quartz swan]

yesss

power to photoshop

cant wait until ai makes it so i dont need to open PS anymore though 💀

if it ever gets there

[atomic mortar]

Eh its been a few years only, give it a few more

[quartz swan]

fair enough

[atomic mortar]

Flux and 3.5 can do semi coherent text sometimes

[knotty heart]

then how are they are doing it https://huggingface.co/spaces/Kwai-Kolors/Kolors-Virtual-Try-On ,

[atomic mortar]

How many devs & money do you have?

[quartz swan]

luck,img2img,prayer,inpaint,time,selling your soul- who knows

[atomic mortar]

And its not exact the same, more like 80%

[knotty heart]

one

I am starting out

[knotty heart]

[atomic mortar]

Well ill be frank with you, we get a lot of people with that same idea (clothes change) and unless you got a lot of experience, time and effort that exceeds ours its probably not possible for you atm

Unless someone will hand hold you the entire process

I say keep trying, use custom nodes in Comfy and maybe even make your own

But you will need to learn a LOT

[quartz swan]

break out google- youll need it

[atomic mortar]

A lora can make a clothing, character style etc similarities.

But you need impainting

[knotty heart]

Thanks, I was trying on comfyUI, made some progress with realVision Model but getting issues in masking , and in getting complex patterns ,

[atomic mortar]

Loras, Segmentation and impainting at a good level where its automatic

And the clothes they use is not complex (model you linked)

Its simple clothes

[quartz siren]

Many open source versions, that’s closed though. One open source version: https://github.com/Zheng-Chong/CatVTON

[atomic mortar]

oh dope i didnt know there was a open source model now

[knotty heart]

Can't use commercially.🫠

[atomic mortar]

@knotty heart if you can do comfy this could help

[knotty heart]

Yeh I know

[atomic mortar]

ah yeah. thats a problem for most people here who want this exact idea

[knotty heart]

I have seen all the workflows on openart but all are using. Flux model
Which is non commertial licence

Fine tuning or creating my own model are the only options that are left with me

[atomic mortar]

well you need a Masking generator (probably YoloV8 could do it if finetuned)

and a custom model. but i think a realistic 1.5-SDXL could work

and custom nodes

[timid sorrel]

hi

[frozen flower]

Hi everyone. I'm trying to run a workflow that would separate male bboxes from female bboxes in order to inpaint them separately using facedetailer. I'm having a hard time accomplishing this as I'm working with anime faces and the presets appear in the "hf transformers classifier provider" node of impact_pack were trained on real faces. I'm looking for any advice you can provide.

Additionally, I came across this anime gender classifier that seems promising, but have no idea how I would integrate it with Comfyui Impact_pack, any advice on that would be much appreciated as well.

https://huggingface.co/DOFOFFICIAL/animeGender-dvgg-0.8

[pine flower]

Hi I am a recruiter so I need to know about tech skill help me

[echo kite]

sucks that A1111 is dead

since it was one of the only webuis I know of to support CPU generation on AMD, even if it is slow

[atomic mortar]

i mean it still works for the models you use

with CPU generation you cant use anything better then SDXL extremely slow

iirc

[atomic mortar]

Hi, i honestly recommend watching a video as its very unlikely someone will invest their time for someone elses business

and tech skill is very broad, if you can narrow it down you might have better luck!

[echo kite]

imma try forge one last time with a lew arg flag I haven't tried before

[echo kite]

yeah if there is a tutorial out there youtube is goated for stuff like this, learned how to make a basic initramfs OS with the linux kernel that way ^^^^

[echo kite]

welp didn't work

even a1111 isn't working anymore

gonna have to just quit SD for now until I install a new OS like I plan to later

[pine flower]

Hi

[dense token]

This is the first I'm hearing of this, what's going on? Looks like there was a release last week, v1.10.1

[echo kite]

appearently that's what people are saying

and it hasn't gotten updates in a bit, doesn't even support SD 3.5

[atomic mortar]

it depends on the branch you use mostly but its not optimal. theres like a billion of tutorials out there and its not because its so good

a few branches just are broken. dont update or simply arent optimized for the newer stuff

and checked just now:
https://github.com/AUTOMATIC1111/stable-diffusion-webui/releases

[dense token]

The entire ecosystem is a mess right now to be honest. Even ComfyUI keeps breaking crap or it never worked right to begin with.
1 to maybe 5% of people at most are programmers, and the rest just want to type in stuff and get text / images / video / sound.

[atomic mortar]

Jul 27, 2024 last update by author

no idea who W-e-W is but its a singular bug fix

[atomic mortar]

check the github link i posted

its litterally there

what has that to do with anything

[dense token]

No one asked

Blocked & Reported. Enjoy your time on Discord, as short as it may be.

[karmic brook]

sorry for that yuckiness! banned em

[dense token]

It's okay. I just wish Discord had an 18+ age requirement sometimes, instead of a 13+ one. =/

[atomic mortar]

i wish discord had a minimum account age requirement setting to join a larger server

[dense token]

That would help cut down on Ban / Mute Evasion.

[fervent thunder]

there used to be foooocus

it isn't updated any more though

it was pretty close to being a box you type in to get image

[atomic mortar]

Thankfully things got a bit more advanced

But sadly old guides still exist

And make everything quite confusing for a lot of people while the process is straight forward

[dense token]

Let's be honest, Focus was never good. Any time you messed with advanced settings, they weren't saved. You had to set them back up every single time.

[fervent thunder]

I mean you're not rly gonna be able to make something that is both simple, for non-programmers and also technically good at the same time

[atomic mortar]

You know, i don't mind there being a barrier of entry

Theres so much slop being posted on non ai sites its unreal

Slop as in low quality images

Many mistakes etc etc

[dense token]

There are actually a fair few that do both, that use ComfyUI as a Backend and something else for the Interface.

[atomic mortar]

SwarmUI is one of them and it's quite good

[fervent thunder]

recommending swarm to newbies hasn't gone well most times I tried

[atomic mortar]

Theres litterally a ? Everywhere

You download a model, put it in models/stable-diffusion and your ready to go

[dense token]

Did you see that they're even using low quality slop images for Official Production Posters now? Marvel's Fantastic 4 is one of them.

[atomic mortar]

Yup, its such a shame too because a nicer image isnt THAT hard

[atomic mortar]

At a certain level of not reading the guides/manual i just recommend people to use a online generator that handles most of their needs

People's reading comprehension is so extremely low nowadays. So i shouldent be too surprised

[dense token]

This is true. We were cracking jokes about people not reading the cards in TCGs

[abstract quarry]

I always recommend InvokeAI for newbies, it's an awesome ui

although I'm using comfyui myself most of the time, I found invokeai the most intuitive ui so far

[fervent thunder]

oh I forgot about invoke, that one is nice

[atomic mortar]

I work in IT. I had a defective printer so i put on two stickers. One: "out of order" and two: "use printer in X hallway" this man walks up. Stares at both notes. Lift both notes up and tries to use it. And ask me if its broken right after

Invoke looks nice but seeing the comfy spaghetti on their frontpage worries me for beginners

[fervent thunder]

yeah its like being in Italy- you have to hide the spaghetti

[dense token]

I'm laughing at this because I've seen this very thing happen. I also made notes, it helps to tape the lid shut too.

[fallen axle]

Is Max here?

I just want to say hi to @vapid dove because I want to say hi to @vapid dove 😈

And just be a minor annoyance

[fervent thunder]

sounds good

[vapid dove]

oh hey

[woven cloud]

I would like to make new friends, can some of you please and thank you send me a friend request if you want too. I watch anime, I do sports and I love to cook!

[gritty lava]

sounds like...

a sCaM

also

[vapid dove]

I am suspicious

[gritty lava]

indeed

check 3 #🏞｜general-with-images

[dense token]

He's in another server I'm in, but it does sound bad. Apparently we're talking about two different things

[gritty lava]

kinda feel like theyre gonna try to be nice to u, then say how they got a new "gift card"

and wan u to accpet it xd

[dense token]

Those bots seem to be popping up more and more don't they?

[gritty lava]

i aplogize if the person is legit, but, the exclamation mark at the end, the fact theres a new phishig site that just appeared, as shown in general images

just red flagsd

[fervent thunder]

yeah cos people keep falling for them

same with fake stable diffusion sites

someone here the other days used one and it wasn't possible to convince them

[dense token]

Some people are just hard headed like that.

[fervent thunder]

I compared to pricing to vast.ai pricing and for the price they were charging for 1,000 images you could make like a million images

[gritty lava]

seems like they may be coming up more, due to gen z slowly turning into boomers ourself, millennials

from wut ive learnd tho, these scammers are the type of species pf scammers that are like, the bottom of the barrol

literally after old ppl

bc skill issue

[dense token]

Like Old Ladies selling Avon because they think they'll make money in an MLM scam.

[gritty lava]

xdd

btw not safe on the chrome web store either, pretty obvious tho, bc google, but, u can have ur session IDs and cookies stolen from web extensions ffrom chromes own webstore

i had one steal my steam acc session, claming to be ad blocker

after some rooting around when the hijacking took place tho, i narrowed down what had caused it, and, from checking dates and stuff, found the last thing i installed was that "ad blocker"

afer removing and changing pw, all was good

[dense token]

I'm surprised we don't have advertisements stealing passwords considering they can turn on your phone mic / camera.

[gritty lava]

all they ende up doing was selling all my steam cards or wutever

ya fr q-q

i believe we will eventually tho

as websites aquire more abilities through java and such

end up with adverts that aquire some kinda cookie

using some exploit

at that point, i might just not go on the web anymore xd

internet gets more unsafe everyyear it seems like

ud think itd be the opposite

[gilded comet]

hi

[sharp relic]

how do i download

[autumn apex]

Do you have a guide on how to do animation/videos in ComfyUI?

[untold pewter]

I'll hit you a dm

[atomic mortar]

Theres a guide in #🤝｜tech-support pinned messages. Though if that guide is difficult i dont recommend ComfyUI

[versed yoke]

@vapid dove hey do you think that stability will ever release there creative up scaler they talked about like 1y ago? or something similar?

[proud bluff]

türkçe bilen varmı ?

[rough swift]

yes

[proud bluff]

bu stable diffusion u nasıl öğrenebilirim acaba ?

midjourney gibi resim görselinden resim elde etmek istiyorum ( tasarım )

[eager shard]

Just use brave man, or the other VPN browser I forget their name

[atomic mortar]

Automod idea: if 50$ is mentioned within a minute of eachother you get auto banned lmao

[fervent thunder]

brave is likely worse than chrome for security

not for privacy, but for security

[fervent thunder]

IDK if its competitive any more

it might have a cool attention trick or something though

it looks a bit like controlnet SEG because it seems to keep the identity of objects but change the appearance fully

it could be attention map trick cos attention maps and SEG are loosely linked anyway

[dry bear]

ey @lucid bobcat , what XY Plot are you using? I'm looking for a robust one and allow me to plug any parameter.

[wet grotto]

hi, stable diffusion has a official youtube channel?

[fervent thunder]

not sure, maybe

[gritty lava]

remember to like and subscribble

and jingle that lil bell icon to be notified when stable releases another

xdd

[dense token]

subscribble... Okay, who showed you my prompts?

[gritty lava]

xD

[lucid bobcat]

I'm using a custom node salad. There are some nodes for image grids. You can provide the x and y labels. But for the individual tiles I use loop nodes and list nodes for parameters. I would loop over x and y and use the indices to pull the parameters from a list that I created before the loop.

[fervent thunder]

which looping nodes do you like?

I need to find some

[lucid bobcat]

Can't remember the name but they're blue 😄

[fervent thunder]

do nodes have colours? 😮

[lucid bobcat]

Yes you can change the color when you right click.

[frozen bay]

Hello

Anyone avaliable to answer some questions...?

[fervent thunder]

oksy

[atomic mortar]

It depends on the question lol.

[frozen bay]

https://tenor.com/view/jiseok-kim-jiseok-big-ocean-bigocean-kpop-gif-16919206117458777151

Thank you

Well I am somewhat new to AI text to image generation, outside of Midjourney of course. I find it to be a tad...boring. I'm looking for more of a challenge! So I decided to look into other options! I came across this video: https://www.youtube.com/watch?v=n233GPgOHJg

It spoke on the history of stable diffusion, and then went a tad bit into flux.

From my understanding, it appears that SD is more fined tuned and fleshed out than Flux, while Flux is continuing with it's development as SD's owners aren't currently developing any new tools.

Question:

1. Which is better to learn at this point in time SD or Flux?

[dense token]

Whichever one you want. Look up example images for both and pick your favorite. Granted you have the VRAM to do so..

[frozen bay]

2. If I wanted to take picture of my self amking a pose, then wanted to create an AI image with that pose as a reference, which would be better for that purpose?

[atomic mortar]

and TBH your mostly learning a UI. it also comes down on your hardware

[frozen bay]

(taking notes)

[atomic mortar]

both would work with a controlnet

what GPU do you have?

[frozen bay]

And this controlnet...what is that?

[dense token]

I've only seen people using image to image with SD or SDXL, I've not seen any examples for Flux but I also haven't looked.

[dense token]

Flexible wireframe skeleton you overlay on images IIRC

[atomic mortar]

hmm basically controlnet does a lot of things but for the sake of your question:
it turns your pose into a skeleton and makes the generated person into that pose

either through a deptmap or a skeleton wireframe

[fervent thunder]

as the others said you sort of learn to use the models in general rather than learning one model

[frozen bay]

Mm! Now this is something that's interesting!

[atomic mortar]

but yeah my recommendation will 100% depend on the GPU you have

[frozen bay]

So these tools basically give you greater control of the image you'll be generating through controlnet! Amazing!

[frozen bay]

OS: Microsoft Windows 11 Pro
Processor: AMD Ryzen 9 5900X 12-Core Processor, 3701 Mhz, 12 Core(s), 24 Logical Processor(s)
Virtual Memory: 22.4 GB
Physical Memory (RAM): 32.0 GB
GPU: N-RTX3080

I'll be upgrading to a 5090 soon.

Based on the specs, which model or tool would be best?

[atomic mortar]

a 3080 is pretty good. yeah you can run flux and SD 3.5 pretty well

i personally use SDXL since im on a 3070TI

[frozen bay]

Now about that. Is 3.5 better than SDXL?

[atomic mortar]

Flux is nice for realism. SD3.5 i havent seen a lot

[frozen bay]

Or XL

[dense token]

I'm currently on a 3060TI and SDXL still works wonders, even at generating 4K images. And yes I mean generation, not upscaling.

[atomic mortar]

XL is older

but still good

[frozen bay]

And I can do everything on 3.5 that I could do on 1 and XL?

[atomic mortar]

hmm it really depends on what you want to do

[dense token]

bit of a learning curve on 3.5 tbh

[frozen bay]

Hm

[frozen bay]

?

[atomic mortar]

SDXL has been around longer so theres more resouces avaible. but SD3.5 is newer and in a lot of things better

such as characters. styles etc

[frozen bay]

I see

Do I have to download two different things to use either?

[dense token]

If you mean UI, no. ComfyUI can handle everything you throw at it, or will be updated to handle it.

ComfyUI isn't for the casual user though...

[frozen bay]

Isn't that the interface that looks like some type of mindmap?

[dense token]

lol, sort of yeah. I take it you've used X-Mind before?

[frozen bay]

I have no idea what that is (yet) lol. I just use mind maps, like https://mural.co/, for my work XD!

[frozen bay]

OH! This looks nice!!!!

[dense token]

Xmind was the first brainstorming / mindmap software I had ever used, back when it was completely free.

[frozen bay]

May have to try this out honestly!

[dense token]

Back to this question, SDXL and Flux are two different types of models, so yes you'll have to download multiple things if you feel like trying them out.

[atomic mortar]

ComfyUI is really complex but SwarmUI makes it a lot easeir if you want to be able to run most models

since it also uses comfyUI in the back but has a simple front

[frozen bay]

I see. Do you have a good place to start when it comes to properly installing these? There are so many youtube videos out here, I have no idea where to start

[dense token]

I don't, sorry. I just kept trying until I succeeded.

[atomic mortar]

if you do decide to use SwarmUI i recommend using the official installation method:
https://github.com/mcmonkeyprojects/SwarmUI?tab=readme-ov-file#installing-on-windows

If you decide to use other methods. i recommend the tutorials of CS1o.
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[dense token]

For an alternative to what he said, check out Lykos Stability Matrix. I keep forgetting about this.
https://github.com/LykosAI/StabilityMatrix/

One stop shop for all the UIs, and also includes a CivitAI model browser

[frozen bay]

Oh whoa!!! LMBO!!!

I guess I asked for this lol

Thank you all! Guess I have to do some research!

[uneven axle]

This server is only for A1111 right? no comfyui?

[fast sage]

for flux, does anyone have a upscale example? Or what should I be using?

[dense token]

There's #🤝｜tech-support if you need help and an actual #🧣｜comfy-ui channel for devs / other help I guess.

[uneven axle]

Didnt' see comfyui channel sorry

[dense token]

How about no. *blocks you*

[ruby parcel]

is openpose only 2d?

as in it doesn't know how far or close the hand marker is etc

[atomic mortar]

you could try to use depth midas etc for that

[dry bear]

Could you please share it with me?

[ruby parcel]

i see

means i need a openpoese editor for the 2d, then i would need to edit a depth map for the hands whether it is further back or front?

[atomic mortar]

Nah you can depth map directly

No open pose needed

But it depends on your use case

[ruby parcel]

hmm

i'm thinking something along the lines of eidting the stick pose in 3d space, then i get the "2d sticks image", "depth image of the corressponding body"

so that i can input to 2 controlnets and do something like "hand infront" "hand behind" kinda images

[atomic mortar]

Hmm upto you but it handles it pretty well already

[ruby parcel]

may i ask a

noob question

hmm

kinda like a workflow question

let's say i have a general idea of a scene in my head.

i have an ipad with pencil, which i can kinda scribble with. i can't draw shyt... but i can scribble general idea, maybe the head is here, the hands are here, the sun is there.

from here on, is there an advice on what process should i follow to make it go through SD and combine with prompts make something out?

i know there's img2img. hmmm. but this is a scribble.

[slim zinc]

Is Nvidia still the main brand you need/want for SD?

[woven cloud]

Hey guys, I would like to make some new friends, can some of you please and thank you send me a friend request if you want too. I watch anime, I do sports and I love to cook!

[atomic mortar]

yes

AMD works but you run into a lot of jank at higher end models etc

[slim zinc]

Ok thank you. Maybe i am lucky to get one 🫠

[sudden schooner]

im looking to hire someone adept at stable diffusion, who also understands traditional art, to create a model and prompt style for me.

More info in dm. Please send me one if you meet those requirements. This is paid well, and will be a fun project.

[dense token]

Why not just do it yourself? It would be cheaper & a fun learning experience. There are models trained solely on paintings after all.

[lucid bobcat]

Guys like him are fishy af. His twitter is nothing but krypto tweets.

[atomic mortar]

what's the pay and 50% upfont you'll have me for 2 days

[dense token]

That's why I told him to "f*** off" as politely as I could.

[atomic mortar]

did he dm you lol?

[dense token]

No, no DMs here. Guess I didn't give him the answer he wanted.

[atomic mortar]

hes not responding so i assume its a broken bot

always love to see what they have planned lmao

[fervent thunder]

create a model and prompt style for me.creating a model and creating a prompt style are like

two very different difficulties 🤔

[atomic mortar]

Yeah but keep in mind, if legit he knows nothing about stable diffusion so its a reasonable mistake

[fervent thunder]

okay yeah true

[lucid bobcat]

Yes but isn't it weird that someone who knows nothing about image generation wants to start with a fully customized model.

[fervent thunder]

true yeah

the lowest possible price to train a decent model is probably super low now, to be fair

so long as you were willing to refine it with existing apache 2 models

[glass lily]

Guys my time after switching from normal webui to Forge went from 50-60minutes to 26 minutes on upscaling a 3K image to 12K. I'm shocked

[atomic mortar]

yeah forge is better then a1111 and why the hell 12k images

[glass lily]

Big number leads to neurons activating

[dense token]

You shouldn't be that shocked. Forge uses the ComfyUI memory management stuff, and A1111 refused to update his backend, so anyone using A1111 has to suffer.

[glass lily]

Why doesn't Deforum show up?

Not even in the UI Tabs setting

[atomic mortar]

Deforum? is that from a extension?

[glass lily]

Yeah the little video maker

[sudden schooner]

Have been doing it myself, and got somewhere. But I’m also looking for an expert to try. Might produce something I didn’t think possible or couldn’t do myself. Im@not an expert with it, but I can pay someone who is

[unborn hedge]

so whats the hype around illustrious? is it just a really good model??

[glass lily]

Yeah it's the real deal

[dense token]

@Mods We have a DMing bot in here=/ @chrome zenith account is jenniferamanda_ in case they try to run

[outer sedge]

Hey guys, what do you all do for a living? Are you interested in communicating and doing something together?

[dense token]

No^2

[outer sedge]

very boring

[jaunty surge]

Hi guys, any huggingface models that I can use to change the design of a shirt?

like I would like to apply a Texture to a white shirt

[dense token]

You can do that with any model and an alpha mask =/

[jaunty surge]

havnt tried the general models though, I am looking for finetuned lora adapter to do so

and also local image editing as well: applying a mask to the shirt neck area and apply specific pattern or texture

I know general models can do this, but I am looking for a specific trained lora adapter for this

much appreciated if you can give some recommendations

[dense token]

Your use case is so specific that Google will probably be able to help you better than we can.

[jaunty surge]

will search on google as well

[karmic swallow]

hi

[atomic mortar]

Oh wow another one! Commercial use?

There's one but it's not allowed for commercial use

[jaunty surge]

just want to try it out for research use, not for commercial

[atomic mortar]

Here you go @jaunty surge

[jaunty surge]

oh no, its a virtual try-on model

I am looking for local editing model for cloth though. Anyways, thanks for the link

[atomic mortar]

Ah well just inpainting then i suppose

Maybe a yolov8 for automatic segmenting The clothing + a comfy workflow to just paste a design

[old lichen]

And here's a question, has Hunyuan3D-2 been deployed yet? Who knows can advise solutions for 4090rtx?
https://huggingface.co/spaces/tencent/Hunyuan3D-2/tree/main

[dense token]

People are using it with ComfyUI, simple Google Search will get you where you need to go. And apparently it doesn't need much VRAM

[mighty zealot]

is anyone able to give me some prompt advice? I'm trying to create flat minimalist iconography that are in the from of drinks, such as coffee, pint of ale etc etc however I feel like I'm just typing random crap in at this point lol

oh wrong channel, oops

[atomic mortar]

Flat vector art?

Nah those questions are fine imo, it keeps general busy as #📝｜prompting-help is hella slow

[mighty zealot]

yeah basicaly

[raven perch]

i already bought a **french **acc with Egyptian followers guys, it's 70k and already accepted in **CRP **so it's ok to use **uk **proxy in this acc to get high RPM or what do you think guys??
and it will get disqualified?

[pale berry]

Is there a fast tutorial for a newbie?

[atomic mortar]

what are you on about?

[atomic mortar]

Hi yes, though i have to ask. what GPU do you have

b ased on that i can give you a solid recommendation

[pale berry]

core GPU

[atomic mortar]

ouch hmm well i do recommend a Web based solution by either renting a GPU online or by using limited free credits / by purchasing credits on CivitAI

depending on what you want to make i think the CivitAI option is going to be the most cost effective and easiert to get into

[bronze cove]

New to SD, just installed last night, using dreamshape 8. Anyone know why I keep getting extra arms and legs in my images? I even put "extra arm" and stuff in the negative prompt. I don't know what I;m doing lol

[atomic mortar]

Dreamshape 8, what model is that?

SD1.5? XL?

[bronze cove]

I have no idea what I'm doing, but it says sd\dreamshaper_8.safetensors

[atomic mortar]

Wheres you get it?

[faint timber]

Make sure your size does not exceed 512x512, the size that Dreamshaper8 was trained on.

[bronze cove]

Stable matrix module browser

[atomic mortar]

Ah its 1.5

[bronze cove]

Oooh, that's helpful.

Thanks, Atom

[atomic mortar]

Yeah that and 1.5 is a older model

Try to stay within 512x512 or its aspect ratios. 3:2 or 2:3 if you want a portrait/landscape

[bronze cove]

What is best model for making fantasy character art? Like, d&d style stuff, realistic, highly detailed, but not too realistic.

[atomic mortar]

Theres no BEST but preference. What gpu do you have

I got a few favourites per release

[bronze cove]

4070

[atomic mortar]

Oh lol, ditch the dreamshape 8 and hop over to sdxl/illustrious

Or maybe even flux

[bronze cove]

let me see if I can find that

[faint timber]

Also, there are small little addons called "embeddings" One of the popular ones is called "easy negatives" This embedding can help with extra feet/hands etc, but the SD 1.5 models are flawed from the start.
I'd suggest immediately abandoning SD15 in favor of SDXL or better if you have a decent GPU (i.e. 8GB or more)

[atomic mortar]

Check on civitAI for models + examples

[faint timber]

I've been getting good results out of the Proteus model on Civitai. It claims to be trained on copyright-free images.

[atomic mortar]

I'm in a rush to walk the dog but yeah i recommend what atom says, a 4070 can handle most models

[bronze cove]

Hmm, I don't know if I;m finding this SDXL when I seearch for it in the module broswer

[faint timber]

Search on Civitai.

[bronze cove]

Is that a website?

[faint timber]

Yep

[bronze cove]

Oh

[atomic mortar]

Also in stability matrix i recommend using either forge or swarm

[bronze cove]

Well, that's where I am I think

[atomic mortar]

Civitai.com = nsfw + sfw
Civitai.geen ,= only sfw

Nsfw only shows up if your logged in + enabled it in filters

[bronze cove]

hmm... So I shouldn't search for modules in the module browser on stablity matrix, or is that what I should use?

Sorry, my brain is really smooth

Cus when I search "Forge" for example, there's like a dozen and they all say forge in their titles but say other stuff too

[faint timber]

Try pulling down this model and placing it in your ComfyUI/Models/checkpoint folder.
https://civitai.com/models/267242/proteus

[bronze cove]

Trying to understand, so download from that link, and then stick it inmy checkpoint folder. Looks for folder

Okay, it's downloading.

[mighty zealot]

so more reading on my side, I'm using the base SD3.5 medium and I assume this is probably not the best for vector art?

[bronze cove]

Hmm, sadly I can not locate a checkpoint folder

Found a modules folder though.

I think I will put it inthe same folder that my dreamshaper is located, that makes sense, right?

That seemed to work, it showed up on stability matrix

So far this module you linked is making nothing but very poorly drawn images.

Like burry cartoons. I'm confused.

[atomic mortar]

what is the rpompt your using?

a prompt like a "cat" wont give solid results

[pale tiger]

recommending swarm to newbies hasn't gone well most times I tried

[bronze cove]

I tried things like, "Hyper-realistic, very detailed, high fantasy image of a female barbarian" Just as a test, but it gave me stuff that looked like a crudely drawn, smeared, picture.

[atomic mortar]

it works OK but the newbie requires to have reading comprehension since everything is really clearly explained by a ? next to the setting

[atomic mortar]

it does depend on the model. what settings you need

seeing how the linked model wants prompts like
"cinematic film still of Kodak Motion Picture Film: (Sharp Detailed Image) An Oscar winning movie for Best Cinematography a woman in a kimono standing on a subway train in Japan Kodak Motion Picture Film Style, shallow depth of field, vignette, highly detailed, high budget, bokeh, cinemascope, moody, epic, gorgeous, film grain, grainy"

and a negative of like:
worst quality, low quality, deformed, malformed, distorted, blurry, unfocused,

Guideance at 7 and the sampler: dpmpp_3m_sde_gpu

[bronze cove]

Yeah, that's hard for a noob like me.

[atomic mortar]

hmm i got some time now, what UI are you using right now?

Forge?

[bronze cove]

I deleted them all atm, downloading something new.

I don't have a lot of time atm though, but thanks for the offer. I was recommended forge, but I couldn';t find it. When I use the browse module option and type forge, it gives me a bunch of stuff with out ratings so I wasn't sure if I should trust any of them.

[atomic mortar]

hmm what i recommend is NOT using stability matrix and following the guide of CS1o (hes in this server).

https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[bronze cove]

Rightnow I;m downloading something called sd35fusion8

[atomic mortar]

first time ive heard of it

youd want the [NVIDIA] Forge-Webui

[bronze cove]

I have no idea, just clicking stuff in the browswer for stability matrix

I'm using stability matrix cus it was a one click install and my smooth brain can't handle anything complicated

[atomic mortar]

well with AI stuff regulairly updates so its handy to get a bit of the know how

[bronze cove]

Can you not make good images with stability matrix or?

[atomic mortar]

you can but as you have noticed. sometimes too many options is just confusing for a new person

[bronze cove]

Can't someone just suggest one of the best options on stability matrix's module browser?

Is there a module out there that makes high quality art without having to add a ton of complicated prompt tags?

[atomic mortar]

... no

its just how AI works

[bronze cove]

rip XD

[atomic mortar]

but

heres what you can do

Put this in the negative prompt always:
lowres, worst quality, worst aesthetic, bad anatomy, old, early, scan artifacts, scanlines, adversarial noise, lossy-lossless,

Put this in the postive prompt:
best quality, good quality, very aesthetic, absurdres, newest, HDR,

then after the HDR etc type your own stuff

its not the greatest but as you get used to it you can edit it as you like

[warm junco]

Fooocus can do that, its basic prompting and selecting a style.
But you can't configure much

[atomic mortar]

hmm i forgot about fooocus

but probably because its so limited

[bronze cove]

Thanks guys.

Can I find fooocus on the stability matrix broswer?

[warm junco]

I think so

[bronze cove]

Hmm, I typed it, 3 O's, no results

[warm junco]

Should be in therr

Its listed on their github

[bronze cove]

I found one, but it says it wont let me get it without an account

Need an api key. Rip

[warm junco]

Not needed for fooocus, I guess your on the wrong track

Maybe dont use stability matrix but try to install Fooocus using my install guide

[bronze cove]

I don't think I'm smart enough

[warm junco]

Its not to difficult:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides
For any questions feel free to ask in #🤝｜tech-support

[bronze cove]

It's weird, I'm looking at civitai.com and looking at pictures made using specific models, I have the same model installed, I'm copy and pasting their prompt for the image into stable diffusion, but the image I get looks nothing like the one they get. Like, it's just such bad quality complared to the examples they give of what you can make with their module, I;m so confused.

[sudden schooner]

which is best for mac? Not automaticq1111

[leaden pendant]

hey guys!
I'm new here, but I had a question about my stability.ai API key
do the credits "regenerate", like, do I have a fixed limited free credits, or do they come back over time?

[quartz siren]

Your settings might be different, what’s the img output and settings?

[atomic mortar]

^ and vae?

not to mention a different UI (background workflow) is different so it could produce different results based on that too

theres one that yeets it thru a llm but i forgot witch one

[dense token]

Most people use ComfyUI with all sorts of nodes that highly affect quality, like HyperTile, Kohya's Hi Res Fix (the real one, not comfynonymous' absolute crap node), FreeU Advanced, etc.

Just having the prompt, model, lora, and negative embedding isn't enough in this day & age.

[leaden pendant]

hey guys, the docs on the website say that I get 150 requests every 10 seconds, but I only got to make 3-5...
like, at all, it won't let me use it anymore

[dense token]

Don't let it get you down @bronze cove. Look at it as a learning experience. "There's much better images to be made, what do I need to do in order to make them?"
*cue the training montage*

[atomic mortar]

i assume the API? i recommend reaching out to their helpdesk

[bronze cove]

There's just so many things, I don't even know where to start, and when I try to watch a video or read something that tries to explain to me what to do, I don't understand it. Rip

[warm junco]

Forge or comfyui

[dense token]

I'm not going to lie, a lot of it is just trial and error. "If at first you don't succeed, never try again try, try again"

[bronze cove]

Ya, i'm tweaking all the setting options it gives, me, trying to figure things out, So far havn't learned a thing though lol

[dense token]

Messing around with settings is best done with only one or two options at a time, that way you can see what looks best in your eyes. Messing around with too many settings at once can often backfire and make you feel frustrated that you're getting nowhere.

[bronze cove]

Yeah, I change one thing, then generate an image a few times. Then I change something else and repeat

[dense token]

That's all you can do. I've probably burned through at least 10k images learning how to make this or that style, what settings gives the best details, etc

[bronze cove]

I think I'm starting to get an image that looks okay

I typically use bing's free image generator, it's quick, free, unlimited uses, and makes some gosh damn amazing art. Really hoping I can get something at least somewhat decent from SD, it's frustrating when bing is so much better and easier.

[dense token]

Oh, I thought you were using it on your own PC. I'm not familiar at all with the online image generators.

[bronze cove]

Oh, I'm trying out SD on my PC, yes, trying to learn how to use it.

But I also use bing on my phone, fast, easy, free, amazing art.

[quartz siren]

What model are you using? Flux/sd3.5 should be similarish quality/better? and definitely better with Lora’s and control stuff.

[bronze cove]

Problem is, everytime someone tells me which model they think is best, I can't seem to find it or it requires a key to download.

Flux required a key

[quartz siren]

Yeah sd3.5 large and flux dev have the agreement thing because you have to abide by their licenses(flux dev is no commercial use, and sd3.5 is commercial use if less then 1mill revenue, you can do pretty much anything with both if not commercial)

You probably shouldn’t download the original models since they won’t fit in most gpus, only with quantization.

First step is what gpu do you have?

[bronze cove]

4070

[quartz siren]

Alright so 12gb vram.

First part is, what model do you want to pick? Pretty much flux dev is best humans, text rendering, prompt following but some strong biases like bokeh, plastic like skin. Sd3.5 large is also good at above things but less, better choice for artistic styles and has a bit more knowledge.

There’s also turbo versions of the models which allow you to gen images faster but slightly worse. For flux, it’s flux Schnell and a good choice too. For sd3.5, it’s sd3.5 turbo but that’s kinda meh imo, I don’t recommend it since it’s slower and worse than flux Schnell.

you can try them out here if you can’t decide: https://huggingface.co/spaces

[bronze cove]

Idk which model to pick, I can't try flux cus of it requiring a key, I could try sd3.5, but I doubt I can find it, I can't seem to find the models people recommend in the stable matrix browser.

I'm looking to make semi-realism high fantasy character art with hyper-realistic detail, but not life-like.

I fond sd3.5, also requires a key. Gosh dang.

Why do the good ones all require keys 😭

[fervent thunder]

flux fp8 is 11.9GB so that will fit

[bronze cove]

But flux requires a key >_<

[fervent thunder]

what is a key

you can just download it

[bronze cove]

When I click on it to download, it says I must enter a key code.

[fervent thunder]

where is this?

the official flux is here

https://huggingface.co/black-forest-labs/FLUX.1-dev/tree/main

[bronze cove]

I'm using the stable matrix model browser

[warm junco]

start with sdxl ones to get into the basic stuff

flux or sd3.5 are not recommended to start with

[bronze cove]

I don't even see a download button on that page.

No idea what to click

[dense token]

Do you even have the accounts connected under settings in Stability Matrix?

[bronze cove]

When I try, it asks for a key

[dense token]

It also gives you instructions on how to generate an API key when you click connect.

[bronze cove]

Oooh, I got it, got it to work. Doh, I thought it was asking for a key, like a cd key, like something you have to pay for.

Thanks!

[fervent thunder]

if you see an arrow pointing down to a line
that means download

this is for almost all software BTW

sometimes will just be a down arrow without a line, as a variation

[bronze cove]

Ya, but there are arrows pointing down on like 12 things

[fervent thunder]

Flux is multiple parts

[bronze cove]

I wouldn't even know how to install it

[fervent thunder]

essentially Flux is four models not one

[dense token]

You're welcome. If you have more issues, just ask. As far as you wanting "semi-realism high fantasy character art", I recommend DucHaiten-Journey-XL

[bronze cove]

Thanks, I think I found it. Duchaitenjourney

v6.4.25

[dense token]

No, that's the SD 1.5 model, what I am recommending is the SDXL model, v1.5.

[bronze cove]

Oh, well I don't think it's in the browser.

[dense token]

You might just have to get it off the civitai website itself. DucHaiten might have flagged his model as NSFW, so click the "..." beside the magnifying glass and Check "Show NSFW Content"

[bronze cove]

Ya, but it's above my brain power to install that kinda stuff.

[dense token]

It's a lot easier than you think it is. Just download the model, drag and drop into the right folder.

[bronze cove]

Witch craft!

[dense token]

Stability Matrix has a Checkpoint Manager too, after you download stuff.

[bronze cove]

If only I knew what that meant or what it did

[dense token]

When you first downloaded Stability Matrix it asked you where all your models are, right? You just dump them all in one folder, SM does the rest.

[bronze cove]

I think I can manage that

[rigid warren]

What's up

I'd like to get into Lora training so I want to ask what's the best method to do it. I only want to create anime characters with Illustrious. Can I do this with 12 gb of vram? Where do I train the models?

Thanks in advance

[clear oyster]

when does SD 4 come out?

[gritty lava]

would have to be a time travler to know dat one

xd

[young bronze]

You'll have to wait for word from SDis

[karmic swallow]

Man plans, and God laughs

[dense token]

Story of my life

[nimble pendant]

hello, do you know if there is a script or something else to sort through loras?

[atomic mortar]

? Like what the lora is for?

Probably not if theres no meta data that specific. I recommend making sub folders in the lora folder

[nimble pendant]

I want to sort it out because I have too many. There are new versions and old ones. I have so many that I was wondering if there was a quick technique.

[atomic mortar]

Well sort by date and delete the old one? Theres no hack/script to do it for you

Since the files are made by many different people

[nimble pendant]

thanks

[odd fossil]

You can use self-hosted Stable-Diffusion sites like Tensor, spend very little money, and train the LoRA you want.

[odd fossil]

Which model.

[modern pagoda]

🤔

[median jewel]

does anyone know how sam in fooocus works? i just provide a picture, provide a text it can create a mask out of the thing i put in prompt. So for example if ahave a pic of a dog in grassfield and i want to cutout the dog and replace it with cat i can just upload the picture to inpaint select sam and write in prompt "dog" and it creates a mask of the dog. I am trying to understand how to get this technology and use it in a python project i have that doesn't inlcude fooocus in anyway is it possible?

[abstract quarry]

segment anything creates the masks for you

it basically splits the image into segments. You can also provide coordinates of points it will then give you the segment the points lie in

however , sam cannot do prompts

you need something else for that

I think there exists several models that can do text inputs and which you can combine with sam (e.g. DINO, grounded, ...)

[potent otter]

hi all, using core model, just started experimenting with api. When i give same prompt, but even with different or random seed, im getting very similar image back, so for example 3 images of the same character on the same balcony, slightly different camera angles 😀 kind of useful but not expected. What do you think i am missing?
{
"prompt": "a sci fi scene similar to blade runner with an android sitting on edge of tall building top",
"output_format": "png",
"style_preset": "cinematic",
"aspect_ratio": "16:9"
}
i expected with a random seed completely different images.

[thorny phoenix]

Is it possible to make videos with easy diffusion?

[dense token]

According to the issues on the Github repo, no.

I guess some people don't understand what SFW means. =/

[desert dagger]

@vapid dove

[atomic mortar]

oh creative. spam discord link in bio. dont click chat

just ignore "ambrose" and move on

[dense token]

You can report the User Profile for those types of people.

[pale berry]

hello

Is there for the free chat?

[desert dagger]

no

[atomic mortar]

Free chat? Like free gens or requests?

[fervent thunder]

yeah I think that is what they mean
the bots here (in artisan channel) are paid though

if I were them I would give SD 1.5 hyper gens away for free

as a hook

[atomic mortar]

If its an interesting requests i sometimes do a few but post them else where but mostly nah

As always chat, spam

[fervent thunder]

in my experience if you do that
its difficult to then convince them that you are not a bot LOL

[atomic mortar]

Nah if someone asks like that i would but people who drop a message like "a cat" get a copypasta

[fervent thunder]

IDK I would have agreed in the SDXL days but Flux does fine with "a cat"

[still glacier]

Interresting idea, but I think it might send the wrong impression. (Some) people would then compare 1.5 gens to latest versions of midjourney, flux, etc... (I'm not part of stability crew, I have no words in those decisions)

[fervent thunder]

oh actually that's a really good point yeah
you would not want someone to see SD 1.5 Hyper and think it is SD 3.5

this issue affects ones like OpenAI a lot cos people thing GPT-4o-Mini is their best model

[still glacier]

gpt naming is .... something

[fervent thunder]

I noticed ChatGPT has switched to giving a few O3-mini answers each day
probably for this reason lol

yeah even Sama does not like the name apparently

but he said they are stuck with it

[pale berry]

I`m a real human ,Sir

So this is a atrisan channel

[low moon]

Only a a robot would need to reassure us theyre human

you sir is a fraud

[pale berry]

haha

interesting

robot do not use big Alpha

soory, my english is limited

[fervent thunder]

its ok

I don't know what big Alpha is

[pale berry]

big letter

how to say A & a

what is the difference

[low moon]

Capitalism.

[pale berry]

Cool

thank you

So , what is the topic usually here?

art?

art can`t be mesured

[low moon]

nah

were far from that

were just glad WW3 doesnt happen any day and crawl up in a ball and cry every night

[fervent thunder]

Fellas

Is 4060 enough for xl models?

[pale berry]

That`s hard to say ,Madam

[fervent thunder]

Btw, has anyone tried soldering additional vram units on their gpu?

To increase SD experience

[frozen bay]

Question:

I was following @dry rock tutorial on Youtube and Pateon. All seemed to be going well till I opened ComfyUI and ran into an error upon open, "Model Error! Download checkpoint..." I wasn't able to copy the entire message because my powert went out. But I was able to download the file it wanted me to: v1-5-pruned-emaonly-fp16.safetensors, however, I'm not sure where to put it.

[fervent kite]

anyone know if there's gonna be an XeSS optomized sampling method?

[atomic mortar]

its enough yes

[atomic mortar]

you joker lmao, cant be done

[raw vector]

it's possible but difficult, what are you trying to do exactly

[fervent thunder]

I want 4060 to have more than 8gb vram

[raw vector]

Lmao good news

That's one of the few GPUs that allow for addition vram

[tight valley]

hello, i have a question. I made wildcards that activate one after another in line. Using dynamic prompts code i made variables that are defined at the start by a user and then are changed during the wildcard process giving me some sort of story prompt. right now i need to call wild card to create a prompt and then copy past it to S/R to create image. is there a better method? maybe some sort of script or extension for it?

[barren dust]

hi everyone, I was wondering if its possible to generate instrumental melody/music with an input from a single person voice, singing a song just with solo voice? does anyone know if I can do it with stable audio or any other software/app?

[dense token]

That's a good question, I'm also curious what the answer will be. I've been interested in trying to use "Bonne Justice" for this.

[fresh echo]

Hey friends,
I want to learn to work with stable diffusion how can I do ?
specifc video that I need to watch ?
Thankss

[desert dagger]

there are a lot of good tutorial videos on youtube

[fresh echo]

But I want to learn to code with it
No with the UI

[desert dagger]

okay well, i'd be asking for that sort of help on the L3 discord or the Banadoco discord, not on this one. this one's mostly users that are interested in the UI. but @fervent thunder could maybe point you to some programing tutorials

[fresh echo]

Oh well thank you very much 🙏🏻

[barren dust]

bone justice?

[dense token]

No, "Bonne Justice". It's a French poem by Paul Eluard.

[rigid warren]

any advice to inpaint good hands?

[atomic mortar]

A good prayer but generally just a segment/mask over the hand does the trick

[atomic mortar]

What model (version) are you using?

[rigid warren]

Illustrious, mostly generating anime images. most of them got pretty good hands but this character I'm doing now doesn't

I'm working on forge

[atomic mortar]

Hmm yeah illustrious normally has decent hands, yeah it's either a keep trying or inpainting

You could use a controlnet openpose + hands maybe

Since that marks out where the fingers should be or a depthmap midas or another one to get more accuracy

[rigid warren]

I tried with adetailer, but the detectors I used don't fix them

[atomic mortar]

I recommend manual masking tbh

[rigid warren]

do u know where i can find a good model?

for openpose

[atomic mortar]

Hmm civit ai has them for illustrious

Id check on my pc but I'm currently not at my desk

[rigid warren]

i think i found one on other website, thanks

[rough flume]

Where can I find the stability ai playground (not google codelabs)?

[fallow hamlet]

hi

[marsh kayak]

Hello!

[dense token]

Sorry @vapid dove, but we have another one. :(

[still glacier]

yeah don't click those "support discord server"

[fervent thunder]

Illustrious is anime so the hands can fool the detectors
you could try fine tuning a vision model potentially

it would be helpful to use a method that takes into account of the fact that hands are at the end of arms

this sounds obvious but vast majority of methods for hand detection do not do that

you can take an openpose skeleton as a guide and start your vision model search where the arms end

[open umbra]

I hate linux its literally the worst OS I've ever used

[fervent thunder]

same, apart from all the others

[atomic mortar]

I love linux its the best OS ive ever used

Wish i could make the switch fulltime though

[neon hedge]

I love Linux

[white sinew]

Bonjour, i'm eos!!!

[fervent thunder]

salut

[vapid dove]

hola!

[fervent thunder]

if they don't wanna release it
even just them giving a hint about how it worked would be appreciated

I've always been rly curious about what SD3 Ultra was as well, around the time of SD3M release
I know it was a comfy workflow based on SD3L 8B but not sure beyond that

it had higher high frequency detail than regular SD3L 8B

[desert dagger]

it still exists - i still use it

[sour solstice]

Hi, I’m curious about the current state of Stability. Are there any updates, current projects or future plans that can be shared? Haven't heard anything for a while 😦

[vapid dove]

Nothing I can share yet, hopefully some stuff soon

[sour solstice]

Understand, but they are still cooking something, right?

[neon hedge]

does anyone have any good dog breed image datasets other than stanford dogs dataset?
i already have that one (20k images) but i need more data. it needs to have image of dog + dog breed

[eager mauve]

hey how long does it usually take to generate an image for yall

[fervent thunder]

Using my gtx 1660s it would take usually around 1 min

Soon i ll get 4060

And i guess it ll shorten to half of second

[eager mauve]

1 min? god damn

it be taking me a whole 20 mins

lags too

[dense token]

What size? 1280x720 Kemonomimi just took 7 seconds on my 3060 TI

[eager mauve]

830x1210

[fervent thunder]

Why the hell you choose this resolution

Updcalers work with lesser

[eager mauve]

hmm can you send me your settings

[fervent thunder]

At the moment i deleted sd

But i usually use 576×576

[eager mauve]

ight ill try that then thanks man

im just dabbling since im new to this 😭

[fervent thunder]

No problem

[fervent thunder]

there will be a way to get it faster probably

not always but mostly

[dense token]

Just ran another test, 1920x1080, 32 steps, 31 seconds on my 8GB of VRAM 3060 TI

No upscaling or detailing needed, pure generation.