#💬｜general-chat

1 messages · Page 178 of 1

dull junco
#

When you wait 260 seconds for a generation and they grow a second head.. feels bad

kind yoke
#

How can I speed up stable diffusion and can I save the images that generated ones?

celest oar
#

do i learn tech or video editing if i wanna do entrprenuership?

atomic mortar
#

Really asking the same questions over and over huh

tight jasper
#

I was about to ask a question about img2img but suddenly I figured out the answer right after finding this thread just by mere coincidence. is it possible someone redirects me to a thread or external discord servers where the discussion on combination of params for generation is discussed?

dull horizon
#

Hi, I have a proposal regarding marketing and business. Since you don't have a ticket system, may I ask who I should contact?

#

@still glacier

#

@bleak matrix

#

@jade wren @odd cloud @hidden dagger

still glacier
#

This discord is meant for public/community usage.
For direct contact with stability team it s probably best to go through their mail.
https://stability.ai/contact

kind yoke
icy goblet
#

how do i generate ?

slender vault
#

go to tech support and check the pins for guides

gentle prairie
#

really dumb question, i assume the answer is yes, but if i close the web browser while i'm generating images, the genertion is still working right? Since the CMD is still open

still glacier
spark trellis
deep narwhal
#

what UI are you guys rockin for local generation

burnt copper
deep narwhal
#

nope, sorry

burnt copper
#

K

atomic mortar
deep narwhal
#

is it easier to use than comfy? 😭

atomic mortar
deep narwhal
atomic mortar
#

yes in the server configuration tab

gentle prairie
#

these prompt are making me go crazy... was doing fine this morning, had the effect i wanted, now trying it again tonight, doing the same prompt (i copy pasted it), and it refuses to give me the results from before, i don't get it

copper crystal
#

when all else fails, symlinks

deep narwhal
copper crystal
atomic mortar
deep narwhal
atomic mortar
#

No that's all of them. Many features/extensions in forge are either already in swarm or got a comfy backend fix iirc. But if there's something really missing you could hop onto the swarm discord to ask on how to do a specific thing

#

What are you missing?

deep narwhal
#

forgeUI had this cool prompt search and replace script where you could run multiple generations back to back replacing part of the prompt automatically

#

i could live without it but it was a cool feature 😂

livid gale
#

guys
is bluesky popular in us and europe

copper crystal
#

and grids

deep narwhal
#

thanks ill check em out

oak current
#

anyone use pioniko here?

#

idk my pio use fp8 only .. i download fp16 and delete fp8 for replace but cannot run.

eager grotto
#

when training an image ai would uploading just comic strips be okay or would it end up having the ai make a mess of an image

copper crystal
#

training set is 90% of the effort for what the model becomes. it'll learn to draw comic strips then. It'd be better to get high quality crops of each cell

#

depending on your goal

eager grotto
#

a specific artist's artstyle

lucid bobcat
#

if the artist is famous chances are there already is one

eager grotto
#

lora?

#

im using civitai

lucid bobcat
#

have you checked if there is a lora for that artist?

#

a lora is like a plugin for a full model

eager grotto
#

there is but it still seems to have that generic ai look to it

lucid bobcat
#

what's the artist's name?

eager grotto
#

jlullaby

fervent thunder
lucid bobcat
# eager grotto jlullaby

You should try "Jlullaby style PDXL, Illustrious" Maybe with the right lora strength and good prompting you could get the style you want.

eager grotto
#

so would just comic strips still confuse it and have it end up making a mess of an image or would the civitai trainer recognize the details from the body

lucid bobcat
#

"Trained on art by Jlullaby, got some help with this one ;getting rid of as much text as possible, specially watermark logo."

eager grotto
#

well i just dont know how many images they uploaded to train it because the most images used the better right?

lucid bobcat
eager grotto
#

im on mobile how do i go to my generated image?

#

nvm it was just laggy

lunar nebula
#

Hello, I have a short question. I have took some photos from my last shibari session but in my living room. So there is a lot of stuff in the background. Is there a good free (and best local) tool to remove and normalize backgrounds? I have tried the MS Picture editor but it doesn't work well. My plan is to remove the clutter from the background and after that blur the background. Any tipps what tool would be good to use. I do this for hobby and not comercial reason so freeware and easy to use would be nice since it is not my main focus ...

slender vault
#

can try a remove background workflow

lunar nebula
#

I am new to a1111 so I am not firm with that workflow.

atomic mortar
#

Gimp could work if your a bit handy

#

Just paint over the clutter and blur

copper crystal
#

could use layer diffusion extension for forge, but it probably doesn't work anymore since trolls started sabotaging that project in the name of GPL

slender vault
lunar nebula
#

OK, I will look at it. Unfortunatly when you google "a1111 blur background from image" there are only results for commercial tools, no tutorials ...

#

But is that not only for removing background? I want to blur it a lot or make it look more cleaner but it should be the image of my room but maybe without that clutter and extremly blurred. I wish I could show you more what I want 🙂

atomic mortar
#

You can use gimp to remove clutter and blur a background, you dont need ai to do that for you

#

Gimp is free and theres plenty of tutorials that can show you how

lunar nebula
#

I have done that now. But it is semi good 🙂

#

When I get clearance from my friend aka model in this case I maybe ask for some kind of guidance on that as an example ...

#

Everything I tried seems to look unnatural though

atomic mortar
lunar nebula
#

We took some fotos of her in my living room. Just for fun and also so I can test or better "simulate" as if I want to upload them somewhere and also for us to look at. Nothing nude if you mean that 😉

atomic mortar
#

Simulate? You want to make a ai model of her?

#

But yeah for your original request to remove clutter and blur it is still a manual job unless you hire someone to do it. Gimp is a nice tool that can do all of this

lunar nebula
#

No. I mean more that I don't like to try things like working on something with dummy pictures. So we made some photos of her in shibari and took some nice photos. The idea was that we get some experience in such photos - how to take, blur the image, make them look appealing as if we want to upload them on some platforms like reddit or social media (I have no social media) and use those photos as the guinea pigs.

slender vault
#

Gimp, ai cant do this

#

too niche

#

too complex

lunar nebula
#

I have done that for other photos I took in a greater hall so the background was kinda clean. Now with my living room when I review the pictures I see that in the background there are cupboards, some shoes on the floor, the tv and such and some plants. That all is a) diestracting even if I blur that out a little and b) could (for my own feeling) reveal to much about my living room and also looks kinda messy when blurred.

atomic mortar
#

Yeah the end result here is still manual photoshop bud

#

Ai cant do what you want from it unless you transform the room into something else entirely

#

But then the lightning will look off and the image turns messy again

lunar nebula
#

I think you. My idea was more like

  1. Use AI to hammer out the outstanding details like the brown shoes on the lighter floor, the black TV in the background and details in the cupboard with some kinda ok looking substitute. And then blur the whole background to blend it in.
atomic mortar
#

Its a whole project your gonna take on that would be easier in gimp. Just add a png image to substitute to blur. But lets say you insist on using AI

#

What are your computer specs?

lunar nebula
#

a gtx 1070 and 32GB RAM.

#

As said it is more to get some practice and not for effectiveness.

#

But I agree, GIMP is faster and easier.

atomic mortar
#

With a 1070 its gonna take a whole bunch of time

#

Maybe a sd 1.5 model could get ok speeds but from my experience its not gonna be a great one

dull junco
#

if there's not a subreddit for cursed stable diffusion abonimations there should be

vernal sigil
#

hello

#

Anyone able to help me configure stabel diffusion using AMD gpu

#

running into a brick wall here.

#

A full log can be found at C:\Users\kidfo\AppData\Local\Temp\pip-install-6cvo7hm3\scikit-image_3b9981ff0af24673bb8227e0a045d1d8.mesonpy-537bo4m6\meson-logs\meson-log.txt
[end of output]

note: This error originates from a subprocess, and is likely not a problem with pip.
error: metadata-generation-failed

#

damn pip

#

Someone has the info. please help a brotha out

#

video call with screen share maybe?

#

Ill give you 10 sheckles

#

😦

cedar salmon
wind steppe
#

HI, I'M NEW HERE. WHO CAN EXPLAIN TO ME HOW I CREATE THE IMAGES HERE? PLEASE HELP ME.

mortal delta
#

How might i go about creating 2d game assets with local ai and most importantly chacarter movemtent spritesheets, is there a easy way to achive this?

lucid bobcat
#

Why exactly can't we use cfg with flux? I tried and it worked! Even negative prompts worked. And I mean out of the box and not with any trickery. Damn I wasted so many image generations fiddling around with loras because I thought using cfg with flux is taboo.

deep narwhal
#

anyone looking at the new 50 series cards?

#

performance for gaming isnt that big of an upgrade but nobodys talkin about how much better theyre claiming that theyll be at image generation

atomic mortar
scenic stirrup
#

gm⛅

frank rampart
#

has anyone figured out how to make AI comic books? im an artist working on a comic, and would love to use AI art to speed up the process but its been churning garbage for me for a year now. i even tried making a lora with my characters

languid haven
#

Comic book consistency has always been junk, it's far more reasonable to generate panel by panel, using controlnet for scene control/blocking etc.

sharp niche
#

helo11

mortal turret
terse sparrow
#

I read that illustrious is good for prompting details to multiple characters, but I can only find guides on how to prompt multiple characters that the model knows.
For example I want to prompt 2 girls with no names, one with blonde hair, another with pink for example. Both have different body types, etc.
How do I do that without contamination?
I would rather do it without regional prompting, it doesn't work great for me

steep path
#

Can someone summarize what Trump's inauguration means in terms of China policies? Does europe get cheap materials soon?

slender vault
steep path
#

I know, it would help tremendously if the compute costs went down.

atomic mortar
#

This could seriously impact the trout population

fervent thunder
#

why do people keep thinking tariffs would make things cheaper? 🤔

slender vault
#

because the average person is uneducated unfortunately

steep path
#

Tariffs increase costs, locally.

atomic mortar
#

for the consumers

fervent thunder
#

Tariffs don't just increase costs locally, they increase costs globally because the economies of the world are not in isolated bubbles that don't interact

quartz siren
burnt copper
#

So I realize today that there are a lot of people who don’t like the idea of AI art, did y’all know this?

burnt copper
desert dagger
slender vault
burnt copper
#

Talking about how we’re stealing work from true artist or something like that

lucid bobcat
quartz siren
lucid bobcat
lucid bobcat
fervent thunder
#

hello

rugged wyvern
#

People here are predators they support discussion of grape and politics

#

Full of illegal discussion that violates discord TOS

#

All saved and will be shared.

clear fossil
#

So what exactly is this? I don't understand what I'm doing?

atomic mortar
old gull
#

Has anyone got automatic1111's webui working on an AMD 5700xt, on windows? Or am I still stuck to Linux? (Getting back into it haven't installed it in a year)

warm junco
south gull
#

hello

dawn mulch
#

@clear fossilYou mean this place? Its a place to talk about Stable Diffusion, an AI image generator.

gentle prairie
#

i really don't get why some lora don't replicate the artstyle at all, even with the right model and wieight

#

even tried with like just the lora and prompt included with it, and nothing

#

lots of lora works great, and some other it just doesn't do anything

sand trellis
#

hellow [:

young surge
#

@proud thistle sup?

burnt copper
#

G Morning

glacial marten
young surge
glacial marten
young surge
#

Why would they? They got quantity.

#

Also... Who the heck uses their "real" name as a user name?

burnt copper
#

They asked for my address and where I lived, that’s literally the first thing they asked

young surge
#

🤷 I've seen too much spam by users incapable of own thoughts, who gets hacked cause "Free Steam card - Totally not fake". Unless i know you, I ain't adding you.

glacial marten
burnt copper
#

Who does that

young surge
#

Scammers

burnt copper
#

Like I didn’t even know what e-pal was, I seriously thought it was a slang name for PayPal

young surge
#

Jokes on them. I'm broke 23/7

burnt copper
upper plinth
#

would 1.5 still be the best choice for creating minecraft textures?

lucid bobcat
burnt copper
atomic mortar
gentle prairie
#

is citivAI down for anyone elese?

burnt copper
deep narwhal
#

anyone know how to get started with fixing faces in swarmUI? do i use refiner?

rotund wharf
#

short questions 🙂 Am I allowed to make a job offer somwehere in the groups here?

slender vault
#

no

gentle prairie
#

Man i wish there was a way to know which prompt or lora breaks another lora

#

when you have a lot, it's too annoying to test all of them

#

lol

nova shadow
#

I'm on a mission on binan wallet, where can I get my test tokens

floral umbra
#

In the past, for instance when 40 series released, was there any wait between the release and when the new cards worked with generative models? Or did it simply work straight out of the box immediately?

#

Thinking as i'm planning to acquire a 5090, just more curious whether proper support within pytorch is needed and such.

finite cloak
finite cloak
floral umbra
finite cloak
#

... basically any? nothing physically prevents copyright violation, it's the law that prevents that (eg if you include a replica of a copyrighted character in a commercial work you release, the original owner can sue you)

finite cloak
#

Stable Diffusion is a family of models, not just one.
Most people these days use Flux.1-dev (isn't SD, but is a derivative of SD3)

I typed the prompt a photo of Alvin Seville from Alvin and the Chipmunks in with Flux as my selected model, here's what it generated: https://i.alexgoodwin.media/i/misc/0022dd.png which isn't quite perfect but it's pretty dang close

#

You can also do image-prompting. Here's an image prompt for Alvin (using random picture off google) using SDXL (older, weaker model relative to Flux, probably can't do Alvin on its own, but with an image prompt it gets very close... albeit kinda wacky lol) https://i.alexgoodwin.media/i/misc/294172.png

humble iris
#

I ve started a collective SDXL GPUs benchmark. Please consider testing your GPU:
only ComfyUI, fp16
create a template workflow (menu Workflow - Browse Templates - left one) and change the model to ponyDiffusionV6XL_v6StartWithThisOne and the resolution to 1024*1024.
calcute the average it/s of 4 generations
do you think I should calculate price-performance on used or new cards?

finite cloak
#

i literally linked a model trained on the movies

#

so, not hard lol

#

to do it yourself mostly just requires a decent GPU and some time to run it

#

(eg a 4090 can do a small lora train on SDXL in ~30 minutes, or Flux in a few hours)

#

uhhhhh enough for SDXL? Yes. Flux? probably not

#

SD3.5m might be a better base nowadays to fit on a 2060? not sure

#

I can't personally give much info about training, but there's a lot of people out there who can. OneTrainer is a pretty nice training UI that supports just about everything

gentle prairie
#

what is clip skip? I see it a lot of civitai where it's recommended to be at 2, but i can't find the option

finite cloak
gentle prairie
#

cool

#

thanks

upper plinth
#

I'm still overclocking my 5090.

#

I can't wait to see how fast that shit is gonna gen with all those sweet new tensor cores omggg....

slender vault
#

Im not impressed with the 50xx

fervent thunder
#

5090 is around 20-40% faster but uses 30% more power so its not amazing yeah

#

the B100/B200/B300GB200/GB300 and NVL72 are the impressive parts of Blackwell, not so much the consumer chips this time

hollow oxide
#

I want to generate (AI) ice text. But I need to use a specific font. I create a text with this font, for example in Inkscape, and export it as a black and white mask. Now I need to give a task for ice generation using this mask. Is this possible? I use SD on my computer via the web interface. I know that I need to use Img2img but I don't understand exactly how.

fervent thunder
#

probably just canny controlnet would be enough if you tile up to a large enough resolution

lucid bobcat
#

You will also need to prompt for ice.

#

I can imagine that there's even a node in comfyui so you don't have to use inkscape to create the text (unless it's a custom font)

hollow oxide
#

I use Ukrainian text. Without the sample of text it don't work properly

fervent thunder
#

img2img can work yeah but you may as well add a control net it would help

atomic mortar
#

Adobe firefly does something similar too for your specific use case if you own a adobe licence by any chance

balmy magnet
#

hi

slender vault
#

beep boop

final loom
worthy nymph
#

Hello

grand rain
#

how do i tell if an image is poisoned or not?

north girder
#

Hi everyone, enjoy your Friday

#

anyone knows how can I use sketch to image?

livid gale
#

is there a model optimized for 16:9?

vestal dew
#

What is the go to opensource AI lipsync now?

lucid bobcat
low moon
#

And a smell test.

#

Ice cream cowboy.

tender gull
#

has someone ran the hunyuan-video model yet?

finite cloak
#

many people have yes lol

low moon
#

yes it works even n 12 GB VRAM

#

Is it perfect no, is it close? yes. Will it be good enough with image to video? yes.

unborn hedge
#

kling ai is kinda nuts, for local, is the closest option to quality hunyuan?

quartz siren
unborn hedge
#

aw man, i only really use image to video, i hope that update comes out soon enough, idk if my 16GB Vram card can handle it though

quartz siren
#

Yeah one of the hunyuan devs said supposed to come q1-q2 this year, not sure what’s the exact time.

sterile parcel
#

Looking for some help. This the right place? Stable Diffusion related. My mental health help is coming from another discord (bad joke)

#

If someone sees this and is willing to shoot me some quick advice on a settings fix, please ping me. I will love you forever and may send you a gift basket

pseudo granite
#

"When I try Google corabo, install the extension, and start it again, it says ""An error occurred while loading the script"" in the code."

If you know how to solve this problem, please give me some advice

atomic mortar
sterile parcel
#

I can do that

atomic mortar
#

Yeah and i recommend describing the problem while mentioning your pc specs, what UI you use and a screenshot of what's going wrong

verbal island
#

👍

devout hull
#

Hello... Yesterday my stable diffusion webui was working fine.... today i get a torch error, unable to use gpu or something like this... what happened? Like I used it 10 hrs ago and now something broke?

#

Now I deleted the venv folder to see if maybe it fix itself... but this is strange.

#

Webui forge is working using flux so... I don't think it's os or hw related

warm junco
leaden cargo
#

Hello, can anyone help me?

atomic mortar
vast nimbus
#

Hi everyone, I’m looking for the vae-ft-mse-840000.vae.pt file. Does anyone have a working download link or know where I can find it? Thanks in advance for your help!

warm junco
#

There is no support server

lucid bobcat
vast nimbus
lucid bobcat
#

or .chpt

finite cloak
#

that's an SD1 VAE tune

#

there's been 3 versions of stable diffusion and several competitors since then

frank rune
#

Does anyone know where I can find a list of API parameters for Forge? The FastAPI docs are not extensive enough for me.

split acorn
#

Hello

unique sage
#

is there a place where I can download upscalers? I like latent antialiased, mainly because of the slight blur which makes my stuff look very nice, but it doesnt allow me to go beyond 1080x1080 upscaled by 1.5, since at that point it deforms bodies and limbs quite a lot. I tried some 4k upscalers which work fine even when i go to 2160x2160 (after upscaling x2), but theyre way too clean and i dont like it much. is there some latent upscaler that goes to higher resolutions without deformities?

finite cloak
unique sage
#

tried it, there's nothing there that im looking for ;<. unless you know one on that website that fits what im looking for

dense shoal
unique sage
#

nah, I generate hundreds of pics a day, I need something that work in stable diffusion on every single generation

crystal moat
#

is it normal for images in a batch looking nearly exactly same ?

atomic mortar
#

Cause it shouldent

crystal moat
atomic mortar
#

Oh if its random or -1 its good

#

Are you using a controlnet etc

crystal moat
last jay
#

Hi guys, can anyone please point me to where i can learn to edit images using Stable Diffusion using code and not a webUI? I want to create a tool to help me in my Design work. Thanks in advance!

slender vault
#

webui is code

#

just frontend gui

atomic mortar
#

Assuming the tool is a image to image comfyUI can do it or has a node for it

#

Or its yet another starting business with a unrealistic expectation of ai

slender vault
#

yea lol typically is

quasi spindle
#

Hey everyone!
I’m planning to create something exciting—maybe a bot, an app, or a tool that solves real or fun problems ! 🤔
What’s one thing you’ve always wished existed or a problem you face that you’d love solved? Let your imagination run wild! Drop any ideas, and feel free to bounce off each other’s suggestions!

slender vault
#

For people to stop coming in and having unrealistic expectations and asking others for their ideas on furthering fueling these ideas

atomic mortar
#

tech talk and coding convos? screams bot

lethal plinth
#

hello ^^

atomic mortar
#

hii

lethal plinth
#

how are you ?^^

crisp trellis
#

I am trying to build a computer that will locally run an LLM like llama 3.2 90B.... Would a computer running with 5090 and 192GB of ram be able to run that large of a model locally?

atomic mortar
# crisp trellis I am trying to build a computer that will locally run an LLM like llama 3.2 90B....

3.1 70b and llama 3.2 90b have the same text model. you don't need 90b if you're only using chat

Option 1(very cheap): 64gb ram - run Q5km with 1t/s on CPU

Option 2 (mid): 2x P40 (2x 24gb VRAM) - run q4km with 6-7 t/s, llama.cpp only

Option 3(top): 2x 3090/4090 (2x 24gb vRAM) - run exl2 with exllamav2 - faster prompt processing and generation, also can train diffusion and text models.
(source 2 reddit posts)

crisp trellis
#

i was wondering if it can run on 32gb, or what is the lowest amount of vram I need to run the 90B model

worn swan
#

hi In which chat should I ask about the possibility of generating women using Automatic1111 Web UI

warm junco
lucid bobcat
#

Is there a node that can do 4k depth anything v2? controlnet aux is very limited. I already hacked together a working node but I'd rather use something more professional.

finite cloak
fervent thunder
finite cloak
#

sure?

fathom gate
#

90B is very large though. There are smaller versions. Choose wisely if you actually need that high complexity. In a more focused set of roles, a much smalelr model can be as good as the larger ones

lucid bobcat
#

Does anybody know how to add noise to a latent in comfyui so it still converges with 100% denoise? I'm using KSampler (Advanced) with flux dev. I tried multiple different noise nodes but no matter what parameters I choose, the image will not converge when the starting step is set to 0.

fervent thunder
#

using RF-edit from Fluxtapose node pack to add noise can be good

#

unsampled noise is calibrated by default

#

if you are just wanting to add stuff like gaussian, perlin, laplacian, simplex etc then its a case of noise scaling, specifically that it has to be variance preserving, as in the variance preserving SDE from Song 2020
if you are able to set the variance by hand it is not impossible to calibrate by hand

lucid bobcat
# fervent thunder if you are just wanting to add stuff like gaussian, perlin, laplacian, simplex e...

I only want to add the default noise, nothing fancy. I found out that KSampler (Advanced) will always return the same bad latent when adding noise is turned off and start step = 0 and end step = 1, no matter the input latent. Is this normal or a bug? I thought I could use this node in a loop, doing one step at a time. But it looks like it's impossible because the first step always returns a garbage latent.

fervent thunder
#

if you use sampler custom advanced with either random noise node or disable noise node it works fine

lucid bobcat
fervent thunder
#

I'm talking about a different node, rather than ksampler advanced I am talking about sampler custom advanced

#

skipping penultimate sigma can help, a node in here does it: https://github.com/blepping/ComfyUI-bleh

#

some of the stock sampler+ksampler combos will add a ton of extra noise to the image without it

crisp trellis
crisp trellis
#

I want it to be knowledgeable on various topics of course

#

Not specialized, but at minimum surface level

fathom gate
#

for running in RAM, that should pretty much just happen automatically if you allow it

#

192GB should be ample - and even though its an order of magnitude slower than VRAM - it's probably more than enough for a single user to be happy with

#

the latency on the start of the answer might be annoying if integrating into voice commands or whatever - I don't know what you are using it for

crisp trellis
#

Yeah but I'd like the responses to be generally quick and not take minutes for a response. A couple seconds to have it think while it loads and a reponse is fine

crisp trellis
#

I'll be converting voice to text and input that as well

fathom gate
#

well there is a visual llama3

crisp trellis
#

The 9B/90B models

fathom gate
#

not sure how small it comes in tho

#

I don't have much experience with visual input. It's a different beast - and while very useful, the AIs aren't god-tier at it yet like with text

crisp trellis
#

Well it's going to be provided some sort of text so it will know atleast a littlke context of the photo

lucid bobcat
#

I don't think the node you mentioned will help in my case. I use plain old Euler simple.

#

I'm starting to believe this is a bug

fervent thunder
#

yes even when add noise is disabled
the penultimate sigma issue happens at the end of sampling though

#

yeah it could well be a bug

#

everyone has different setups so its hard to compare

lucid bobcat
fervent thunder
#

hmm that's a strange one

#

is sd available on amd cards?

#

yes but its tricky

fervent thunder
#

yeah

#

its possible

#

same for LLMs

#

@fervent thunder i have a 7900xtx what would you recommend me to get for anime generation(or any art rrly)? also one for chatting with llm bot

#

because there is like so many options 😭

mortal delta
#

dumb question, but can any ai help to make a pcb bored or something of that nature. im just curious.

fervent thunder
#

probably Flux Dev for image
LLM is tricky but Qwen 2.5 14B or 32B would be fine

fervent thunder
#

it could have absorbed some hardware books or something

nimble grotto
#

hello

solar wraith
#

Heyy guys , I wanted to build a headshot generator app using open source model .

( Without finetuning)

Which model can I try out ?

atomic mortar
atomic mortar
#

But if you want to get your own picture into a ai image you would need to use face swap, a dedicated server (because its for an app) so you need to finetune!

lucid bobcat
atomic mortar
#

There are auto complete algorithms for extremely simple ones but tbh a stable diffusion discord is probably the wrong place to ask

#

Since there are pcb communities

solar wraith
atomic mortar
#

Youd need a few models. Or flux with a few loras. But then you need to create a few workflows within comfy to facilitate that

#

But i see that this is yet another million dollar idea so my advice ends

solar wraith
#

Okayy thank you !

#

But wait , that means flux can handle my usecase properly right ?

atomic mortar
#

If you got the face swap nodes and the know how to configure it so it works like you want it to

languid haven
#

I still don't get what's the advantage of Illustrious

#

When I prompt for it to sit reclined, it does it consistently 4 out of a batch of 4, whereas pony hits maybe 2/3 times, sure. But it can't seem to shift the style away that very specific anime look

slender vault
#

you save so much on tokens cos you dont need all that score bloat crap

languid haven
#

I use embeddings for that score stuff

#

Pretty sure it's 1 token for score_5 - 7

slender vault
#

ive always been an advocate of pure prompting power, i find embeddings and such to be a hindrance

languid haven
#

So is the use of illustrious better when the pormpt has less than 75/150 tokens?

#

Neither one is really great with natural language as far as I can tell, Illustrious has a little bit of edge

#

on that front

#

Is it mostly because visual styles? As pony is obviously fixated on that broad jawline stuff

atomic mortar
#

God i hate natural language prompting

#

"the background is jubilant and serene but her gaze is deterministic. She stands confidently " and other bs

warm junco
languid haven
#

Well s--t

#

I don't know much about booru tags

#

I just prompt whatever, for pony, by cutting my language into tags

#

Anyone got some good example prompts for Illustrious here? I can't figure out the right way to use it

warm junco
languid haven
#

OK, I will give it a shot, thanks.
BTW, would you say that Illustrious is better at booru poses than Pony?

warm junco
torn swallow
#

Hello fam, been away from AI image generation for a lil while, what is the best way to run stabil diffusion these days?

warm junco
atomic mortar
#

But if you mean UI its comfy, swarm and forge

silent sinew
#

Is there a way to put real life object on image generation?

slender vault
#

elaborate

#

you want your face? your dog?

#

something specific etc

fervent thunder
#

or Qwen2vl-Flux

#

or ACE_Plus (is also Flux)

silent sinew
silent sinew
fervent thunder
fervent thunder
slender vault
#

You dont need flux, you can train a lora of a specific item or person

sick cairn
marble portal
#

Does ram amount mater a lot with running stable diffusion locally?

fervent thunder
#

yeah at least 64GB, or otherwise at least 32GB is good

#

you can get away with less but it gets trickier

marble portal
#

Okay I have 7950x3D, 64gb ddr5 and a 5090 soon, so I guess next upgrade I’ll make is my ram amount

warm junco
mortal delta
#

I have a rtx 3060 now, i want to try my hand at better llm models, ive tried 32 sized models but they are slow. is there a way to run bigger models locally and have it be at good speeds or something?

fervent thunder
mortal delta
warm junco
mortal delta
untold pewter
slender vault
fervent thunder
#

ollama is fine ye

#

with GGUF 5_K_M and IQ4_XS do extra well for their size for some reason

#

so if you can find 5_K_M or IQ4_XS versions of things then that is good

atomic mortar
slender vault
#

who knows

#

i dont use it for those purposes

#

😛

#

but its a reasoning model, itll display its though process

#

so not great for immersive conversation

#

that being said, still a dang good model

#

for a side hustle project

fervent thunder
#

maybe someone uncensored it

atomic mortar
#

Oh me neither lol. Tough some guidelines are a bit too strict

fervent thunder
#

when people say uncensored they can mean different things though

fervent thunder
#

can someone guide me thru this whole AI thing? i got 7900xtx and i want to generate stuff locally

atomic mortar
#

but you need zdula to run it properly

fervent thunder
#

this the best i can generate locally right?

atomic mortar
#

the best would be comfy but i think it could be a bit too complex for a beginner

#

but for you i would recommend forge as theres more tutorials availible

fervent thunder
atomic mortar
#

comfy just has big learning curve and its not everyones piece of cake as its mostly puzzling lines together and if you dont know what your doing its more frustration then anything else

fervent thunder
#

oh alright then i will go with webui

foggy yacht
#

Has anyone been able to train a good flux finetune for style? I mean full finetune not a lora. On style it's easy to train on a single subject. I'm trying to train a style with 200imgs, but no luck.

#

Uh what?

#

Oh

marble portal
#

How do I know if my pc can run the 671b deepseek model

atomic mortar
#

ah hey there i got the awnser for you

#

you cant

#

ill tag you in the list of requirements in general with images

slender vault
#

youd need like 4 or 8 a100s lmao

marble portal
#

It seems like once the 5090 with 32gb comes I’ll be able to do the 60 one at least

slender vault
#

yeah lol

marble portal
# slender vault yeah lol

How are you liking it? I haven’t had a chance to use it yet, waiting on their new image generator to be posted to ollama to use it

slender vault
#

im hoping they do a chat version or something cos reasoning models display the thinking process which is nice but i cant incorporate in my workflow because of it

fervent thunder
#

non-chat LLMs are rough yeah

near silo
#

My friend has 6xA100's, and he runs it at low precision

fervent thunder
#

unlucky cos its only just too big for 16xH100

#

1300/16=81.25 LOL

slender vault
twin igloo
#

Does anyone have an NVIDIA GeForce RTX 4090 and has created video with it?

atomic mortar
#

lots of people have. whats the question that comes after?

fervent thunder
#

ya, im sure most can

slender vault
#

considering dont know the age

#

hard no

fervent thunder
#

frfr

fervent thunder
#

Hello

slender vault
#

beep boop

bronze basin
#

So what is our opinion of Janus

fervent thunder
#

-googles what a janus is-

#

looks to be a british investor thing

#

dunno tbh, i do wanna invest in sum, but'll prolly be a savings account

atomic mortar
#

Janus pro, im waiting for some more examples but promising so far

slender vault
torn swallow
#

Need to reinstall all the stable diffusion stuff and wanted to get the best possible UI and way to use SD 😄

#

I have not used SD in over 1 year

barren elbow
#

generate a photo on this topic: Basic methods of primary surgical treatment of wounds

atomic mortar
#

Seeing the question, forge is best for you

#

Check pinned in techsupport for guide

torn swallow
atomic mortar
#

Yeah then forge-webui

fervent thunder
#

I need to download Python to use Anaconda3 to use Pytorch to install torch-directml to use comfyUi to use a training ai to use a model to generate pics

#

(am never leaving my room)

marsh wasp
#

I used 16GB before with Stable Diffusion - It WORKS but uses your swap file a LOT. and if you have that on a SSD, not ideal.

fervent thunder
#

once you have loaded the models its ok

#

just make sure they stay loaded lol

atomic mortar
#

You mean you want to pay someone for a ai job? I recommend civit ai for a bounty or fiverr

#

If you just want a few Images, once im home i can help you with that (free)

warm junco
fervent thunder
warm junco
#

?

#

My guide covers multiple webuis with zluda.
Directml is not recommended

#

If you mean more than one python version. Most if not all ai stuff with AMD works with Python 3.10.11 64bit

fervent flame
#

hello

atomic mortar
#

hello new person on discord.

lucid bobcat
#

"person"

upper plinth
#

Has anyone run Deepseek locally here?

#

Is that shit really the next best thing for uncensored LLMs or is that just hype

atomic mortar
#

but if you want The NEXT BEST THING you need about 1.3tb of vram but the 7-13b models are okay

upper plinth
#

ok but not revolutionary?

#

cuz we already have tons of 7-13b llms

#

so Im wondering what makes this one special

atomic mortar
#

its comparible to OpenAI at the full model

upper plinth
#

so small deepseek = full ChatGPT 4?

atomic mortar
#

and actually opensource instead of "open" just in name

#

what no

#

at the FULL model

upper plinth
#

oh

atomic mortar
#

we can run small models

upper plinth
#

so its just hype. got it

atomic mortar
#

its good but not WOW. the Janus model is pretty nice tho

#

as its comparible to DallE

upper plinth
#

given the fact that it tanked Nvidia stock I was expecting something crazy

#

the only thing crazy about it is its price and thats it

atomic mortar
#

stockmarket is bad thing to look at as its a giant bubble and investors are twitching

#

make model
nividia stock drops
people realize they need nvidia cards to train etc
nvidia stock rises

upper plinth
#

rofl

#

it makes me wonder why if China is so advanced technologically as the media wants us to believe, they still can't produce their own GPUs.

#

we need someone to compete with nvidia's shite business practices

atomic mortar
#

nvidia is both a goldmine owner and selling pickaxes, they make a ton of money

upper plinth
#

I can guarantee you the real price for 5090 will be closer to 2600 pre-tax since it will sell out in the first 6 seconds. AMD nowhere to be seen. We need new players

atomic mortar
#

we got AMD but people didnt give it a fair enough chance

#

and they fell behind

#

and the problem in computers is

#

we litterally made a rock think (not to oversimplify we used lightning to help)

#

good luck trying to improve on that

#

as a startup? nearly impossible youd need investments of trillions for a chance

#

the machines alone are a gigantic investment and youd not be there even

upper plinth
#

china has the money

#

and the interest is there, look at the recent shortages there have been

atomic mortar
#

but not the technology. lithomachines are resticting htem

#

them*

#

and for chip production you need spesific conditions

#

they are catching up slowly though

upper plinth
#

yup. I give it 5 more years tops

#

for chinese branded CPUs and GPUs

atomic mortar
#

but not any area is simply suitable witch also limits advancement speed

upper plinth
#

well there are fabs in taiwan, israel, costa rica

atomic mortar
#

what do they have in common

upper plinth
#

all owned by the same companies

atomic mortar
#

and the biggest problem of all

#

lack of talent

upper plinth
#

there was an interview not long ago that mentioned china has more honor students than the US has students

#

if there's any country that has insane talent it's china

atomic mortar
#

but they dont get the oppertunities and get crushed by the insane work culture

upper plinth
#

that is true given that probably half the country is still in poverty. but just give it some time

atomic mortar
#

now we all get mad at a printer right. how much babysitting it needs

#

now imagine instead of paper its 10 micron spheres being shot around 50k times in a vacuum at a very fast speed being shot by a laser

deep narwhal
#

im lookin at upscalers and it looks like people have been using the same ones for like 2 years

#

is that right?

atomic mortar
shy sluice
#

I have some credits I bought before artisian came out can I use those or do I need the membership?

#

I used them on dreamstudio

deep narwhal
#

thought i was just couldnt find the new ones

atomic mortar
#

i prefer using 4x-AnimeSharp.pth - 4x_foolhardy_remachri.pth and 4x_NMKD-superscale-SP-178000_G

#

both are on civit ai but the anime sharp not iirc

eager shard
#

What's the utilities to make stable diffusion work on amd cards? I just dropped $1800 on a 128 ram PC with two 7800xt for topaz and running local llms

atomic mortar
#

personally i recommend swarm UI or webui-forge

eager shard
atomic mortar
#

you NEED to follow the AMD guide or its going to be slow

eager shard
#

I have a 3080 that I use a ton for this but I have zero clue how to do amd

upper plinth
#

128 ram is an absolute waste of money

#

you are better off getting 32gb of VRAM for less than half of what you paid

deep narwhal
#

128gb ram doesnt even cost that much if youre planning on rocking 2 video cards

#

its like what 300 bucks?

atomic mortar
#

it depends if you use it for different cases aswell then its not a waste

#

imo

upper plinth
#

only if your sole purpose is to run turtle-speed llm's which most suck compared to subscription-based models

atomic mortar
#

ai isnt the only use case to have high ram lol

upper plinth
#

especially with deepseek pro being like 1/10th the cost of GPT pro

#

it certainly is

#

I have 32gb and never ever do I use it all

#

maybe only in minecraft lmao.

#

unless ur running a server or sum shit

deep narwhal
#

what did you spend the $3 you saved on

atomic mortar
#

thats your use cases yeah. virtual servers, processing large amounts of data (3d animation etc)

upper plinth
#

a 5090

atomic mortar
#

as 3d animation will use any ram you give it

upper plinth
#

vram is worth its weight in gold. ram is useless except for super niche cases

deep narwhal
#

this man is high

cedar salmon
#

so some can be more capable instead of hitting a wall, sounds like a good reason

deep narwhal
#

of all the things to hate on, why hate on having more ram than you need? lmao

atomic mortar
#

im jelly of his build lmao

upper plinth
#

because unused ram is wasted ram

#

= wasted money. you are better off getting multi-quadros

atomic mortar
#

you say wasted. i say he might have other use cases as he spent a lot of money on that build

cedar salmon
#

should of got a tablet

last jay
finite cloak
eager shard
#

Two 7800xt = 32gb vram

lucid bobcat
#

What is Janus good for? Is it only image captioning? I'm not impressed by the image generation capabilites. And the 7B model is terribly slow.

atomic mortar
#

Speed is mostly on specs your pc has no?

eager shard
broken forge
#

Hello, what is currently the best checkpoint in forge application to generate very realistic images and characters ?

#

flux1-dev-bnb-nf4-v2 ?

deep narwhal
#

i love the way you prompt for pony models

#

but im pretty sure you can use the same language for sdxl now right?

#

last time i use sdxl you had to write a whole dang book for a prompt

eager shard
#

i forget which pony based model i use let me check

#

autismmix confetti

sleek stream
#

yo is there any alternative to reactor?

lucid bobcat
#

If you modify the comfyui node for Janus image understanding you can pass a batch as input and make it choose a favorite. Could be handy for automatically picking the best seed. Just a tip for anyone interested.

calm hearth
#

Hey does sd have a vanish brush to like scratch things out of a pic? I can’t find anything regarding that, maybe an extension?

gritty lava
#

my old mate used to use the inpaint thing

#

to do exactly that

#

how tho, beats me, they was jus good at it after practice, id give u their ID but, um, we're not on good terms anymore so, cant xd

broken forge
#

will someone help me ? I would like to generate a character from a prompt but want to later create more pictures with the same character (face hair eyes body) to then have a lot of pictures for fluxgym

#

in forge( stable diffusion)

junior shuttle
#

Are there any good models for virtual clothing try on? eg. user provides photo of custom model and garment. Something like Kling.ai?

#

Preferrably one that is capable of handling graphics on shirts and such as well

lucid bobcat
#

Are there any RTX 5000 series Stable Diffusion and other AI models performance tests out yet?

atomic mortar
atomic mortar
atomic mortar
junior shuttle
atomic mortar
junior shuttle
#

Unfortunately no

atomic mortar
#

Its an million dollar app idea again so honestly its gonna take a lot of learning or paying someone

#

Theres a app thats in beta rn thats trying to do exactly that but its costing them a pretty penny

#

Infact last week we had someone with the same idea lol

#

You could try a mix of a controlnet canny model + inpaint with automatic segmentation and hope it transfers properly

junior shuttle
#

I've been looking around for something I can run locally on hugging face. Theres a few that I've tried directly in my browser but the results arent promising. I just want to try on some things Ive made but I get what you mean, if there was one it probably would be million dollar idea.

atomic mortar
#

But running stable diffusion localy, its not that hard in its self

#

But what you want? Thats a challenge

junior shuttle
#

Right lol

atomic mortar
#

You could lookup a controlnet impainting tutorial on youtube for comfyUI but its missing over half the things youd need

cedar salmon
junior shuttle
atomic mortar
junior shuttle
atomic mortar
#

But thats the current state of image generation .its impressive but were not there yet

fervent thunder
#

there's flux ones

edgy summit
#

Could someone recommend a more recent model, SDXL or Pony that does Fantasy / Dark Fantasy / Armored characters well?

edgy summit
#

I can't find a model willing to do lots of blood, even for dark fantasy purposes

violet prairie
deep narwhal
# eager shard autismmix confetti

ive just been jumping around using whatever model pops up under the most downloaded for the month 😂 i think i should settle on one and learn it inside and out

eager shard
edgy summit
#

tried it just made a blob of tomato sauce.

eager shard
#

😭 I mostly just use autism mix confetti and it does everything I've tried

#

It's just a wallpaper machine at this point

bleak matrix
#

Good morning, everyone! How are you all today?

minor dragon
bleak matrix
#

Ah, so, real life arrives! rip

#

I hope your day improves ☀️

undone garden
#

Is there another gradio based platform other than auto1111 and forge that's getting popular or did everyone just move to the trash heap that is comfyui? Both haven't been updated since last summer.

dim oxide
#

even yesterday.. just update it after downloading. i'm still sticking to forge

#

btw, i got a question... how does Adetailer picks "top K masks" ? i'm getting some random inconsistencies in img2img, same image, different seed, two people - sometimes the top 1 k mask is person on left, sometimes person on right... whats going on there?

atomic mortar
slender vault
atomic mortar
#

i get disliking the noodles but comfyUI is the most sophisticated ai software availible to the community tbh

warm birch
#

hey i want to install stable diffusion

#

but there is still the error that couldnt lounch python

#

but i installed python 3.10.6

#

does someone else know what the problem is? thank you

atomic mortar
warm birch
lucid bobcat
# violet prairie Latest breakthrough of AGI! TL;DR: It is all about money and evolution. You can ...

Is self-promotion even allowed? Also I read a little bit of your paper and it's complete nonsense. It reads like a kickstarter scam.

print("Hello world!")
From the point of view of classical textbooks, this process is simple, as if the meaning of the execution of
the program is completely within the code itself, without any further connection in real life
So you think "print" needs no further context about the printing press, modern printers or monitors? Sounds like dialogue for a bad sci-fi movie. Did you even write this yourself or is this an LLM?
Sorry to sound so harsh, but don't sell your blog post as a scientific paper.

undone garden
undone garden
atomic mortar
slender vault
#

This arguement is flawed

#

You dont seem to know about workflows

#

Maybe try seeing what the tool can do before disparaging it

atomic mortar
#

in swarm once you get a workflow going to the point where you only need to edit the prompt you can even save it as a "simple" tile to use

#

or well just save the workflow you made once? its not that much work and it allows for much greater results then forge ever could

slender vault
#

Yup

#

I got a whole folder full of workflows for specific tasks

#

It boils down to people having unrealistic expectations with ai

#

Also, comfy allows for more complex workflows that produce interesting results

undone garden
#

But it's pointless to argue about it, you do you

idle axle
#

Comfy has two major problems. First, it’s fundamentally a backend tool that enthuasiasts are trying to force on people who want a front end tool. Second, workflows result in combinatorial explosion.

slender vault
#

lol what

#

how does a .json file containing the logic "slow" things down

lucid bobcat
idle axle
lucid bobcat
idle axle
#

That's what I mean by "backend tool". It's something that should be used behind the scenes but not forced on the user for 99% of situations (IOW, don't do anything ridiculously limited like Fooocus).

lucid bobcat
# idle axle See my comment about combinatorial explosion. Those "yo, what I did but now with...

Under the hood Comfy works no different than other UIs. I don't know what you mean by combinatorial explosion. You can bypass nodes, you don't have to create a new workflow for every combination of steps. You can even convert part of a workflow to a group node, which will get rid of all the node salad. I can do things with Comfy that I can't do with other UIs. For example in Comfy I can make a XY plot for absolutely every parameter imaginable. I other UIs I'm restricted to the supported parameters.

idle axle
# lucid bobcat Under the hood Comfy works no different than other UIs. I don't know what you me...

The problem is comfy forces the user to deal mainly with the generation process. Most people couldn't care less about the generation process (unless they really need to fix some specific thing) and care about the results. Basically, for 99% of situations I don't want to have to see a single node or have the UI spend any space on any of them or force me to load some random workflow when I just want to enable a setting. I just want settings in a convenient UI that's been optimized for settings and prompts instead of for connecting nodes. I want a frontend UI, not a backend UI. And I specifically want a frontend UI that hasn't been dumbed down (see Fooocus).
A1111 / Forge are janky but they (almost certainly accidentally) ended up finding a rather good balance between presenting the relevant things while having quite a bit of flexibility. Under the hood Comfy approach is good but I really really really don't want to have to care in the slightest about the flowgraph when there is no reason a decent UI couldn't fully handle that for 99% of situations.

reef escarp
#

Are there not bot channels anymore?

#

With the dream command?

livid glacier
#

how does the artisan work?

#

i rmbr back in 2022 it would message you the result

#

but its been a while

#

new account btw, nice to remeet yall

reef escarp
#

Yeah same

lucid bobcat
# idle axle The problem is comfy forces the user to deal mainly with the generation process....

Yes If you have a user friendly UI with all the features you need then use that by all means. The advantage of Comfy is that you can do more exotic things with it because you don't have to wait for the devs to implement every new functionality or feature. Or if you have a crazy idea you can just do it in Comfy, even if it means that you'll have to create your own nodes. I think everyone should start with a simpler UI and only switch to Comfy once they feel the need for more control.

idle axle
undone garden
# slender vault how does a .json file containing the logic "slow" things down

Because it's an added step, especially since you still have to tailorthe workflows somehow to your current needs, as there is no workflow that is truly universal. It's a useful tool if you want to experiment, it's a hassle if you know exactly what you want to do and you have to add-connect-bypass nodes and whole sections or swap workflows entirely to make it work instead of just checking a box.

slender vault
#

lol

#

you clearly dont know what youre even mad about

undone garden
#

I'm not mad at all, despite your best efforts

atomic mortar
#

i mean if you use specific settings for realism or for anime so you got two workflows

#

and you can just

#

workflow > open

#

done

#

but id like to think forge as a jack of all trades but comfy workflow a master of a trade

cedar salmon
#

some like to fill out forms some like to connect noodles

lucid bobcat
slender vault
undone garden
idle axle
#

Well, that was quick.. I lasted 8 minutes with Invoke before I got fed up and ctrl-w reflex hit (not helped by the fact that it kept showing distracting animations on the screen with no way to turn them off from the UI)

undone garden
idle axle
idle axle
undone garden
viral palm
#

Can you run ComfyUI/Flux on a Mac these days?

lucid bobcat
idle axle
# lucid bobcat The two sets of data you mention here are the image you feed into the model and ...

Cropping is exactly what I mean by "focus". Say I'm working on a 2048 x 1024 image. I want to inpaint a small region with good details but it needs some context so I have a 256x384 area I want the model to see for context. Currently it requires way too many hoops instead of just selecting a rectangular focus region and drawing the mask. openOutpaint almost gets it right but doesn't understand non-rectangular regions once you decouple the generation resolution from region size (and also lacks regular controlnet features).

atomic mortar
#

swarm has a fantastic impainting tool

lucid bobcat
idle axle
#

I have to wait until tomorrow for the vast people to update their Swarm template. I'll give it a try then.

atomic mortar
#

i happen to have the UI open if you want to see the impaint window

idle axle
undone garden
atomic mortar
#

i am working on a shitpost however so yeah

undone garden
#

It's essentially what adetailed does

#

But you can do it manually and adjust the attention and make it as big as you want on inpaint

lucid bobcat
undone garden
undone garden
idle axle
undone garden
#

Yeah it works with controlnet too, I use it with lineart all the time

idle axle
undone garden
fervent thunder
#

Hey, if I showed you guys an image, could an expert tell me if it was made using Stable Diffusion? I am trying to figure out which software was used so I can download that one but don't want to waste my time until I am certain.

#

would really appreciate

#

skoomadentist has been typing for 10 minutes lol

idle axle
# undone garden https://streamable.com/d3ww4l

Now add the step of generating a controlnet preprocessor preview (properly aligned and without downsampling) so you can do trivial edits to it (eg. clean up scribbles or make space in lineart). At least my A1111 and ReForge don't let me do that without copying the original to-be-inpainted image elsewhere that can run preprocessor on it at full res, copying the result of that back and so on. And the frustrating part is that all of the sub-tasks have already been solved, it's just that nobody's bothered to make them actually work (I suspect most people are fixated on generic "pretty" images instead of goal driven generation tasks).

fervent thunder
#

nooooo it wasnt for me

upper plinth
#

5090 drops tomorrow woooooooooooohoooooooooooooo

idle axle
#

How much does it cost?

upper plinth
#

2,000 if and only if you can get one at msrp.

#

suspecting it will sell out in <6 seconds

#

my bot might not even be fast enough.... we will see. hyped af rn

idle axle
#

I'm curious, why the hype? Is the extra 8 GB really that important over just renting a 48 GB vram VM from cloud?

upper plinth
#

boatloads of extra tensor cores

#

ppl say it's lackluster bc it's not a big jump from the 4090 but since Im upgrading from the 3090 for me it will be night and day

idle axle
#

How much do 4090s cost nowadays?

upper plinth
#

well they will cost less by tomorrow. fyi you still cannot buy any new gpu at msrp except during launch

lucid bobcat
upper plinth
#

amazon still scalps them and they are almost always sold out on newegg etc

idle axle
undone garden
#

This is what you're trying to do, right?

idle axle
#

I'll give swarmui a try tomorrow when the new template becomes available. Hopefully it can solve that better.

undone garden
lucid bobcat
undone garden
#

Short side

eager shard
#

I still hate using comyfui but since it's built in swarm I still use it occasionally, mostly for upscaling to 8k as some random comfyui workflow i ripped off the web made it fast in comfyui where it took multiple hours in every other ui

calm hearth
#

The UI in ComfyUI is ironic /s

slender vault
#

^

arctic sun
#

Hey guys! I’m new here (relatively new to ComfyUI). Lately, I’ve been wanting to make a profile picture of myself in the design style of Marvel Rivals artwork. Is there a way using style references, Loras, etc, to achieve this? If so, could you walk me through what your process would be for this kind of thing?

languid harness
#

?

junior shuttle
lucid bobcat
#

They want $400 to upgrade to 32GB? Holy cow leave this website immediately. Buy a laptop from a company that doesn't rip you off.

desert dagger
junior shuttle
#

Its definitely not my first choice just what I have on hand at the moment

quartz swan
#

there should be plenty of style loras for it out by now AbigailThink

junior shuttle
idle axle
#

I think I paid 400-450e or so for the laptop, including that ram + 1 TB ssd

calm hearth
native wedge
#

does anybody know of any sd android apps that ACTUALLY work? the ones that i find crash or don't work 💔

old lichen
#

Guys, where can I find a workflow for working with flux generated icons?

tepid rampart
#

Hi, I am Lee from India

dusty trellis
#

Does anyone know what tools and workflow these types of videos use? Music, I get. Then, image generators to video generation. Then using a video editor to put the clouds and foreground in planes and loops, but how do they get the trees to merge so well into the sky, the butterflies flying around and pool water movement? https://www.youtube.com/watch?v=pvxUHpf1pxQ

fervent thunder
#

masking model for the trees

#

the butterflies and water are likely traditional CGI not deep learning

arctic sun
gentle prairie
#

mmmmh

#

100 batch count usually took me like 1 hour, been that way for month since i've been using around the same settings

#

i started a new one and suddenly it's 2 hours?

#

i don't have anything else open, or i didn't change the resolution

#

what gives

#

can certain LORA make generation longer?

#

it never happened before

#

what is happening... now the generation is not doing anything anymore, it stopped at 11 %... it's still running but it's not using VRAM anymore, like it completely stopped

#

NEVERMIND

#

somehow i turned off tiled VAE

#

my bad, lol

gentle prairie
#

How do i change a lora's name? as well of the prompt itself

#

there's one lora who has __ in the name, but that messes with dynamic prompt

#

i changed the name to just one, but selecting the lora still gives me __

bronze pine
#

Does anyone know what's happened to reActor? Their GitHub page has been taken down

atomic mortar
bronze pine
#

Sad af, was a proper useful tool. Is there anything else like it?

median jewel
#

if i wanted to train a lora for sdxl what gpu is worth getting, 5080 looks kinda good, its also for gaming, im a student so i aint got that much money

warm junco
atomic mortar
#

I think i still have the original because i never updated my swarm that had the extension

drowsy robin
#

Does anyone here know if Mistral-Small 3 24b competes against deepseek's distilled models?

#

It just got updated literally half an hour ago.

# 
- 24B params, 81% MMLU
- Latency optimized: 150 tokens/s
- Competitive with Llama-3.3 70B, Qwen-2.5 32B, GPT4o-mini
- Apache 2.0
bronze pine
long zodiac
#

Good morning everyone, sorry to be a bother but I’m looking for a solution to getting rid of an issue I’m having. It seems all my art is coming out with nothing but gray images. Is there a way to fix this?

slender vault
#

use a vae

long zodiac
#

Thanks Ellie

fervent thunder
#

I just found out that Override settings are a thing
Uh what do they even do

coarse zephyr
#

who do I tell about a spam bot from this server?

modern pagoda
upper plinth
#

welp today was a total letdown

#

looks like another 2 years before 50 series becomes viable to purchase. god I hate this world

drowsy robin
#

The point of that I thought was to lower the cost at once by spreading it across payment dates

upper plinth
#

viable as in availability

drowsy robin
#

Oh.

#

It's all bought already?

upper plinth
#

it sold out in 5 seconds. or less

#

yes

#

all of them

drowsy robin
#

Damn.

#

Well, I'm a unique one since I'm on arc still myself.

#

Alchemist, too.

#

🤷‍♂️

upper plinth
#

oof

drowsy robin
#

Not oof. I can run flux no problem at 3it/s

#

And deepseek 14b

#

its been heavily improved since there's now pytorch 2.5.1+xpu and intel xpu branches for torch 2.7

#

I think the main thing arc suffers on now

#

is the fact it's a 32-bit architecture. This can be bypassed by slicing the tensors under a 4 gigabyte rate.

upper plinth
#

what is the performance of that GPU comparable to?

#

I have a 3090 so I can live but man I really really wanted to taste that 5090

drowsy robin
#

It's a 3060/3060TI comparable

#

in games*

upper plinth
#

interesting

drowsy robin
#

can do VR, AI is no problem now mostly

#

I got the Arc A770 LE 16GB

vernal tendon
#

Halo Worldhabby

gritty lava
#

i know whoooo i wanttt to take me home, i know whooo i want to take me home, take me homeeee 🎵

calm hearth
#

Apparently there were only a few thousand 5090s shipped in all of NA was sold out everywhere for me, anyone here able to get one and wanna flex on us? Lolol

atomic mortar
#

My local pc enthusiasts/expert forms were pretty mad and only a few managed to buy them

floral umbra
#

Is there a tool yet to convert safetensor models to tensorrt models in parts/chunks yet? Where it'll combine the model at the end, as flux is too eavy, even when using shared memory, or ram as vram extension, it eats just too much.

viral shale
#

Hey peeps! An AI novice here. Need some kick starter insights and advice on where and how to's for starting with Stable Diffusion for Architectural Visualization. I have never used SD earlier except for a few times in hugging face.

compact knoll
#

Hey yall. i'm looking to learn and improve my skill. all tips tricks and walk throughs are accepted.

lucid bobcat
atomic mortar
#

I mean, its just someone asking for advice despite the nsfw profile

lucid bobcat
atomic mortar
#

Staying neutral till i see links or dm requests

main snow
#

i have been exposed to the glory of prefectpony

#

that shit goes HARD

lucid bobcat
atomic mortar
#

Especially in servers oriented for an older audience

lucid bobcat
atomic mortar
#

The ones i had meetups with didnt lol, still yeah odd profile for this kind of server but some people just have the npc energy

#

And this one didnt respond to any of this@compact knoll

#

Ill bait sec

compact knoll
atomic mortar
compact knoll
#

odd.. i'm not a bot

#

i promise

lucid bobcat
compact knoll
#

i thought it was to learn more about stable diffusion

#

ai editing?

atomic mortar
#

And to use it on your pc locally

#

What specs does your pc have? Or are you looking to use a cloud based solution?

compact knoll
#

well my pc is rather up to date but ... i do ai mirror edits of my imvu toon and i want to make them better

atomic mortar
#

What graphics card do you own?

compact knoll
#

nvidia geoforce rtx

#

?

atomic mortar
#

If you open task manager (control alt delete) the second tab (looks like a graph icon) shows you the model name

compact knoll
#

rtx 5080

lucid bobcat
#

lol no way

atomic mortar
#

50? Made a typo?

#

Aint no way you got one this quick yet not knowing lol

compact knoll
#

.< ok

#

double checks

atomic mortar
#

? Can you post a screenshot of it in general with images?

#

If it is i would be incredibly jealous due the shortages of any 50 card rn

#

Had a 5090 in my basket and the store crashed so i couldn't buy it sadcat

compact knoll
#

sorry i was mistaken

atomic mortar
#

Hmm and youd be looking to ai mirror videos or still pictures?

compact knoll
#

i do still photos

atomic mortar
#

You could probably run it albeit slowly

#

What UI are you using right now for your current images?

compact knoll
#

android .... apple made the ai mirror images more pixalated

atomic mortar
#

Hmm you used an app for it then i assume

compact knoll
#

that is what ai mirror is love

atomic mortar
#

I mean you can make ai mirrored images on other platforms/apps too so just asking

compact knoll
#

i'm still learning love.

lucid bobcat
compact knoll
#

no .... it's my phone

atomic mortar
#

Well i would recommend following one of the guides CS1o made in the pinned messages in the #🤝｜tech-support channel

compact knoll
#

thank you

atomic mortar
#

Webui forge is the better one for you. Image to image isnt hard but you need the proper models for it

#

Lmk once you get to the step where you have installed it

compact knoll
#

is that pc or phone

atomic mortar
#

Pc

#

Phone cant run it locally

#

Phones simply don't have the capacity for it

fervent thunder
#

a phone could run flux

#

its just annoying to set up and would be slow

atomic mortar
fervent thunder
#

not sure

#

not neccesarily

#

people run fairly big AI stuff on mac

#

but modern iphone and modern mac are similar chips

atomic mortar
#

The base silicon sure but the vram amounts are still gonna bottleneck it

#

O iphones dont even have vram

#

Its not gonna be able to load in even too lol

fervent thunder
#

the way apple silicon works is that

#

it has neither VRAM nor regular DRAM

#

it has a unified memory

cedar salmon
#

These things exist on a smaller scale already, Samsung Galaxy AI Sketch To Image, Google Pixel Studio, Apple no idea neve even heard of that brand

fervent thunder
#

IDK if they are gonna use them but OPPO and Samsung posted diffusion models and upscalers on Arxiv

#

they are not open source but I am not sure if they are coming out closed source either

lucid bobcat
#

If you can run AI on a raspi you should be able to at least run it on an Android phone.

#

I mean I'm sure someone has already done it.

dark hawk
#

Any telegram user here?

#

@fervent thunder

atomic mortar
#

Since his claim "a phone can run flux"

atomic mortar
cedar salmon
atomic mortar
#

"22.3 minutes from submission to output, and my phone felt like it was melting"

#

doesnt sound ideal

#

but add image to image and a possible control net

#

wew

fringe pivot
#

Does stable diffusion 3.5 have to run on comfyui? kind of miss the layout i played around with in 1.4 tutorial i followed lol

lucid bobcat
fervent thunder
#

why stable got more expensive

fallow tide
#

can i use stable diffusion on an amd?

gritty lava
#

bots, bots everywhere

#

geez, we need some bot-be-gone spray

#

put a roach trap by the front door of the server anroud the walls and windows

#

maybe a new way of verifiction on the server would be handy, a way to like, make it effort intensive so as to drive the bot deployers away, like, those new ReCaptcha things that make u have to solve a friggin puzzle

fervent thunder
#

it would be a ton of work though to do it properly I don't think its worth doing

fervent thunder
#

My phone gallery is insanely big
It's almost at 40k pictures and it's a nightmare to find anything
Is there any ai image search thingy that can accurately find what I am describing

atomic mortar
warm junco
fervent thunder
#

Ok thanks

#

Does stable diffusion create a completely unique name every single time it generates an image or can two different images have the same file name

#

It gives them names like 00039-2127574113

warm junco
#

example: 20250201135213.png

peak orchid
#

Is there a way that I can automatically remove all unused nodes from a comfyui workflow for an image?

lucid bobcat
#

If you hold control and drag your mouse, you can select multiple nodes at once.

peak orchid
lucid bobcat
upper plinth
#

Anyone know how to do a tiled texture on either 1.5 or Flux? I remember A1111 had a checkbox for tiled, but it's gone now.

atomic mortar
lucid bobcat
atomic mortar
atomic mortar
#

yeah manual labour bud

peak orchid
upper plinth
#

swarm?

atomic mortar
# upper plinth swarm?

yeah swarm webUI has it under sampling. you could look for something similar within A1111 (tough i recommend using forge webui or swarmUI)

upper plinth
#

Oh I see

#

Does anyone know if this is a thing for Flux? Its just that Flux blows everything out of the water at this point

#

also wondering if anyone has trained a lora for flux and if it can be done on 24gb VRAM

atomic mortar
#

i think i mentioned it above but:
its a web-ui thing the seamless tileable

upper plinth
#

flux is only on comfy tho

atomic mortar
#

nope lol

lucid bobcat
atomic mortar
#

A1111 is pretty outdated

upper plinth
#

damn yeah I havent used this in a while

#

does forge have any speed advantage over swarm

lucid bobcat
upper plinth
#

how? which UI?

atomic mortar
#

some like the UI of forge

lucid bobcat
#

You can do it in Comfyui. Basically overlap the sides of the latent after each step and blend.

atomic mortar
#

hes on A1111

upper plinth
#

im open to updating to superior UIs

#

havent used this in probably like a year

atomic mortar
#

i do NOT recommend comfyUI as a person who isnt really familiar with this/noodles

#

as its node based

upper plinth
#

Im already on comfy but yeah I have no idea how to do what mario is saying

atomic mortar
#

SwarmUI lets you not use the noodles while still having its advantages

lucid bobcat
atomic mortar
#

SwarmUI has such a node :^)

#

SwarmTileableVAE