💬｜general-chat

[fervent thunder]

Chatting about it is cool. But there are so many unqualified people on the internet who act qualified

[desert dagger]

you're justifying your actions. that's not what your first question was. there are laws in place for using someone's likeness without their permission. suggest you contact an intellectual property lawyer

[fervent thunder]

I think even qualified people have a hard time with it. The closest thing I can see online is andy warhol

I have not been a practicing attorney in years. But that is not how it works 🤣

[fervent thunder]

I cant make political commentary without using her likeness,

[fervent thunder]

thats exactly within my budget, you're hired. I will give you 20% of all winnings!!

[desert dagger]

i don't care, okay? you want to know what can happen to you, ASK a QUALIFIED lawyer

[fervent thunder]

I just hired an attorney, let me ask

[desert dagger]

you just hired someone that said they aren't practicing and that already gave you the only advice you're going to get from them. go ask a lawyer. one that's practicing and is familiar with the actual point of law you are interested in.

[fervent thunder]

And you have to be careful with attorneys that cover this particular field. There just aren't many of them. Not any money in what you are asking about.

[fervent thunder]

see we each get something out of it

Just saying I have known a few over the years, and I would not have hired them to do my laundry. So do your research.

[desert dagger]

or wind up owing someone a lot of money and possibly sitting in jail, your choice

[fervent thunder]

in the short term I have removed the posts, as far as ethically I feel like what I did was protected as parody. As far as legally, I know thats an uphill battle

I feel like when you say someone is beyond parody, beyond critisim, thats a problem

We still are not China yet. So you should be able to express yourself without fear of retaliation.

Or maybe that is wishful thinking on my part.

[desert dagger]

paradoy's a touchy subject too, and you should probalby research whether you, an individual, can actually qualify to say you're creating parody or not

[fervent thunder]

I was a suit salesperson for 10 years, lawyers are blood suckers! thats why you're the best person to hire, you dont want the job. You would actually want to help!

[fervent thunder]

its social commentary, my research lead to andy warhol case

im not selling the art

[desert dagger]

you're an individual. i suggest that instead of arguing and justifying yourself, which is the only reason you are here, that you go do some real research.

[fervent thunder]

im just creating a discussion, what are you the gate keeper?

Yea. There may be some legal websites that could help. I have no idea, but it seems like there are some where you can ask an attorney a question.

[fervent thunder]

thats a sales funnel 😛

[desert dagger]

remember, this is not a legal forum, so you're creating chaos

[fervent thunder]

for sure

I mean like 80% of the loras are someones likeness

where do you think I got kamala harriss lora, so who is really responsible

this is absolutely relevent

and I imagine there will be some legal cases in the future with this, if not already in the works

whats the difference between drawing someone and using ai to create them though?

if im running stable diffusion locally what makes that different than using my eye balls and a paintbrush to copy what I see

That's a question for people smarter than me. But good luck with that because even our smart people are idiots today it seems.

If you live long enough you realize the people making the rules are really not that intelligent. So...

I think the reason I care is because I got banned for "sharing someones nudes" and it makes me really second guess if thats really the case

🤔

Kamala nudes? 🙈

the funny thing is that reddit has a kamala harris nsfw forum, so I think I got targeted for my political commentary rather than making parody

she was not nude lol

imo it was a SFW photo

maybe its racist to imply twerking is nude, shes a strong black woman who can twerk

I guess with the shooting it is an overly sensitive time right now

well this happened a week ago, but yes its highly relevent subject

one interpretation of my social commentary could be one of empowerment. Another one is sarcasm. Kamala harris made an iconic apperance on BET which inspired my social commentary

ok, back to my laundry, good luck !

😁

[fervent thunder]

thats my job now!!

[grave stag]

What's the best write a text prompt like female centaur in a forest with black hair covering her chest, holding a spear in an anime style and bam outputs a result that is good I can use for rp refs? If its stable diffusion I'll learn how to use it but last time I tried o failed miserably on getting anything I could say oh yeah that is 100% anime character from show I watch with the right LORAs. If its a different thing to use then I'll look at it and 100% try it if they have a free option or a free trial.

[loud solar]

Any experience with AuraSR upscaler?

[fervent thunder]

yeah it was evaluated on the upscaling discords and its bad

the best upscalers are still ATD/RGT/DAT2/HAT-L currently

[loud solar]

Thanks!

[fervent thunder]

which are literally just very fancy and fat versions of Swin-IR

the best thing in the world is probably a massive private GAN somewhere
its not really profitable to train GANs though

but GANs are amazing

[loud solar]

Long Wikipedia article ... a lot to learn 🙂

[fervent thunder]

[fervent thunder]

wonder if sd works on redhat with amd

yeah rocm

[fervent thunder]

theres a ton of various little things as well beyond rocm that u need

like for example I was not able to install ROCM on debian at all

[pure orbit]

can you download a model to generate Music locally on your own pc like we do with StableDiffusion??

[fervent thunder]

is sd3 proprietary or will there be checkpoints for it? I saw some discussion about the licensing for it

[desert dagger]

sd3 medium (2b) weights are open source. but as it's unfinished, creating checkpoints for it (that work correctly) is not that simple

[pale latch]

it has commercial restrictions. any commercial use and you need to pay the stability membership fee . if your company makes more than $1m in revenue then you need an enterprise license

[median spear]

Hey, can anyone make me an AI image? Or at least help me try to make one? I’ve been trying to find someone that can make it for days now lol

[fervent thunder]

what do you need made

[fervent thunder]

good to note, thats a problem I would be happy to have

[median spear]

I’ve been trying to find someone to be able to turn this album cover from Ken Carson to beast boy. Here’s sort of what I’ve been wanting

Nevermind lol, I can’t send pictures here for some reason, but it’s the “A Great Chaos” album cover

[wet grotto]

hi there

exist contronet union to SD 1.5?

[wise imp]

How to generate images?

[fervent thunder]

what?

sometimes what we imagine dosent work in reality

[fervent thunder]

takes work

https://github.com/AUTOMATIC1111/stable-diffusion-webui?tab=readme-ov-file#installation-and-running

[median spear]

Can I DM you the images? I can’t put them in here for some reason

[fervent thunder]

#🏞｜general-with-images

[median spear]

Ok, I’ll send them there

[dark oriole]

I've got an annoying, seemingly random issue where it'll make "twins" of a character/subject and no prompts seem to remedy this.

Example: I'll prompt a picture of like idfk, Sephiroth or something, and it will just put 2 of him into the image, ruining it.

Any way around this?
I've already using Hires Fix, was told it'd help cut down on this but it hasn't

@ me if you know anything

[wicked gust]

has anyone been able to generate QR's using the new controlnet++?

[desert dagger]

what aspect ratio is the image you're trying to create?

[fervent thunder]

that happens when the size is too big

[dark oriole]

Different aspect ratios, but typically 16:9 or at the very least in a landscape orientation.

I know the AI will just... do whatever and try to fill space sometimes but is there really no other way to at least suggest to it "Hey, I only want one character here"?

[desert dagger]

that would be why. the AI is trained on 1:1 ratio images. so when you have a ratio like 16:9 you just gave it 2 square image tiles to draw in. so it draws one subject - in each tile

no, you can't prompt that out. you can use out painting later to expand the image if you want

[fervent thunder]

you have to use outpainting once you get an image you want

some checkpoints do better with bigger sizes, btw

I start with 1024 x 1024 on a lot of the good checkpoints, and depends on the prompt it does work

[dark oriole]

Outpainting has been a very mixed bag for me, but guess I'll re-learn how to do that -.-

[fervent thunder]

yeah I dont use outpainting yet myself

[desert dagger]

you can also inpaint out what you don't want. just select the second character and prompt something like trees or landsape over them

[dark oriole]

It's needlessly complicated

[fervent thunder]

its dependent on the checkpoint as well

[desert dagger]

i use photoshop's gen fill to do that with

[fervent thunder]

I feel like with time the check points willl continue to improve

[desert dagger]

they will, but it'll help to also use SDXL not SD1.5

[fervent thunder]

yes I use most of those but there isnt as much selection

[dark oriole]

Is SDXL just better?

[fervent thunder]

seems like it

looks more like midjourney lol

generally speaking try to read the checkpoint and lora notes, you want to try to match their settings to some extent

like if its a virtically trained drawing your gonna get poor results generating a landscape picture

[desert dagger]

depends on what you want. it was trained on 1:1 ratio images too, but much larger than what 1.5 was trained on, and it doesn't have as much of a problem with wanting to create stuff in each tile if you ahve a ratio that's not 1:1 like 1.5 does

[pale latch]

pretraining was square ratio but later stages had bucketing on sdxl

[autumn oyster]

hi, I'm pretty new to this. Can anyone explain to me what the difference is between the stable diffusion verison on github and sdxl is?

[desert dagger]

SD_XL is one version of stable diffusion

[autumn oyster]

is sdxl the one that most people use? apparently from what I read it has creates the highest resolution images

[desert dagger]

that's a very simple question with an incredibly complicated answer, actually. a lot of people use SDXL - i'm sitting here right now using SD 1.5.

[autumn oyster]

and why not sd 2.1 or 3 or whatever?

[desert dagger]

2.1 has some problems. i don't like fighting with it. SD3 2b is what i've used almost exclusively since 2b was released, but i have some prompts that only work correctly in 1.5 and i was using one of them, so that means using 1.5

SD3 2b (medium) is my prefered model now, however

[autumn oyster]

are certain models based on certain versions of sd?

[cedar salmon]

yes

[tulip yarrow]

How many people do you think will still use SDXL after they release an SD3 Lora trainer

[desert dagger]

everyone that has prompts that only work in sdxl

[pale latch]

there was code to train loras on day one. people are just figuring it out still

i made a lora but its balls 🥹

[tulip yarrow]

imagine if it's like soccer balls 😭 (i know what you mean)

[raven scaffold]

hi

[tulip yarrow]

Alright so for some reason civitai rated one of my (non-X-rated) videos... X-rated

It was just a generation process

[ebon skiff]

How to create an image by prompt

[ebon skiff]

Rip griddymastah

Tq

I am new to discord

I don't know how avail all the things

Tell me one thing

It's premium? Or opensource

Tqqq

Ok I agree I am a noob

But using foul language is the biggest thing

I know

It looks like you

If you know the things to share

Tq

Go away

[desert dagger]

reported to discord and the mods for spamming and harassment

[ebon skiff]

@frigid sinew which standard you are studying 🤣

[desert dagger]

because that's what you're doing.

[ebon skiff]

Appreciated @desert dagger

[desert dagger]

no problem.

i care, why?

reported for insulting and harassment

[ebon skiff]

Can someone please remove this cockroach @frigid sinew

[desert dagger]

@ebon skiff dont' engage with him

[ebon skiff]

Keep all your accounts in your a** hole

Ok @desert dagger

[echo kite]

is there anything like diffusion toolkit for linux?

[echo kite]

nvm the guy who made it also made a crossplatform version super recently

[fervent thunder]

are there any mods in this discord? lol

[warm junco]

Banned him

[desert dagger]

thank you

[acoustic dagger]

Guys just need help with something

After you've closed Llama 3 LLM how do you re-open it?

[lilac forge]

why is my wildcard not showing up?

I make the prompt like this:

score_9,score_8_up,Pony YeiYeiArt Disney Princesses, lora:Artgerm_XL_PONY:0.8,

Below is a snippit of the wildcard I'm trying to use:

Pony YeiYeiArt Disney Princesses
lora:AnastasiaXLP_character:0.8,AnastasiaXLP,short hair,red hair,blue eyes,coat,scarf,ponytail,fingerless gloves,hat,yellow dress,ponytail,black dress,bare shoulders,necklace

lora:ZeldaMarinComission_Character:0.8,MarinXLP,hair ornament,pendant,hibiscus,blue dress,red sash,orange hair

lora:MaleficentXLP_character:0.8,MaleficentXLP,colored sclera,green eyes,colored skin,green skin,horns,cleavage,cape

lora:SarahHawkins_Character:0.8,SarahWaifu,brown hair,long hair,blue eyes,white shirt,red apron,mob cap,pink nightgown

But it just shows up like this (in other words, there should have been one of the entries where the wildcard's name is)

score_9,score_8_up,Pony YeiYeiArt Disney Princesses, lora:Artgerm_XL_PONY:0.8,

And I do have the Wildcard and Dynamic Prompt extensions installed.

Also, when I get rid of everything in the prompt but the name of the wildcard, like this:

Pony YeiYeiArt Disney Princesses,

It does actualy work, and this shows up (it's a random pull from the list in the wildcard text file):

lora:SnowWhiteXLP_character:0.8,SnowWhiteXLP,puffy sleeves,dress,yellow skirt,corset

[bleak matrix]

Good morning, everyone! How are we all today?

[warm junco]

What Programm do you use?

[acoustic dagger]

Windows CMD

And I've installed Llama 3 on the official website

[warm junco]

Did you installed ollama?

[acoustic dagger]

Yes

I just want to open the program to type to

[warm junco]

Then you can open up a cmd and type
ollama run llama3

To start the ollama service (if its closed by accident) you have to type
Ollama serve

[unkempt hatch]

where can I get controlnet blur for 1.5?

[mortal sphinx]

hi

[boreal bay]

Hello, everyone. I am very new to stable diffusion. I really hope you could help me.
There is an app called AI Morph (the one that was made by Daily Joy studio). (It's not and add, cause this app sucks)It was great, until now. This app lets you transform any picture in anime style. It was the best amount the other, accurate in everything and detailed. Very beautiful. But now it doesn't support NSFW. When i put some NSFW there it says "The result may not meet our guidelines. Please try something different." Can you please tell me, if apps like this are using something like stable diffusion or are they creating their own ai? Cause their pictures was the best in everything. And now I want to understand can I make something like this in Stable Diffusion?

[desert dagger]

just use one of the anime loras from CivitAI for stable

[boreal bay]

Is there a way to control how accurate a generated image will be to original one? I mean, I want a picture with same colours, clothes, pose, facial expressions, everything exactly the same but in anime style

[desert dagger]

sure. first question - how are you running stable diffusion?

[boreal bay]

On windows, thought bat file that opens SD in browser

[desert dagger]

you're going to want to watch youtube tutorials on stable diffusion control net

and style transfer

[boreal bay]

What would you recommend? What to start with?

[desert dagger]

i would suggest you start by watching https://www.youtube.com/@sedetweiler everything on his channel. and once you've finished that, you'll know more what direction you want to go

[boreal bay]

Okay, thanks!

[pseudo bough]

hey i havent been utilizing ai tools the past 3 months much what are some of the biggest advancements that have come out

do we have a reliable video creator now

[autumn oyster]

when i'm using a model, should I be using the tags that the model was trained on? I notice that some of the really good pictures I've seen mainly use the tags and not some random words

[pale latch]

depends on the model. some train the text encoder hard. some don't.

some have stronger keywords. some are more generalized

if you're using anything from ponyxl, there are some tags you need or else eveyrone will call you a newb and laugh.

style_up_9, style_up_8, style_up_7, style_up_6 . I think . then the lower numbers in the negative. if you don't do this you're a total poser

[desert dagger]

you'll also have really poor quality images

[sage reef]

im kinda sad that Scott is making less videos these days, I know he is very busy, i always enjoy his vids 🙂

[desert dagger]

agreed, though i wish he'd show up here once in a while

[wet grotto]

hey, T2I adapters are only to XL?

[desert dagger]

spammer

[pale latch]

yeah i have no idea how to use it

[desert dagger]

@frail sonnet can probably explain how to use the pony models

[signal burrow]

hi, how can i run the face-to-all?

the sdxl

[knotty steppe]

Hi, I am new here and start learning how to create pictures, I am wondering where I can share them after creation. Is there a social networking website for gen pics?

[stiff valve]

I recently build a tool for Local AI and integrated SD into the interface, as well as LLM support (both local and online). I started a readme, but need to update it with the additions I've made the last few nights. There is also an ai-compose script that is built for podman, and for team red (AMD) users. I hope more people find it useful, as I have made it my go-to tool now
https://github.com/CaptainASIC/AI-Garage

[tiny lintel]

Is stable diffusion paid?

[stiff valve]

only if you're being scammed

or if you're using a cloud run, i guess you gotta pay for time, but the aoftware of stable diffusion is free and open source

[buoyant merlin]

SDXL require more vram, correct?

[stiff valve]

kinda, but it really depends on your models and workflow imo

[buoyant merlin]

i wonder how further can i push it with 20gb vram before it explodes

[hearty prairie]

A tiny dancer crafted from noodles and leaves performing an elegant ballet on a paper-crafted dance floor, surrounded by a soft morning fog in a lush, green woodland, with dewdrops sparkling on the surrounding leaves

[stiff valve]

20gb is pretty good, you can get pretty far into prompts

[buoyant merlin]

ah nice, the card im running is pretty much the workstation version of the 3070 ti but with 20gb of vram slapped on it

the rtx a4500

havent tried it out throughly yet

[stiff valve]

i have 24gb and only really get into trouble when i try to make 2kx2k, or batches of 8 images at a time, etc

[buoyant merlin]

interesting

4090?

[stiff valve]

7900xtx

[hard kernel]

Which is better 4090 or 7900xtx both have 24 gb vram

[stiff valve]

personal preference imo

team red or team green

i guess OS would have soem impact, in linux, aobut the same, in windows, nvidia is probably a little easier, but since ROCm for windows is now available, that's leveling out too

[hard kernel]

7900xtx is half the price of 4090 + available in my area which awesome

[stiff valve]

part of the rason i went team red

i can run 2x 7900's for the rpice of one founders 4090

[hard kernel]

Yha stable Swarm ui enabled us to do that

[stiff valve]

i havent bought the second card yet, but its on the list

[dusty jasper]

Guy's I'm curious what Ai do they use in generating Ai images of Dwayne and Arnold that trending on YouTube

[tidal sleet]

Hey guys! Not sure if this is the correct channel for the following message. Please advise if not.

New to this field and the information is a bit overwhelming.

I'm looking for a consultant to accelerate my learning and help build a foundation for generative AI in which I want to apply to my architectural design and visualization process. Also looking for those confident in the aid of producing visualizations. I don't have an exact budget, but I'm happy to work out an hourly rate.

Based on my research so far, I'm looking for the following:
(They do not need to be solved by 1 consultant or 1 session. If you have solution or specialization in 1 of the specific areas, don't hesitate to reach out.)

Comfy UI setup and optimization for desired results of:

1. Architectural Smart Inpainting.
2. Setting up connection to cloud GPU
3. Exploration on ideal "models". Possibly training our own.
4. 3D Box editing/ Depth Conditioning
   Through comfy UI if possible, however it's fine if it's not possible. We would like to integrate 3D programs or images from 3D programs to allow for precise attribute prompt editing using "loose control net" and 3D "box editing" https://arxiv.org/html/2312.03079v1

Feel free to reach out to me for any questions or interest

[late solstice]

I participated in an artificial intelligence film competition organized by Civitai. I created the images, videos, sfx and music entirely with artificial intelligence. You can watch it from the link below. If you like it, I would be happy if you vote 🙂

https://civitai.com/images/19990254

[frail sonnet]

The trick to using Pony is the same as all other checkpoints, use loras to make it decent. This is especially true with Pony. And/or use Autismmix or Duc Haiten's Pony instead. Most people seem to use "Score_9, Score_8_up, Score_7_up". Some even put the lower scores in the negative prompts.

if for some odd reason you don't want nsfw (prob don't want to use pony in the first place in that case though lol), you would add "rating_safe" it works sometimes.

I just learned something new today, apparently you are supposed to use it with clip skip 2 or higher.

@pale latch

[fervent thunder]

Does fooocus have api ?

[fervent thunder]

☕

[remote roost]

Guys how to make Stable Diffusion black

[stiff valve]

what?

[remote roost]

I just want to make Stable Diffusion Dark and I did It then no matter

[glacial cloak]

Hey, is anyone else having trouble with sdupscaling?
I just started with local installment 2 weeks ago and saw that you need a comfy ui workflow to really get what you want and just downloaded a workflow.
Now when i go to the manager and install missing nodes everything works fine but not the sdupscale :/ its installed and the list shows it as installed but when i load in the workflow it states that it is still missing
Uninstalling and reinstalling wont work
Been trying for 2 weeks and by now im pretty desperate

[fervent thunder]

are you referring to this ?
https://github.com/ssitu/ComfyUI_UltimateSDUpscale

[glacial cloak]

Yes

[fervent thunder]

try install directly from github

[glacial cloak]

I installed it via the mod manager though is that a problem?

Ok thx i will try out

[glacial cloak]

hm, ive got home now and followed the instructions from the github
it just said copy the link and paste it into URL for extension's git repository, which i did but this time the upscale does not even show up on the manager anymore :3

i found it on the normal extensions page, but its not in comfy ui

[fervent thunder]

what happens if you go in comfy manager

and use the "install from git url" ?

do you see an error window pop up?

and does the terminal say anything?

[glacial cloak]

yes it says it already exists

aka it says check in terminal and the terminal said it already exist

[fervent thunder]

are you confusing SD upscale with SD ultimate upscale?
because that happens sometimes

also what happens if you physically download it and place it in the folder

[glacial cloak]

i think its the ultimate i need
whats the difference? didnt knew there were two

[fervent thunder]

they are just two completely different things with similar names

[glacial cloak]

just checked to be safe, yes it is the UltimadeSDUpscale

[fervent thunder]

did you try to physically download it and place it in the folder

[glacial cloak]

not yet

i guess

[fervent thunder]

that would be a great idea to try

[glacial cloak]

in what folder does it belong if done that way?

dumb question im sry :3

[fervent thunder]

ComfyUI/custom_nodes/

[glacial cloak]

ty

hm, but it still say not found

[fervent thunder]

did you restart and refresh?

[glacial cloak]

yes

[fervent thunder]

someone on github said this:

https://github.com/ssitu/ComfyUI_UltimateSDUpscale/discussions/65

[glacial cloak]

if i press install missing nodes it is listed there but its already installed so i can only uninstall, update disable it

[fervent thunder]

"For some reason there is a conflict between this and reactor node, which I don't have myself. Try disabling one."

what happens if you try to add it to workflow? its just not there?

[glacial cloak]

yes its just a red window

what is meant by reactor node?

[fervent thunder]

they mean this
https://github.com/Gourieff/comfyui-reactor-node

what if you delete the red node
and then just try to add a new node
and search for it

[glacial cloak]

the upscaler is not on the list :/

[void jasper]

can you run stable diffusion without the python, pytorch, miniconda, and other stuff your system downoads and takes huge load in diskspace? Like could you run stable diffusion over your java IDE instead?

getting bit of tired that the whole python stuff, which i only installed for stable diffusion is like 50 GB of mess

[fervent thunder]

no its not possible

that's just the nature of python

[void jasper]

dont like that, feels like it auto installs a million hidden programs onto my pc

and then i wonder why disk space is gone

[fervent thunder]

how much disk space does your system have in total?

[void jasper]

512gb

while the system ssd has 230

miniconda alone is 16gb

[fervent thunder]

there are probably ways to optimise it
just looking at the list of dependencies for comfy UI
it feels like this could be done in under 50GB for sure:
https://github.com/comfyanonymous/ComfyUI/blob/master/requirements.txt

torchsde
torchvision
torchaudio
einops
transformers>=4.28.1
tokenizers>=0.13.3
sentencepiece
safetensors>=0.4.2
aiohttp
pyyaml
Pillow
scipy
tqdm
psutil

#non essential dependencies:
kornia>=0.7.1
spandrel
soundfile```

[void jasper]

i dont even use that ui

i have the old original thing with the 1.0 models

[desert dagger]

most of the diskspace is probably gone because the size of the models, not the stuff like python

[fervent thunder]

yeah I am not seeing something massive in that list of dependencies

[void jasper]

yeah still. not a fan of installing a whole programming language, just to run 1 thing 😄

there has be an option somehow to make image genration on your local machine without python

[fervent thunder]

haha wow someone has done it
https://github.com/leejet/stable-diffusion.cpp

there are others, this one has control net working apparently:
https://github.com/axodox/axodox-machinelearning

[desert dagger]

you get to install all sorts of stuff you don't want in order to run anything, but remember - with auto1111 or comfy or the stable webUI - you are programming, you're just using an interface to do it with.

[frail sonnet]

Definitely not, the first time i tried installing it I kept getting wrong python version and lack of Cuda errors lol. Seems it needs those to run.

My comfy uses 334gb, however I have 85gb in my checkpoints folder; 208gb in my loras folder, and 30gb in my output folder

[fervent thunder]

208gb of loras wow

[frail sonnet]

I'll be attending loraholics anonymous next week 🤣

I prob put some checkpoints in the wrong folder tbf lol

[void jasper]

like i want become java dev, and i am curious if you can do all the fancy stuff with AI everyone do in python in java too

cause i am not fan of the commandline stuff with pip torch and all. i rather have my big IDE like eclipse and do my stuff there

[amber bloom]

you should check out PyCharm, it's a nice IDE for python that can automatically install dependencies and such so that you don't have to deal directly with pip

[frail sonnet]

It's not that I don't like coding, it's that it doesn't like me 😉

Do glif or huggingface run on some java setup at all? I love the glif SD options lately 😄

[amber bloom]

But python is the standard for machine learning etc. there's nowhere near the same library support for Java, C# etc

[void jasper]

still, i rather stick to the language i know, then just going python, cause "its hyped"

like python is not even object oriented,

so dunno 🤷

[frail sonnet]

Could you do the auto install stuff using the C etc. that it comes with, then so all the additional addons and finetuning in your prefered programming language? This is completely internet conjecture, I know nothing about such things 😄

[void jasper]

yeah i find tons of java APIs to use some webservice.. but there you obviously need buy credits...

so thats not realy an option

need a Local option that runs on PC xD but looks like there is only stable diffusion for that

[cedar salmon]

a search shows some people i dont know the right terms im not a programmer, are importing pytorch to java and go from there

[cedar salmon]

its about direct support not hype

[void jasper]

na the whole AI thing, for some arbitrary reason everyone use python

[amber bloom]

Protip: Software development if all about being flexible and willing to pick the right tool for the job, even if you have to spend time learning it.

[frail sonnet]

SD has lots of competitors rn, but most of those are based on the SD model...

[fervent thunder]

there was a flip

after 2010 or so

machine learning was all in C/C++

and then it became python

after scikit-learn and tensorflow etc

[frail sonnet]

Java Libraries for Image-to-Text

OCR Libraries: Java has robust libraries for Optical Character Recognition (OCR). These are excellent for extracting text from images:

Tess4J: A Java wrapper for Tesseract OCR, a powerful open-source engine.
Apache Tika: A toolkit that includes OCR functionality alongside parsing various document types.
Deep Learning Libraries: While not specifically for image-to-text, you can use deep learning frameworks in Java to build your own models:

Deeplearning4j: A popular Java library for neural networks.
ND4J: A scientific computing library for Java, providing the numerical backbone for deep learning.
Using Stable Diffusion with Java

Java API Wrappers: There might be community-developed Java API wrappers for Stable Diffusion. These would make it easier to interact with SD models directly from your Java code. Look for projects on GitHub or other code repositories.
REST APIs: If a Stable Diffusion model is available as a REST API, you can easily send requests from Java using standard libraries like java.net.HttpURLConnection or libraries like Apache HttpClient.
Choosing the Right Approach

OCR vs. Deep Learning:
OCR: Ideal for images with clearly typed or printed text. It's simpler to implement.
Deep Learning: Better for complex images or handwritten text, but requires more expertise to build and train models.
Existing Libraries vs. Custom Development:
Existing Libraries: Faster to get started, but might not be perfectly tailored to your needs.
Custom Development: Gives you full control, but requires more time and knowledge.
Example: Using Tess4J (OCR)

Java
import net.sourceforge.tess4j.*;

public class ImageToText {
public static void main(String[] args) {
Tesseract tesseract = new Tesseract();
try {
String text = tesseract.doOCR(new File("your_image.png"));
System.out.println(text);
} catch (TesseractException e) {
e.printStackTrace();
}
}
}
Use code with cauti

I asked Gemini advanced lol

it does lie sometimes though

[fervent thunder]

you wanted to ask for text to image

[frail sonnet]

I did

damn gemini advanced

[fervent thunder]

🤣

[void jasper]

yeah i can try that

[frail sonnet]

text2img

[fervent thunder]

I had the same experience with Gemini

[void jasper]

probably need some OXXN or so

[frail sonnet]

Gemini advanced has even made me fake comfyui workflows when I asked it to make me some. They were in JPG format ROFL

I had asked it this, shouldn't have used short form words I guess "if someone doesn't like C++ and python, but prefers to code in, and use java, is there an img2text comparable to table diffusion that they can use? Or a way to use SD via java?"

I think Claude 4o lies far less

It's extra funny when it helps me write code for stuff! It does apologize a lot when the code doesn't work though 😄

[cedar salmon]

i dont have a thing with them, curious are their older models still available? like 2.1

[fervent thunder]

probably not

yeah it looks like Claude 2.1 is gone forever

[frail sonnet]

how about via freedomgpt?

[desert dagger]

?

claude is right here https://claude.ai/new

and working just fine

[fervent thunder]

it is but

only 3 or 3.5

they don't offer 2.1 any more sadly

[wet grotto]

what re you talking about, LLMs?

I like a lot 🙂

I use then with kobold and silly tavern

[fervent thunder]

nice

I do everything in cloud

but silly tavern is cool

[wet grotto]

cloud?

[fervent thunder]

not sure how to explain cloud

[wet grotto]

is a backend, frontend? do you have link?

[fervent thunder]

I use AWS and Azure

[wet grotto]

ahh ok I understood, nice I never used it

[fervent thunder]

there's Google's one now too

its growing

but IDK much about Google's version

[wet grotto]

a backend from google? never heard about it

[desert dagger]

i wouldn't either. it doesn't work as well. why offer something that most people aren't likely to use but that will tie up your compute time?

[wet grotto]

do you like claude?

[fervent thunder]

yeah I understand its not worth the compute

[cedar salmon]

i mean they started off not even reading what you wrote

[fervent thunder]

its a bit like stable diffusion checkpoint or lora
its different flavours

and sometimes you want a certain flavour for a certain thing

[cedar salmon]

why use 1.5 when we have 3 duh

[fervent thunder]

when 3 came out

it was way stronger but

the writing style is not the same

[cedar salmon]

ya 2.1 was way more creative

[fervent thunder]

there's currently not anything that can mimic 2.1's writing style exactly

I think they should open source their older models over time

[wet grotto]

gemini is free api, btw

[desert dagger]

yes. use it when i need an LLM. usualy opus, once in a while the sonnet model

[desert dagger]

ummm what?

[cedar salmon]

its quite clear and i agree

[wet grotto]

the LLM from google? is gemini and is free api

[fervent thunder]

not the LLM

what I mean was GCP

its the Google version of AWS and Azure

[wet grotto]

ah ok

[fervent thunder]

its a bit hard to explain but they are like "managed platforms"

with a big mixture of things

[wet grotto]

well it interest me a lot but I need to make a class of this things to learn

I'm architect, it's not my area

but I love it

[fervent thunder]

if you can use stable diffusion then that's already more than most non-techy people

so you are off to a decent start

[wet grotto]

thanks 😄 hahaha

[desert dagger]

claude is anthorpics, it's never been open source. curious how they would have gotten their hands on any version of it. i'd be really careful if i were you

[north bronze]

Hi, is it possible to run stable diffusion in an i5 laptop without Nvidia or AMD GPU? thanks

[wet grotto]

if you cant try colab
https://colab.research.google.com/github/TheLastBen/fast-stable-diffusion/blob/main/fast_stable_diffusion_AUTOMATIC1111.ipynb

[fervent thunder]

yes absolutely, in CPU mode

[frail sonnet]

github is awesome! Also huggingface has Taesd and flash instances you can try for free.

[fervent thunder]

it may take 20 minutes for an image but it will work with just CPU

[north bronze]

thanks for all the replies, I'll try them out

[fervent thunder]

its not that bad waiting long time for image
I actually wait like 30 minutes even though I rent data center GPUs
because I choose to use slow samplers

it gives you time to work on the workflow more

[wet grotto]

you can do a image with 30 secs on colab, try it

[fervent thunder]

not with the samplers I am using

[wet grotto]

sorry I confuse you with CardiB XD

my fault

[fervent thunder]

lol

its ok

[frail sonnet]

Does anyone know how I can check a safetensors checkpoint I made, to see which version of SD, it is thought of as?

[fervent thunder]

not sure, when I searched on that I just found another person asking the same question 🤔

[frail sonnet]

Gemini advanced didn't give me any useful replies unfortunately

[fervent thunder]

comfy in general isn't documented

and you have to go read the code

its pretty rough

[frail sonnet]

I merged Pony and SD3, in theory. I know SD3 is in there somewhere since no nsfw lololol, but aside from that, I have no idea how merged things really are.

[pale latch]

you brought the pony xl tencs over to sd3? it'll probably just disalign everything in the clip layers is all. there's no reason that would bring anything from the unet to sd3

none of the alternate tencs i've seen seem to do very much improvement. the t5 is there being prompted still and i think that'll do most of the heavy lifting in these cases. a proper test would be to run sd3 with only the clip layers and no t5, and see if alternate tencs merged in do anythign then. Other people have been doing this since release day and i've not seen any valuable results there

you don't even have to merge the models. you can just extract the clip layers from sdxl models and load them separately in comfy.

[chilly token]

Fucking tensor size errors man

[thorn hare]

is there some formula what model size should be used based upon gpu vram and regular ram size ?

[pulsar venture]

Hi, how dora's quality is going?

[lean timber]

Does anyone know of a LoRA or model which is good at producing crosshatch illustrations, similar to this style?:

https://i.imgur.com/hJuO8JS.jpeg

Or is there any good up-to-date guide on how to train such a model myself? I did train a LoRA using a jupyter notebook like a couple of years ago, but forgot the process lol

[rustic willow]

klingon death metal video for odyssey project

https://civitai.com/images/19928073

udio music, comfyui SDXL wf, capcut

vid2vid mostly with clips from Psycho Circus - KISS as the driving vids

[fervent thunder]

dora is only mildly better than lora

[pulsar venture]

But better

[fervent thunder]

I think its better to train a full checkpoint model

[ruby drift]

(new to generative AI in general here) I'm a bit confused about how prompts work. Let's say you use "harvesting potato", but want a potato harvesting something. How can you prompt for that nuance? The way prompts are handled makes me think they're keywords only, and not sentences.

[fervent thunder]

it depends on the model

some are more like keyword tags

some really are sentences

they are much more limited than text models though

this nuance you talk about it called "coreference resolution" in machine learning

T5, which is the text encoder for SD3, can definitely do this

take a look at this

https://huggingface.co/jtlicardo/flan-t5-small-coref

its a T5 fine tune, specialised on the task of coref

[ruby drift]

Alright, I'll get in reading!
Is huggingface reliable? I saw something there yesterday and didn't feel like creating an account right away

[fervent thunder]

what did you see?

they are not microsoft/google tier

but they have been around for a bit now

and made some libraries that get used a fair bit

mainly the huggingface transformers and diffusers

[ruby drift]

Alright. I'll try to assimilate that information!

When running on a lower VRAM system, if you don't get an OOM exception, is the result affect in any other way that the time it takes to generate?

[fervent thunder]

just time

the result has errors
but they are not affected by memory

you can reduce errors by using solvers with a higher order, solvers with multistep, and solvers that have ways to deal with stiffness. But you don't have to learn about that for a while you can just use the default solvers

[ruby drift]

(running automatic1111 on my older PC since I don't want to install too much experimental stuff on my work PC)

[ruby drift]

Ok, I'll keep that chapter for later then!
Thanks again

[fervent thunder]

automatic1111 won't actually let you load fancier solvers anyway you need comfy ui or diffusers for that

[ruby drift]

Alright, I'll give that one a try

[craggy kindle]

question, would my gtx 970 be able to use a SDXL fp32 model?

i read that fp32 is more vram intensive but that the quality is better

[ruby drift]

Let's say you're satisfied with a generated image, but would want stuff generated "around" the image to make it larger (so akin to zooming-out), how would you do that?
Use the generated image as a latent image, but somehow set that image as a part of a blank latent image?

[desert dagger]

outpainting

[ruby drift]

Great, thanks

In something like ComfyUI, if I fix the seed, it's pointless to increase the batch size right?
I would need to feed in a fixed seed, then set the KSampler to increment?

[desert dagger]

yup. it'll randomize the seed for anything other than the first one anyway

otherwise, all you get are multiple copies of the exact same image

[ruby drift]

Oh! So I can't work on multiple at a time. I guess it's not a usecase though. Increasing the batch size seems good for "discovery" rather than for actual refining.

[desert dagger]

if you're outpainting, you don't want to do more than one at a time anyway. with tweaking you need to just work on one till you're happy with it, and then save it and move to the next

[ruby drift]

Yeah

So to outpaint you go from KSampler -> Decode -> Outpaint -> Encode -> KSampler?

[desert dagger]

well - if i'm doing it, i save the image, pull it up into photoshop, and outpaint there...

[ruby drift]

Oh, you don't generate the outpainting?

[desert dagger]

I am, photoshop's gen fill. I don't want to do that to everything just automatically, i'm only doing that to the ones i really need that for

[tulip yarrow]

Does civitai support multi-concept loratraining

Can you do it within a single dataset but using different trigger words for that image

[mortal sphinx]

Hmm..

You guys think it’s possible to train for a specific character, do that twice, then start pumping results with just those two characters generated together?

[astral snow]

Hello best discord servers for learning how to run and use AI locally on my machne easily and the best programs?

[desert dagger]

this one

[astral snow]

Mkay

Easiest way to get sillytavern up and running on my manjaro OS with all the features without having to type in a billion commands?

[desert dagger]

@warm junco might know

[astral snow]

Hmm

Cheers

[little saddle]

join this server now the frist 3 prople that ask me to join my discord channel i will give admin to

[still glacier]

yes

[still glacier]

Depends what you're looking for. Something easy to use, something complex but powerful, a middle ground. Depends of your hardware also.

[astral snow]

Got Nvidia Geforce RTX 4070

So GPU should be good

[still glacier]

yes

[astral snow]

Easy to use for sillytavern friendo

[lusty creek]

gm

[still glacier]

never used sillytavern, but If you want something easy give fooocus a try https://github.com/lllyasviel/Fooocus and for something a bit more complex but way more powerful auto1111 https://github.com/AUTOMATIC1111/stable-diffusion-webui/

[light fractal]

I have a question. There is a function for inpainting an object into the selected area based on a text description, but is there a way to inpaint an object based on an image? For example, I have a photo of the table and a photo of the cup and I want to add the cup from the photo to the table photo, is this possible?

If this cannot be done with stable diffusion, are there other neural networks that allow this to be done?

[fervent thunder]

PowerPaintV2 node with BrushNet SDXL
combined with canny edge and depth map control nets
and some IP adapters for style and colour etc
feeding an attention map to the IP adapter would also help
and if you want to, you can guide the model further using your input image by placing the cup on the table and covering it with velvet noise

if all of that doesn't work then you need a lora

[wet grotto]

hi there

the difference between a normal model and a inpaint model is just to use inaint on I2I with inpaint?

[fervent thunder]

yeah

otherwise they are same

[amber bear]

福建的传统土楼，斑驳的外观，在明亮的月光下，有护城河，门前有老人，有小狗

[wet grotto]

but I use inpaint model only when I use brush? is it?

[unkempt hatch]

why is it that when I watch the image generation, controlnet doesn't seem to kick in until 40% of the way in

even though I have it set to start immediately

[wet grotto]

what?

[fervent thunder]

in the initial steps, guidance in general doesn't do that much

you can actually turn off guidance for the first few steps and get a similar result

[unkempt hatch]

Why is that? It seems like it should be the opposite

Because what ends up happening is ghosting. 40% it doesn't use controlnet, then when controlnet kicks it, it makes a completely different image and I get two overlayed, extra ghost arms etc

[fervent thunder]

there was a paper that looked at trajectories
and turned off the guidance at different steps
and that's what they found- that guidance in the initial steps and the last steps wasn't useful

its best in the middle

if you are getting extra arms the best fix for that is hidiffusion

or koyha deep shrink

or generate at a lower resolution and then upscale

anything to shrink the latent

also more steps for lower sigmas

[mortal sphinx]

Wow really? How can I start?

[lyric vault]

hi!!!! i am Omaley nice to meet you all!!!!

there is any good tutorial about Stable ?? if it was in spanish it would be great!!!

[still glacier]

First I'd just look on civitai or other websites to see if someone didn't already train a lora for the character you are looking for.
And when it comes to prompting multiple specific character, (one of) the easy approach is to use some kind of "regional prompting" tool so you can fit (at least) two prompts in one image. One for each of your character.
eg of guide + tool of regional prompt https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111/wiki/Regional-Prompt-Control (very ""anime"" examples, so make sure you're alone when reading it ^^"), you can also use controlnet for that but it might be more complex.

[unkempt depot]

I am trying to use StableDiffusion AnimateDiff in forge (forge is basically A1111 with extra features). I am using canny as controlnet and using the animatediff extension. I uploaded a video under the animatediff extension and then tried to click generate but got the huge following error:

Wait, the error is so big I can't upload it lol, I'll upload it bit by bit instead

Nevermind, this error is far too big, I don't want to spam the whole chat full lmao, is there any other way I can upload this? I tried putting the error on a .txt file, but I can't upload .txt files

[fervent thunder]

You might try making a post on Reddit or somewhere else you can directly link

[warm junco]

You can show it in #🤝｜tech-support
But forge is outdated.

[unkempt hatch]

I don't understand this last part

[unkempt hatch]

But I can literally see it happening. It does like, 12 steps without guidance, and products a completely new image. Then at step 13 I see the controlnet kick in and try to force it to match the lines, but it's already been denoising for 12 steps, so it basically creates 2 unrelated images

[fervent thunder]

I need to look for the paper more, I couldn't find it earlier today. I read it a few days ago

[desert dagger]

you have an artist sitting there, drawing a picture. the first few steps, he just makes large lines and roughs in the image. who cares about guidance. the last few steps, he just tweaks fine details. who cares about guidance. it's the middle when he's actually drawing the image, not just roughing it in or tweaking the details that guidance matters

[fervent thunder]

what is the total step number?

because 12 steps is different depending on the total step number

the sigmas are the main metric because of this

[fair crane]

Hey

[unkempt hatch]

At artist draws a man sitting on the left. Suddenly, when he's 40% done, he's told to draw that man on the right. The finished result is two ghostly men

[desert dagger]

naw, he's been told to draw a man, and handed two pieces of paper, taped together. he draws one on each piece of paper, just like he was told to do

[fervent thunder]

if you are using comfy you can check when controlnet kicked in

I actually prefer to kick it in slightly in the middle

if I find the paper where I got that idea from I will post it

the idea was that in the first few sigmas, controlnet just goes to waste

and then you don't need it in the end

because the end is more about fine detail

[unkempt depot]

Wait forge is outdated?? What should I use instead?

Btw I posted the error in #🤝｜tech-support

[unkempt hatch]

But I'm telling it to kick in at step 0, and it's kicking in around step 20/40

[fervent thunder]

how are you confirming when it is kicking in?

[ruby drift]

Now that I got around to play a bit with SD (through ComfyUI), anyone has some recommended resources for me to read to better understand what I'm doing, and how to move forward?
I'm asking specifically because if I look at something like this, https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0, I have absolutely no idea if that would be something relevant for me to try, and how to go about it if I want to.

(When I try to find information, all I get are either tutorial that say "click here, click there" without substance, or too deep that I can't grasp anything.)

[nova lily]

So if you wanted to i2i only a portion of the image, you would have to go to inpaint instead right? or is there another technique out there?

[mortal sphinx]

Way out of my understanding but amazing that it’s possible. I hope tools become easier over time.

[desert dagger]

start your explortion by seeing what individual words do. just use the word apple as your prompt. generate. change the seed, do it again

[warm junco]

Yea inpaint is made for that

[plucky isle]

i cant use models in stable diffusion

[desert dagger]

you don't have a choice, you can't use stable diffusion if you don't use a model. more explination of what you're having a problem with, please, and how you're running stable

[plucky isle]

I upload a model but it does not create an image

creates an image with its own model before loading it

but after uploading the model it does not create an image

[desert dagger]

let's back up. how are you running stable diffusion? are you running it on a website or your own machine?

[plucky isle]

automatic 1111

i have rtx 4060 ti 8gb

is it enough

[nova lily]

Is it possible to put upscalers on XYZ plot? whats the option called?

[desert dagger]

it should be, though it might run slow. I dont know Auto1111 though, i use comfy. @warm junco is probably who's going to have to troubleshoot this

[warm junco]

Yea its enough.
But you need to edit the webui-user.bat and at the line commandline_args= you add:
--xformers --medvram-sdxl
For the best performance.

[warm junco]

Yea its possible. Its called upscaler I guess

[plucky isle]

how will I do

[warm junco]

Right click the webui-user.bat and edit it

[bitter seal]

Hey everyone, where is the best place to post question sabout animatediff?

[warm junco]

Hey, In #🤝｜tech-support

[low moon]

https://huggingface.co/spaces/multimodalart/AuraFlow

check it out XD

[foggy halo]

auraflow is good actualy

except it still kinda struggles with woman lying on grass

[low moon]

the things AI struggles with: hands, text, women laying on grass...

[foggy halo]

yeah even the text isnt nailed perfectly

just hope a finetuned model

[low moon]

yes it's v0.1 after all

its nice to see a new model come out tho like almost weekly

hold on to your butts and papers, what a time to be alive

[wet grotto]

T2I contolnet work both 1.5 and XL?

[civic berry]

does anyone know how to fix some elements in my generated image?

[desert dagger]

that's what inpainting is used for - or photoshop generative fill

[nova lily]

It doesnt exists unless its called something else

[steady atlas]

yes for sure

[low moon]

we need a pony room

for instance what are those scores we have to add to the promts?

in fact we need 2 pony rooms one sfw and one nsfw XD

[fervent thunder]

you don't really need a room

because you just need to do one thing

add score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up and score_8 to score_8, score_7_up, score_6_up, score_5_up, score_4_up

that's the only trick

people really exaggerate how hard that is for some reason

[sage niche]

What can this do?

#What can this do?

[low moon]

is that all negatives or positives?

but not everyone adds them in promt i see

some add only a few,

whats the criteria

[fervent thunder]

all positive

would recommend almost never using a negative

on any model

and if you do, just one word

you have to add them all, adding only a few is not correct

since score_9 is higher quality than score_8

you can just do the score_9 string which is score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up

but you can't just do score_9 on its own

it explains mostly here https://civitai.com/articles/4248

But wait, turned out I messed up a bit! What I described above is how PD V5.X used to do things, in V6 I wanted to also be able to say - "hey, give me anything 80% good and up". But score_8 tag would only give us images in range 80% to 90%. Perhaps using both score_8 and score_9 would work but I wanted to verify that, so I changed the labels form simple score_9 to something more verbose like score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up and score_8 to score_8, score_7_up, score_6_up, score_5_up, score_4_up. In reality I exposed myself to a variation of The Clever Hans effect where the model learned that the whole long string correlates to the "good looking" images, instead of separate parts of it. Unfortunately by the time I realized it, we were way past the mid point of training, so I just rolled with it (I did try to use shorter tags after the discovery but due to the way we train it didn't have as strong of an effect).

[still glacier]

no thanks ? don't drop random url without any context

[fervent thunder]

please report it seems virusy

[still glacier]

you know what, I'll delete it for now to be safe. If you can provide more context we'll see. This whole website seems dodgy.

[gusty mist]

Sup, why doesnt ksampler efficient have a denoise setting?

[desert dagger]

What?

[gusty mist]

the normal K-Sampler has the “Denosie” option, whereas I don't see it on the ksampler efficient sdxl node.

[desert dagger]

I guess that's what efficient means, just a few settings, and they expect you to use a different sampler if you want the additional settings

@warm junco can you get rid of this spammer, please

[gusty mist]

alrighty, thx

[fervent thunder]

there's so many ksamplers that are slightly different

[cobalt ether]

Which version of SD is used when generating images with the ultra feature?

[fervent thunder]

ultra is secret pipeline

I worked out that it at least has one second sample stage

with either latent upscale or noise injection

[desert dagger]

you can probably find the answer to that by just searching this discord for the word ultra and reading through the posts

[fervent thunder]

the only other clue I have is that the one called "Core" was confirmed to be a fine tune

[cobalt ether]

Doesn't look like a fun rabbit hole

[fervent thunder]

its kinda gotta be secret

because that's one way they monetise

[desert dagger]

it's not secret, and it's got nothing to do with monetization. it's still being developed.

[fervent thunder]

oh maybe I misunderstod

I thought the idea was they would charge for ultra

[desert dagger]

it's in beta, that's why you can use it here or on the website - or via the api

[desert dagger]

they're charging for the artisan access, not the specific models you can access via that

[fervent thunder]

at least at the moment they charge for ultra on the API also

[desert dagger]

they're charging for API access - you're using their GPUs, so you're paying for compute time

[arctic sedge]

Hey @karmic brook ! Hope you having a good week. Do know if SAI are in talks with the CivitAI team, to host SD3 again at all?

[tawny halo]

dm me if u can help me in nlp python and transformer to create a chatbot nlp dm me if u can help thx its for AI tools

[hushed grotto]

=)))

[tawny halo]

wdym

[hushed grotto]

@karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook

[cobalt ether]

https://tenor.com/view/dies-from-cringe-meme-cringe-imagine-gif-23477312

[hushed grotto]

no barking

[cobalt ether]

Sure you can say it, but why are you

that doesnt seem like a good insult

[hushed grotto]

shutcho ahh

imma give you teeth

[cobalt ether]

are you going to threaten me with head

[hushed grotto]

with teeth

[cobalt ether]

you are a rambunctious one

[hushed grotto]

imma give you good head

[cobalt ether]

not with that foul mouth

I will have to exercise you first

[hushed grotto]

shutcho ahh

[cobalt ether]

LORD HA MERCY THE DEVIL HAS THIS ONE

[hushed grotto]

@karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook @karmic brook

[cobalt ether]

Im sorry to inform you but I believe fruit views you as inconsequential

[hushed grotto]

im mad

[cobalt ether]

why

[hushed grotto]

when im angry people run away

RUN

[cobalt ether]

Do you want to talk about our feelings

[hushed grotto]

imma give you extreme teeth

[cobalt ether]

Not only do I provide civilian exorcisms, I can also provide emotional counseling

[hushed grotto]

ok

thats it

[cobalt ether]

im not the tooth fairy

[hushed grotto]

im angry

RUN

[cobalt ether]

stop offering me your teeth

im neither the tooth fairy or your local drug dealer

you dont have to blow me

[hushed grotto]

@karmic brook @karmic brook @karmic brook @karmic brook

[cobalt ether]

perhaps he has children of his own to deal with and cant deal with the ones here also

sometimes you need to apply perspective

[pale latch]

I've always felt that he was just winging it. Caption dropouts on this is the obvious solution

[desert dagger]

let's see if i can get discord to ban your account from all servers

[pale latch]

He'll have another account rolling by the end of day

[desert dagger]

not if they ban his ip

[pale latch]

or just delete his own posts. discord won't action then.

[desert dagger]

he can hope he deleted the ones i reported, and let's see if he guesses right

[pale latch]

aren't you supposedly a technically inclined man? you know that's a nothing effort right? not only could i change my home IP right now, i could also just use a vpn. IP bans tend to fk with a ton of users too so networks typically avoid them now. consider if he was at his highschool using discord. Every other discord user at his school would get banned too since discord see's their entire subnet as one address

[desert dagger]

sure. but not everyone knows how to spoof an ip, do they? and this guy is probably, 12? 11?

too young to be on discord, and i'm looking for proof of that to report too

[pale latch]

HWID bans don't even effectively work. Just load it in a VM or run a bootloader that changes it.

the best way to manage bans is accounts. That's about it.

[desert dagger]

and do you honestly think discord cares if they ban a school full of kids? nope.

[desert dagger]

i'm very well aware of that - providing you know how to spoof the hardware address - these aren't skill sets most kids have.

[cobalt ether]

35

[pale latch]

This is just a cop delusion. chasing a speeder to let him know he's speeding.

vpn's are basic to figure out. you'd be surprised what 12 year old trolls can figure out

[desert dagger]

not you.

[cobalt ether]

I know

[desert dagger]

may i point out that my comments seem to have gotten the correct response out of the individual? though it's too bad i had to say this where he can hear it just to get it through your brain?

[cobalt ether]

his functional age, but his mentality might be more in line with your guess

[pale latch]

I'm all for him leaving too. But ip bans are still meaningless and reporting is often not as effective as you think. Best you can hope for is that server moderators do their duty

[desert dagger]

let me try this again. thanks for totally blowing the cover on what I was doing which got him to stop. now, will you please shut up

[hushed grotto]

Im sorry guuys

[cobalt ether]

I forgive you

[marble pecan]

hello

[hushed grotto]

ion wanna get banned

[pale latch]

lol making things up isn't how to teach people to be better people. just breeds resent. Also, i don't believe you . IP bans don't matter at all

[cobalt ether]

I will ban your whole computer!

[hushed grotto]

??

why

sorry

[cobalt ether]

Because I can

[hushed grotto]

noo

[cobalt ether]

and all your families computers

[hushed grotto]

really>

[cobalt ether]

and if you dog has a computer, banned

[hushed grotto]

thats possible

[cobalt ether]

yes

[desert dagger]

oh i did report him. every single message. i could have done so silently, however there was a reason i said what I was doing publically and it wasnt' so you could pretend to be a hardware tech

go back to making sphere loras

[pale latch]

lol this troll is sneaky. Now he's pretending to be 12 since the man is so sure. been around long enough to have seen this play a million times before.

[cobalt ether]

kidding though, I wont ban your dogs computer

[pale latch]

aw tears. Dude you were leaning on ip bans like they meant anything. Just saying.

[desert dagger]

i'm really fed up with you and your condescending tone.

[pale latch]

i don't make sphere loras either. spheres are a specific perfect geometric unit. i got balls

[cobalt ether]

same

[pale latch]

i tend to reciprocate like that and certain personalities are often very grated by it

[desert dagger]

yeah, well you can start by not assuming that you are the be all, end all, computer tech and that no one by you knows anything. that would help

[pale latch]

IP bans not working are like, first day of tech school lesson # 1

[cobalt ether]

whats lesson #2

[desert dagger]

kid, i was building computers before you were born. drop it

[pale latch]

the user always lies

[cobalt ether]

theres a different between building computers and the internet though

just to be fair

[desert dagger]

yeah, i know. i'm a network technician as well. and i'm done with this conversation

[cobalt ether]

Alright, been a pleasure

[desert dagger]

talk to me after you wire a building

[pale latch]

hmm. anything before i was born is long obsolete now a days. ip spoofing was even easier back then since networks had a lot less ssl and authentication headers

[cobalt ether]

I wouldnt survive long enough to talk to you

how could you suggest I do this untrained

are you trying to murder me

[desert dagger]

shows what you know

[desert dagger]

the problem you have, is that you keep assuming comments are directed at you, instead of the person they really are directed to

[cobalt ether]

yes it is a problem

was that commented directed at me?

[pale latch]

He just wants to live while he's alive you know

[cobalt ether]

oh god did I do it again

[desert dagger]

that one was, yes. but just about everything else you've reacted to wasn't

[cobalt ether]

was that one directed at me?

[desert dagger]

who's on first?

[pale latch]

this one passively was. just to be clear

[pale latch]

was that directed at him or me? Can't be me because you are fed up with me and fun antics don't make sense to try to engage with me in at this point. I'll assume it's the steel horse iron riding cowboy

[cobalt ether]

is this directed at me?

[pale latch]

the steel horse part was, but if you notice i had used discord's reply function to show which message it was mainly directed towards. features we didn't have back in the irc days

how you like my 2 spaces after a period styles?

[warm junco]

its called Hires Upscaler

[frail sonnet]

Well damn

"We hope this message finds you well.

Adobe Stock has refined its content policy regarding Artist Names. If you have referenced an artist’s name in your content’s titles, keywords, or text prompts for Gen AI content, the content will be removed from Adobe Stock.

For additional information about the updated policies, please see our Account and Submission Guidelines. All content must comply with the Adobe Stock content policies.

Best regards,
Adobe Stock Artist Relations"

I don't think Rembrandt currently holds copyright but whatever

[desert dagger]

just send adobe stuff that doesn't refrence artists then

[narrow kernel]

That's not it, it's the images found online of the art that companies own, like museum websites that had their data scraped by bots. The museums themselves might also have some kind of agreement for tourists along the lines of "we own the rights of any pictures taken inside the building and reserve the right to enforce that at our discretion" or some shit along those lines

To cover instances of just using someone's Facebook photos of the same art

[elfin shuttle]

Hi

[desert dagger]

hey

[sage reef]

[fervent thunder]

what happened

[low moon]

Was Fork today's popcorn moment?

[fervent thunder]

whats the trick to getting them to approve any art

does it have to be a certain size

[low moon]

Adobe went from awesome photo editing and stuff to total corporate tyranny.

[desert dagger]

went?

[low moon]

well yeah

[frail sonnet]

It has to be a certian size, format and pixel amount. I forget all that now since it has been forever since I sent in any art. Also they don't want the images too large, so I had to rezie everything just for them. Kind of a PIA honestly.

[fervent thunder]

too much competition

seems like it might be better just to start your own art blog

[tough tree]

whats a Good Ui to use for someone new to SD?

[desert dagger]

comfyUI

[fervent thunder]

Hi everybody is there any game recently

I got sad and joined a support group then read over other peoples issues and suddenly felt more grateful

[fervent thunder]

Excuse me I should send it to other group

[brave matrix]

I assume there api services that you can use to render sd images in the cloud?

Can someone tell me the best ones?

The cheapest

[fervent thunder]

vast.ai

is almost always cheapest

at least here in europe

especially if you use it at funny times

[brave matrix]

I meant like less so gpu rental and more so like a cloud SD renderer

That you can make calls to

Does that make sense?

[fervent thunder]

https://comfy.icu/

this is ran by some guy on reddit

has serverless API

[brave matrix]

Nice

Thanks

Do you know of any others?

[fervent thunder]

no that's it

cpus work, just wait longer lol

[desert dagger]

search google for online comfyUI

[dusk scaffold]

Some news? It's been a long time since I've faced some news about AI.

[fervent thunder]

no news rly

world is waiting for GPT5

[dusk scaffold]

Reddit banned my IP and hardware. I have no idea where to find news about AI. Seems that it's not already turned into something more private and closed

[fervent thunder]

what did you do on reddit?

[desert dagger]

all the news is on twitter, anyway

[dusk scaffold]

Twitter? It's censored as fuck. 99% of people that I follow are nsfw artists and 1% of what they show me is nsfw art os something about it

It's geo and user censored

[dusk scaffold]

Talk and share thoughs without censorship. Some people hate it

[desert dagger]

twitter is where the scientific community is, and that's where the news comes from

[dusk scaffold]

It changes from user to user and region to region. It's censored as fuck

[desert dagger]

you asked for news. twitter is where the scientific community is, and that's where the news comes from

that's where you'll find papers posted, new developments posted, etc

[dusk scaffold]

Twitter? It's censored as fuck. 99% of people that I follow are nsfw artists and 1% of what they show me is nsfw art os something about it

[desert dagger]

so you're not really after news, then, apparently

[silk condor]

I want to do an upgrade on my old GPU in my secondary setup to hold stable diffusion and local ai etc.. Should I go for an Nvidia GPU with 12 GB or should I look into taking a AMD with 16 GB instead?

[fervent thunder]

always Nvidia for AI

unless you are willing to troubleshoot a bunch of low level library stuff

to do with pytorch, flash attention, ROCM etc

[silk condor]

So even if AMD has more ram it's better to go for an Nvidia card? 🙂

[fervent thunder]

yeah

[desert dagger]

amd is a cpu, not a gpu

[silk condor]

Sorry Radeon 😄

[desert dagger]

yeah, the world runs on nvidia now

[silk condor]

Okay so Nvidia.. Would it be better with an 3060 with 12 GB ram or a 4060 with 8GB? They both have GDDR6 but 4060 runs on 2500~ mhz and the 3060 runs on 1800~ mhz

If we just go with default and no overclocking etc.

[fervent thunder]

the higher VRAM amount

[dusk scaffold]

In twitter no

[fervent thunder]

another news source is Arxiv and then the big conferences

[primal dirge]

Guys, we got a huge problem

Hacktavists are Targeting comfy

And stable...

[desert dagger]

pretty sure the admins of both are aware of that

and there's nothing we can do about it

[primal dirge]

Oh good

Still we're in their cross hairs

And furries of all people too!

[desert dagger]

you've had way too much coffee today

[primal dirge]

I hate that you're correct about assumption

But the group is null bulge...I wish I was kidding

[austere carbon]

Hi I’m wondering if anyone can give me a rundown of this guys workflow for his videos… I posted on Reddit and wasn’t able to get an answer. I understand the concept of feedback loops, but the complexity and fidelity of his videos looks insane, any information on this workflow would be greatly appreciated!

https://www.instagram.com/reel/C8luO4VM3l1/?igsh=MzRlODBiNWFlZA==

[desert dagger]

i'm pretty sure that just about everyone on the internet is aware of them, they made a very bad mistake the other day - and if you happen to be one of their members, it would be a good time to cut ties

[desert dagger]

it's probably video to video

[austere carbon]

@desert dagger do you think he’s starting with traditional vfx and other elements in a video he creates himself as the seed “first” video or is it all ai based? … I work in vfx and i feel this has more going on than just ai (ie:composting elements, fx, ect) maybe I just don’t know the capabilities of the tools yet, but parts of ut read like human made vfx to me. Thank you for your input

[desert dagger]

i'm betting he took videos of his living room and then used that for video to video and edited the clips together

[austere carbon]

@desert dagger I’ll look into using my own videos. And see what I can achieve, thanks!

[desert dagger]

take a look at runway's Gen2 video to video feature

[austere carbon]

That would be perfect as I already have subscription to runway

[desert dagger]

🙂

[austere carbon]

Thank you for your time, much nicer than the response I received on Reddit lol

[desert dagger]

you're welcome. 🙂

[primal dirge]

That's the good word I needed to hear, broski. Thanks

[primal dirge]

Is this all made locally? What was used and its workflow?

[austere carbon]

My only information directly from the creator is that he used “1.5” I’m assuming he’s referring to stable 1.5. I have no other information as to his workflow and can only speculate, if you have any ideas any information would be appreciated, thank you

[primal dirge]

SD 1.5 is capable of video generation? It looks too fine tuned for that

[austere carbon]

I believe he’s using feedback loops

[trim magnet]

yes the only good animatediff models are from 1.5

[primal dirge]

Huh, I see, thanks for the info

[sturdy tangle]

so my entire reason for working with ai is adult graphics and video, i wanna make onlyfans for ai gen characters, am i in the wrong place??

[primal dirge]

Rather random to shoot your motive, but welcome aboard

Going local installation?

[austere carbon]

Do you think his result is achievable using a feedback loop type workflow where he feeds the output image as reference over and over again and changes the prompt slightly? It looks too smooth for that but maybe I’m just not doing it right

[trim magnet]

yes but he probably used another program to smooth the video

[austere carbon]

Nice, I’ll keep trying, would love to be able to pick the creators brain, cause whatever he’s doing looks crazy good.

[trim magnet]

yea u prob need to do several gens until it comes out right and the smooth it out with another program

[desert dagger]

yeah, you probably are

[nova lily]

ah appreciated

[sturdy tangle]

new to discord, but i have no time for foolin around, so if you were talking to me 13012ed, i think i must make a local install, not sure what yet...i done research and a real hot market exists that no one talks about legal but not proper or moral..so i need to make some money. my partner in crimee and other things is helping..dm me ill show you some prelim tests turned out well using mage, but the whole process of stills to stop motion...too tedious...i think content creation is about to ignite but it costs sooo much $10,000 macs 100,000$ server setup is beyond me... Ive got overclocked liquidcooled i5s running like i7s clusters, ubuntu, mac windows android iot ect in a constant state of change trying to devlop a reasonably cost efficient way to run a virtual stable of frisky playmates for the socially challenged. OF is kind of a virtual strip club, good place to waste money. I want some more along the lines of afterhours at Hugh Hefner's. you pay to be a member, afterthat no rules, if you can imagine it you can see and interact with it. Im sure some will have a problem with it, but they can ..whatever.. i intend to retire before 50, buy an island, a satelite, do all the things others only dream. I had this idea when i read Cryptonomicon...but not a bank...lol

[tough tree]

what are some good guides on using comyui for a noobie?

[desert dagger]

i'd suggest you start by watching Scott's tutorials https://www.youtube.com/@sedetweiler

[desert dagger]

you're new to discord but don't have the time to learn how to use discord? you're going to find yourself very confused.

[ruby drift]

Anyone has experience running SD from within a VM with GPU passthrough?
If so, how is that working for you?

[forest dome]

Ancient Beauty

[nova lily]

I just found out the issue, that option doesn't exist in i2i. idk why

[sage reef]

maybe Scott is waiting for the improved model or 8B to come out before making new videos 🙂

[fervent thunder]

does anyone else run tkinker with their stable diffusion or does everyone pipe it through that one github gui with the sliders

[desert dagger]

might be, he's rather busy - but for someone just getting started, he's got plenty to watch already

[sage reef]

of course, i learned a lot from Scott as well

[sage reef]

github gui with sliders?

[fervent thunder]

like ollama but better

[sage reef]

yea but can you link to what you are talking about, im actually not sure

[fervent thunder]

https://github.com/AUTOMATIC1111/stable-diffusion-webui

[sage reef]

if your question is just "how do people usually use stable diffusion", i would say most of them either use automatic1111 or comfyui

[fervent thunder]

yeah with the sliders... I made one with tkinker for setup purposes.. checks the CUDA drivers and installs the checkpoint values.. pretty niffty if you ask me. Computers have come a long way!

[sage reef]

most beginners use automatic1111 and then once experienced enough with the terminology, pipeline, etc, they switch to comfyui

[fervent thunder]

alright cool. I've used both I've just had more luck for my own purposes using python scripts. The gifs, photos, and photo to photo generation works really well!

[sage reef]

oh you mean with just python scripts? yea that works too

[fervent thunder]

Yeah super simple made a filing system with it to save the photos with either preset names or a prompt will ask for the filename

[mossy mantle]

can someone help me im stuck and i dont know what to do

i tried to infuse my cheeseburger with dark matter and it doesnt work it just exploded

[desert dagger]

i'll bet you poured it on rather than spreading it, didn't you?

[mossy mantle]

wrong

i stabbed a syringe full of dark matter inside of it

[sage reef]

is this some AI generated text? 🤣

[desert dagger]

did you get pickel bits in the tip of the syringe?

[south basin]

did turbo/d3-turbo ever release their training code?

[warm junco]

Because in img2img there is no hires fix setting

[lethal spindle]

hi, i wanna play with text-to-video but i am not familiar with these models or how can i train them (lora, dreambooth or something else?). can someone help me with some information about text-to-video?

[tropic frost]

out of curiosity, what method do you use to get more accurate hands/fingers?

[stuck sun]

adetailer has a hand model which works... sometimes

[sleek oriole]

Is there a room/server where I can ask Fooocus questions?

[hardy sedge]

Why do sdxl character loras look much more realistic on an sdxl model than on a pony model? And is there a sdxl model that is like a pony model?

[warm junco]

Pony is based on SDXL.
If a lora is made for sdxl it can work with pony but dont have to.

[warm junco]

If its a technical question then here:
#🤝｜tech-support

[hardy sedge]

It works, but the loras look better on sdxl models. With Pony Models, you can hardly recognize the character.😔

[warm junco]

Is it a real or anime character?

[hardy sedge]

Real

[warm junco]

Then try PonyRealism model

[hardy sedge]

I try many Models all the Same. I think there is No solution

[warm junco]

Have you tried changing the lora weight?

[hardy sedge]

https://www.reddit.com/r/StableDiffusion/comments/1bts2km/sdxl_loras_with_pony_model_seem_to_not_work/

[hardy sedge]

Yes, but then the face is completely broken at 1.5 eg.

Too bad, pony models have a damn good quality.

[fervent thunder]

if you look on reddit
there is someone who has pulled off realistic pony fairly well

their workflow is not simple though

it was a few weeks/months back

ive been putting cat videos on instagram, seems to be the way

nice

these models make great cats

cats work really well with noise injection because it thinks the noise is fur

so it makes them more fluffy

[scenic knot]

Hey hi can anyone tell me how to use stable diffusion server

[lusty beacon]

Guys what is the best sd model

sd3, lightning or xl or sd 1.5

etc

[pine summit]

hello everyone

[wet grotto]

depends what you need, the space/vram that you have, and the content that checkpoint was trained

[lusty beacon]

I have 24 gb vram

wondering which base model is the best one without fine tuning

[wet grotto]

well if you don't need any context you can take SDXL, is the bigger

[fervent thunder]

it depends if you are looking for the latest thing to experiment with, or if you need something reliable for a project right now

SD3 has the best image quality due to the new VAE

but if you need something reliable right now its not viable neccesarily

[rigid hedge]

is BLIP2 better than BLIP? 😄

[lusty beacon]

But it's only working in Comfy

?

[exotic venture]

How do I generate ai image what command is it?

Worthless ass server

[fervent thunder]

it works in Diffusers as well

[desert dagger]

there are only a few channels on this server that you can generate in, please start by reading the informtion here #artisan-faq

[proven spear]

may i ask how to set up stable diffusion in my computer?

[still glacier]

cf #🤝｜tech-support pinned messages

[proven spear]

thx

[indigo carbon]

how do i start making ai ? like all of these projects what should i even learn

[copper shale]

Has anyone tips how to make images look more artistic/less realistic. I hate that almost any image, whatever I prompt, looks super HD with Strong contrast and shadows, looks so bad. Idk where this hyperrealistic trend comes from in AI art but it sucks. Can't make images that look like actual digital paintings anymore

[lilac forge]

Hello everyone, hope everyone's doing well today! I am having a problem with wildcards in Automatic 1111.

I make the prompt like this:

score9,score8_up,Pony YeiYeiArt Disney Princesses, lora:Artgerm_XL_PONY:0.8,

Below is a snippit of 3 of the lines in the wildcard I'm trying to use:

Pony YeiYeiArt Disney Princesses
<lora:AnastasiaXLP_character:0.8>,AnastasiaXLP,short hair,red hair,blue eyes,coat,scarf,ponytail,fingerless gloves,hat,yellow dress,ponytail,black dress,bare shoulders,necklace

<lora:ZeldaMarinComission_Character:0.8>,MarinXLP,hair ornament,pendant,hibiscus,blue dress,red sash,orange hair

<lora:MaleficentXLP_character:0.8>,MaleficentXLP,colored sclera,green eyes,colored skin,green skin,horns,cleavage,cape

<lora:SarahHawkins_Character:0.8>,SarahWaifu,brown hair,long hair,blue eyes,white shirt,red apron,mob cap,pink nightgown

But it just shows up like this (in other words, there should have been one of the entries where the wildcard's name is)

score9,score8_up,Pony YeiYeiArt Disney Princesses, lora:Artgerm_XL_PONY:0.8,

And I do have the Wildcard and Dynamic Prompt extensions installed.
Also, when I get rid of everything in the prompt but the name of the wildcard, like this:

Pony YeiYeiArt Disney Princesses,

It does actualy work, and this shows up (it's a random pull from the list in the wildcard text file):

lora:SnowWhiteXLP_character:0.8,SnowWhiteXLP,puffy sleeves,dress,yellow skirt,corset

Also, the formatting on Discord gets rid of the double _s and turns them into underline. But I am using 2 _ in the front and _ after the name of the wildcard.

[pastel surge]

4070 or 3080 TI for A1111 image generation?

[hard kernel]

I suggest using artistic lora for style and try changing models

[keen kraken]

hey guys ive been using easy diffusion for a while but decided to try Forge because ED isnt supported anymore and forge is faster, i just wanted to ask if there is a way to view prior generated images and if you can see prompt history

[hard kernel]

Seeing old prompt and generation is available on fooocus GUI and available on colab https://github.com/ehristoforu/DeFooocus

[keen kraken]

is it as fast as forge, on easy diffusion it would take maybe around 4 minutes per image but forge is alot quicker

[hard kernel]

It's fast as forge but you can't add extensions into it and your restricted to there aspect ratio, cn, face swap, ip adipter, available in both of them

If you don't have high v ram GPU I suggest use use this in colab it would be 10x faster

[wise ginkgo]

anyone answer u?

[wise ginkgo]

so rocm doesn't work well or some other reason?

amd gpus often have more vram - but, that's not enough reason to get amd gpus?

I'm looking at similar cards although I think I'd try for a 3090 over a 3080 ti since u can only find them used where I am - and there's not much diff. in price

a 4070 Ti super is a bit more

7900 xtx used and 4070 Ti used are similar prices

[fervent thunder]

ROCM can sometimes work for some things in a limited capacity
if you are willing to do a lot of work troubleshooting and fixing issues with underlying libraries like pytorch and flash attention
and you are willing to accept slower performance as well as less features
and bare in mind there might not be any help available because almost everyone uses Nvidia

[wise ginkgo]

this is the universal idea, then? 🙂

do u guys use Stable Diffusion in Linux, too? Because nvidia in Linux can be problematic too- right?

[fervent thunder]

yes nvidia in linux can sometimes be problematic

[wise ginkgo]

I haven't had an amd gpu in years - I'm familiar with using nvidia cards - but, I have used them in Linux - apparently, things are improving

[fervent thunder]

yeah for the most part nvidia in linux is okay

[wise ginkgo]

do you use nvidia gpu in Linux?

I plan on using Arch and Fedora

I'm using Windows atm but I am gonna install Linux on a spare ssd

[fervent thunder]

it should be fine

[wise ginkgo]

I'm hoping to buy a gpu to replace what i have in the next few months

here a 3080 is around $550 and change .... a 3090 is $800 and a 4070 Ti Super is about $900-$1k

oh yeah, what about vram?

is there any problems if you only have 12gb? 3080s often only have 10gb

3090 - 24gb,4070 Ti Super has 16gb so I kinda would like to get that gpu

a 7900 xtx has 24 and even the 7900 xt has 20

[fervent thunder]

more vram is really important

but I would much rather have a 10GB Nvidia than a 24GB AMD

[wise ginkgo]

really? interesting

[pale latch]

amd isn't that bad these days. it has sdp to supplement the xformers advantage now. that was a big deal before. there's still speed differences, but more vram is more capable

and if you're going to go linux amd has rocm there

[fervent thunder]

in some ways yes

but if a new architecture comes out rapidly

then AMD is going to fall behind again

[radiant tusk]

Is there a adetailer (bing-su) for comfyui?

[fervent thunder]

yeah its called impact pack

it has a node called detailer (SEGS)

[radiant tusk]

Does it make the faces more similar to the source model the same way adetailer did for A1111?

[fervent thunder]

which can inpaint on the basis of object detection and segmentation models

[radiant tusk]

Thanks

[fervent thunder]

yes exactly the same

[radiant tusk]

Im gonna try tot set tha tup

is dreambooth still the best way to do quick models of a face or is there a new way now

[fervent thunder]

I haven't trained stable diffusion yet so I will let someone else answer that

[desert dagger]

@warm junco

[frail sonnet]

https://www.nasa.gov/missions/mars-2020-perseverance/perseverance-rover/heres-how-ai-is-changing-nasas-mars-rover-science/

[spark tinsel]

What exactly happens when you put a word with negative weight into the uncondition prompt? (Example:-1)

[pastel surge]

from what i can see the difference is extremely minimal after testing today

if anything the 3080ti pulls ahead by 10-15 seconds from what i can see, but it may be configuration related, i need to make sure theyre identical installs later

[formal badge]

Hi

[nova lily]

In A1111 Have they ever made an extension or option to make the preview image/window bigger yet?

[fervent thunder]

when the pizza party so good i call the pizza skibidelicious

[hard kernel]

I am not from USA but what's the meaning of skibidi?? Why people using it I am only aware of that skibidi toilet yt channel??

[fervent thunder]

its a joek

a meme

[trim magnet]

when something is good u say its very skibidi

[hard kernel]

But skibidi is a toilet how it can be good??

[trim magnet]

it just is

its in a perpetual state of being good like Livy Dunne

[hard kernel]

Very skibidi explanation

[trim magnet]

yea once u become a rizzler sigma from ohio you will understand everything

[narrow kernel]

why ohio?

[placid hatch]

Does running webui or comfyui in fp8 mode significantly reduce quality?

[warm junco]

Nope

[night shoal]

有人吗

[strange coral]

Hi

[placid hatch]

Do you happen to know which fp8 format is preferable?

[radiant tusk]

Who knows how to use live portrait vid to vid well

[mighty creek]

Hey folks I am looking for a solution to make gamified, creative avatars , from user selfie, I tried several solutions with prompting but theres no consistency in the solution , does anyone know any solution for this , or any resource that I can try out ?

[quasi belfry]

hello. I made a Python code to generate images with stability.ai platform API . It works good, eveni if I quickly ran out of image generation mode. It seems that I could only generate 3 images during the trial period, how long I tested stability.ai

The problem is that the images are DALL-E 2 Level, it means that they are not at all conclusive, poorly outlined. I can't use them in my articles. Can you tell me what I have to do to generate better image quality, richer in content?

[warm junco]

If you have a good GPU you can use Stable diffusion localy on your PC to generate images for free.
And you can generate better images because you can use custom models.

[warm junco]

Hey for that look at Controlnet IP-Adapter

[placid hatch]

Is e4m3fn or e5m2 better when running in fp8 mode?