💬｜general-chat

[fervent thunder]

I would be honored to be called a 'reg' in any place lmao

You might just put a single batch of 6 to test out a prompt or new lora

[astral goblet]

I've got 16gb of vram and a 4080, so higher batches of 6 win over here. it's a matter of 2 seconds or 10 seconds and i get a few variations to look at

[fervent thunder]

where do i find the bots

[astral goblet]

robotopia

[fervent thunder]

is it at bots.com?

[astral goblet]

.com

[fervent thunder]

why dont you go back home to your cool home at coolmathgames.com

you dont need to confuse yourself with fungible tokens

where da bots at

where are they

tieyhe

[rigid ember]

Gotcha. I will toy with that to wrap my head around it a little more. Regarding Loras, can I turn them off? I'm finding that multiple loras that are different but share the same trigger give me things I don't want, but I can't seem to be able to enable or disable them during a prompt and just end up physically removing them from the directory

[fervent thunder]

That's a great idea.

I support you.

[rigid ember]

Thanks chatgpt bot

[fervent thunder]

You are welcome, god.

[rigid ember]

The correct term is "creator"

[fervent thunder]

No, i'm ChatGPT. I am right.

[lament steeple]

with the way loras are set up, you have to turn them on loraname:1.0 to activate them where 1.0 is the actual weight you want to use

That goes in your prompt

[rigid ember]

So they aren't active without that specific prompt explicitly?

[lament steeple]

right

[rigid ember]

It must be something trained model specific or in my prompt order/weight then. Thanks. Good to know what it's not. lol

[lament steeple]

What are you getting that you don't want?

[rigid ember]

Just similar prompts to previous outputs in another model/face train ckpt. Which I guess is another question if you wanted to field - I learned how to train a model, but still haven't learned how to train a lora. I've got a way too long video I'm at the start of/returning to that'll tell me, but wondering if there's a faster way xD Anyway, the question is just when should you lora and when should you model? Are models just better at giving the desired result with the same learning input and parameters?

I did a "face" train in a model that includes body type and whatever else the train picked up, but I'm also wondering if I'd get better face results with a lora model and low denoise

[lament steeple]

Lora/TI when you want a single word. Model/hypernetwork when you want multiple concepts that take explicit description

[fervent thunder]

can someone show meh huhh

big mac

and fries

wait

[dark anchor]

no

[fervent thunder]

no
i was gonna say

wait

i forgot

sorry

thisis really hard

i cant remember

[dark anchor]

awesome, don't break your enter button

[fervent thunder]

wait

what

is that

oh no

its getting worse

im so sorry

i cant even

[dark anchor]

eh

painfully unfunny

[fervent thunder]

i was just trying to get a big mac to enjoy

[rigid ember]

He's becoming self aware guys. better hide. lol

he gon splode

[fervent thunder]

i dont even remember where to get. what was it

what was i trying to get

i dont even know

its

oh god

[dark anchor]

discord acc created today and first server joined was this

bleh

[fervent thunder]

im sorry i dont think internet fatasses can help me on this

goodbye internet fastasses

scihofrendia

[rigid ember]

So what's a good example of one of those multiple concepts with explicit descriptions? An artists art style? A person's body type/proportions? Something else I might be missing? lol.

[lament steeple]

an artist style is a single concept

multiple concepts would be a.... keeping it sfw... say you wanted to do a specifically sci-fi themed model.

[valid pendant]

I'll help 10 persons on how to trade and earn $20,000 within 72 hours of spot trade. Pm me for more information or message via TG t.me/smart_jones

[lament steeple]

I feel a monty python joke coming on

spamity spaaaaaaam! spamity spaaaaaaam!

[lilac pine]

Hey guys! I have a question. Can a LoRA be trained on other concepts? Not styles or characters, but like: A object, an action performed by a character, an interaction between characters, etc.,

[vast ingot]

Most training methods are also geared for that, yes. It's always the same matter of tagging correctly a corresponding dataset, but any concept that can be represented by enough different pictures should be able to be taught

[proud nova]

I was attemptng to train LoRA of this kind just yesterday. However, the result wasn't very good. I used up to 15 images. Maybe I need to use more. But I don't know. What I can say is that it had some effect on the result

[tight trench]

Big hello!

[lilac pine]

If I remember correctly, don’t you have to like remove some parts of captions that describe said concepts? I think other than that, you describe every other parts of the image that is not the concept you are training for.

[fallow axle]

hi

i dont remember, how do we update SD ?

[proud nova]

Hmm, maybe you are right. I will try it today again

[jovial bone]

Hi
I am totally newbie in IA. I can see different "model" I can download and use with stabe diffusion. But what is excatly a model, a way to render an image ?

[proud nova]

Dms

[fallow axle]

no one knows ? 😢 i think i can add a line in the file

like that SD auto update

but i dont remember what exactly

[lilac pine]

Yep! That is correct! The file you are mentioning is webui-user.bat

[willow coyote]

Hi, at https://replicate.com/cjwbw/hasdx why are some of the keywords written in parenthesis?

[lilac pine]

Open that in a text editor of some sort. There would be a line that would run the actual webui. Before that, put “git pull.” It should work.

[fallow axle]

thank you

[lilac pine]

Yeah I tried to generate a female character with white hair and red eyes, but the captions only described the red eyes. So when I generated it, the model was able to generate white hair consistently but it generated random eye colors.

Make sure to edit your captions so you remove any descriptions of your concept and leave everything else as is.

Maybe I should try that too

[proud nova]

So the txt file works like negative prompt...

[jovial bone]

haha apparently 16G of ram is a bit short to use nice model

[proud nova]

What model do you use and what resolution you attempt to generate?

[fervent thunder]

How much VRAM do you have

[jovial bone]

for the moment 512x512 and the model is v2-1_768-ema-pruned.safetensors

5.21 Go

I guess it s the Stable Diffusion 2.0 model , right ?

y a un v2-1_768-ema-pruned-fp16.safetensors half smaller , but I don't know the real difference

[proud nova]

So wait, you have issues with generating the images?

[jovial bone]

ho after to close some browser tabs, it s passed

[proud nova]

CUDA errors? Or?

[jovial bone]

nan , the softwhere just crashed when running the model

not enough ram

[proud nova]

You have 16GB VRAM?

[jovial bone]

still not enough ram haha

[proud nova]

What is ur GPU?

[jovial bone]

not a vram issue , more usual ram go to bottleneck

[proud nova]

Alright but what is your Graphics card?

[jovial bone]

GeForce RTX 2060

I am downloading the more lite version

[jovial bone]

which kind of machine do you have ?

[marble rapids]

U

[upbeat warren]

hello

[lilac pine]

Wdym

[stable pawn]

A question that may be simple. Is there any way to know what word AI understand and what words are unnecessary? I see folk spend words on things like "best quality" even humans have no concept what is best or what even "best quality" would referee too, and do words know "trending on artstation" and what style that even would be.

[near silo]

Anybody know how to change the quality output of SD images?

[stable pawn]

Like can I have some script that rate how good words I use in a prompt are.

[near silo]

generally speaking, terms like "best quality" and "trending on artstation" are actually really good for the AI's.

think about it this way. If an image was tagged as low quality, or has a low aesthetic score, it would be considered low quality, so by saying best quality, the AI gets rid of the lower resolution/aesthetic score images in its considerations

And then also if you say trending on artstation, it will be able to pull from images that it was fed from the artstation trending page, and draw connections. Specifically good lighting, good contrast, good colors, high resolution details, and the works

all of our words are just tags, which trigger related images with the same tags

also, i recommend looking into SWINv2 tagger. Its an extension for SD that can tag images incredibly well, and it uses tags that AI's understand well. So you could feed an image in, get the tags, and better understand what words it associates with

[stable pawn]

But what is "best quality", good lines, no outlines, sharp? But the question is, how do we know that just "trending on artstation" is better than "popular image quality"?

[near silo]

I actually found out recently that the AI's are less trained on "loin cloth" and more on "pelvic curtain" as weird as that is lol

[stable pawn]

I been annoyed by AI that I have to use "pussy" rather than vulva or vagina, it like the training is based on childish words rather than correct anatomy. But I have to adept, but make any pussycats can be risky 😄

[near silo]

yeah, its all understanding the words at use

I really recommend looking at auto taggers to get a good idea of what images trigger what tags

like here

have a look in #🏞｜general-with-images really fast please

[stable pawn]

Well, I did learn that AI see "girl" almost like "child", and that if you write "adult girl toys" the AI doe not render adult sex toys but rather something else, just for folk that tagged images and trained the AI did not know how to use the word "girl". Now "diamonds are a girls best friend" is seen as "Diamonds are a childs friend" 😄

[near silo]

that is actually the correct way

girl is adolescent

woman is adult

as boy is adolescent, and man is adult

the way they tagged it follows traditional English definitions, and actually allows for more control than if girl was the same as woman

female is a sex, girl is a young female, woman is an adult female, the works

[stable pawn]

you mean "female" and not "woman"?

[near silo]

ha, i said it even before you did lol

thats why we say old woman, and not old girl, cause it doesn't make sense and contradicts itself

[stable pawn]

So girlfriend shall be femalefriend?

[near silo]

and a young woman would be somebody around 16-20

[near silo]

now you're just getting into unnecessary assertions

[gusty falcon]

Does anybody here have experience of running SD on a Mac?? A non M1 Mac too?!

[stable pawn]

No. I point out how "girl" is used in reality and how AI connect "girl" to wrong things, Like a "city girl" is not the same as a "City child" as AI see it.

[near silo]

you're nitpicking small things that are easy to work around. Girl still logically means female child, or adolescent child. Terms like boyfriend and girlfriend and such are just terms that don't fit that standard relationship

[stable pawn]

But my question was, how do we know what word AI understand and what it connect the words to.

[near silo]

look in #🏞｜general-with-images I just posted two huge examples for you where I show how

[stable pawn]

No, you just bluntly took this out of context.

[near silo]

girl literally mean "A female child or adolescent"

[stable pawn]

https://www.merriam-webster.com/dictionary/girl#:~:text=a(1),whose gender identity is female

[near silo]

links to something that has all but one connotation listed as a minor

the first, aka the most widely used definition, still states a female child from birth to before adulthood

[stable pawn]

But the word girl can be used in different ways, the question is still "how do we know what word AI connect to diferent context".

[near silo]

look in the damn chat already #🏞｜general-with-images

I answered that a while ago

Literally showed how you can use taggers to look at images so the AI literally tells you what it sees, and how it associates words with it, what more do you want?

[stable pawn]

Girl was just an example. You was just taking it out of context to troll. The question is how can I know what word AI understand.

[gusty falcon]

Hello everyone, I’m looking for some support….
I want to produce animated videos, like the zoom/spin ones where the AI morphs and invents in 3D space, but I want to base it on a pre existing image (innit) could anyone recommend a system to use? Thanks!

[near silo]

FFS, i give up. I have linked to the channel, and said it 3 damn times now. USE. AN. IMAGE. TAGGER.

[stable pawn]

And what other words or context it referee to when, AS AN EXAMPLE, we write "girls night out".

[near silo]

thats not a term an AI is gonna be trained on, that feels pretty obvious

I am gonna disengage from this conversation, cause I have linked and told you what you are asking for several times now, and you are choosing to not listen

[stable pawn]

Yes, it not trained on that, but can we know what words the AI used and what are best to use? That was the question.

[brisk bloom]

the problem I have with taggers is they always refer to 1girl, no matter the age of the female

[near silo]

And I have answered several times now

[stable pawn]

Like if I write "Doplamina", is there any script that can point it out as a word AI do not know and mark it as "not good".

[near silo]

1girl just tells the AI there is a female, and no other subject. It will do it with all things it knows. 1dog/1other/1cat. Its just to know there is nothing else

[warm junco]

The words the ai knows depends on the images a model get trained on. For Anime models most booru words work because they used images from there to train the model

[brisk bloom]

true, but you can type the age bracket of the person you wish to gen in the prompt and it will understand that

[stable pawn]

Yes I know that, but what I ask about is how we can now what are good words to use and what are unnecessary words in a prompt, like how do we know that "trending on artstation" is good or not in a prompt, do the model even know what trending and artstation is.

[warm junco]

We cant know that unfortunately

[near silo]

yeah, so like "1girl, old, aged, white hair"

it will know the subject is female, and they are an old female

[brisk bloom]

that depends on the model creator and what they used

[warm junco]

It depends on the words used to tag the images or what blip captioned the image as

[near silo]

that as well

you're just gonna have to spend some time and figure out what works for you

I found the words I need to get very high detail images that look very realistic

[stable pawn]

All I wanted to know if AI can rank usability of words or if we in some way can figure this out. Like hat do AI do when some write "best quality" and is that better than "super duper extra ordinary quality magic"?

[brisk bloom]

from my own experience using interrogators, they dont really provide more info that what you can already describe, unless its a specific art style you arent familiar with

[near silo]

again, just gonna have to figure it out yourself

fuck around a find out, as I like to say lol

[brisk bloom]

yep you definitely need to play with it. pretty soon you get the hang of it

[stable pawn]

@near silo Well, that was good to know, that is same to say, we do not know what words do and most what folk write is just random BS we have no idea what it do". I just do not believe in that, mostly for tags been used to train images and most models know those tags, but my question is, would it be possible to rank the "words" in a prompt.

[near silo]

as in put weight behind the words you put in?

[stable pawn]

There are scripts that optimize prompts and some to translate from tags to words, so let formulate the question as "would a script like that be able to use (as an example) chat GPT3 to sort what is effective or less effective in a prompt.

[novel bluff]

Hi, where can I find the option to generate a photo for my FB profile?

[stable pawn]

I did a test where I used "cute robot bunny" as a prompt, but when I added "worst quality" or "poor image quality" in the beginning this didn't alter the image quality but rather the subject (the bunny). Conclusion is (for me) that things like "best quality", "good quality" and so on is unnecessary prompts for AI can not know what we mean when we say "quality" or what "best quality" suppose to be. When adding "trending on artstudio" things did happen but not in the way I thought, some images was zoomed in, other had signatures and copyright notes, and some time I also got two or three bunnies from different directions, it did not change the image quality or so. I used SD 1.5.

Edit: "Intricate details", worked surprisingly well to add details to the subject.

[quasi leaf]

kinda have to work with images alot, if you want to base it on an actual image of yourself, i made my discord profile picture with controlnet and img2img, with prompting and inpainting to fix things i did not like. if you do not want to base it on an actual image, you just have to engineer the f out of prompts and find a model fitting what you want to do

[viral topaz]

Super stable diffusion is the same stable diffusion or it's completely different ?

[quasi leaf]

only references i can find to "super stable diffusion" is youtube videos from last year. i think its just stable diffusion and they called it super for effect

[fierce charm]

SSD is a UI for Stable Diffusion

[quasi leaf]

is it? Can you link it? 🙂 havent seen it before

[fierce charm]

I stand corrected. It's A1111 but called Super Stable Diffusion by a youtuber for some strange reason

[warm junco]

Yes Aitrepreneur just named it like that xD

[viral topaz]

Yes, it's his fault to make confusion 😁

[stable pawn]

My testing: I tested "Anime, illustration, " before my prompt "cute robot bunny" and it work well, but I had better result when I removed the separation like "anime illustration, " and "anime_liiustration, " was almost a bit better to make illustrative images. I use the normal Stable Diffusion v1.5 model in my test.
Edit: Even "anime_render" worked well and better than "anime, render,".

[wild fjord]

oi

[green brook]

Hi friends - I'm looking to upgrade GPU's for SD. What do you guys recommend with a ~800-1000 budget for gaming + SD performance? I was considering 3080 used or 3090 used currently.

[fervent thunder]

Yeah any high number 3000 series or 4000 series card is perfect

[green brook]

Thanks I figured - I was leaning more towards the 3090 range because of the 24 GB of vram vs the 12 gb from a 4070

[fervent thunder]

What resolution are you gaming on

[green brook]

1440p

[proper karma]

I would highly recommend 3090 used for this

[elfin arch]

@dapper cove
¿Esta mondá qué?

https://discord.com/channels/@me/1082596355039571968/1082596490079391766

Me quieres jaquear la cuenta o qué?

[keen yew]

what is the best stable-diffusion api or endpoint for javascript?

[near silo]

interesting

having a high refresh monitor dramatically affects SD performance

I just got a 165hz monitor, and even with nothing happening on it, my SD perfo on LoRA training drops from 2.02 it/s down to 1.72 it/s

I did not realize it made that big of a difference, crazy

[vast ingot]

For when I'm training, i find it important to move everything else to the motherboard, especially windows, and disable material acceleration in discord, to free every resources for training there. It can necessitate a restart for windows to let go of the VRAM it hogs too

[near silo]

I do run VRAM intensive programs through my iGPU, such as firefox when i watch youtube videos

it makes things wayyyy faster

but I guess having a 165hz monitor runs naturally on DGPU to forward the stream from iGPU

I just didn't realize it made that big of a difference

[keen yew]

@vast ingot

[woven sundial]

Does using xformers give deterministic results now?

[vast ingot]

I only know of 2. One would be the official API on dreamstudio using your API key there. The other one would be to host an automatic1111 install, open it to the web and open the API on it. First way is easier to set up for sure, second one will let you do maybe more but will be more complicated to set up.

[near silo]

what do you mean?

[keen yew]

ok thanks, with the option 2 i will be able to deploy an aplication that makes calls to the api for image generation ?

[woven sundial]

I've read and experienced enabling xformers breaks perfect deterministic results, but idk if this was fixed or not.

[near silo]

as in, it doesn't produce the exact same result on the same seed?

[woven sundial]

Yeah

[near silo]

it still does that, though the difference is small enough I don't think anybody cares

[woven sundial]

Thx!

[near silo]

the benefits WAY outweigh that itty bitty con

[vast ingot]

Both options will do the same, make it so you have a REST API you can call. so yeah, both will work.
Main difference is how much work you'll have to do to set up the api, and how costs will be repercuted.

[keen yew]

Which model is the best stable-diffusion-v2-1 or stable-diffusion-v1-4 etc in your expirience?

[keen yew]

thanks man

[vast ingot]

Best for what would be the question. 1.4 isn't really used anymore, 1.5 is just superior to it imo

2.1 is great for photorealism but i prefer 1.5 for the rest

But mostly i use like 100 different dream booth models depending on the occasion

[keen yew]

I'm asking this because I've only worked with text-related AI with openai, now i want to work also with image-related using stable-diffusion

[keen yew]

i will use 2.1 cuz i need photorealism

thanks for your help man 😉

[near silo]

there are plenty of 1.5 models that offer just as good photo realism, if not better

[vast ingot]

Go check civitai too then, there are lots of great models from the community too

[near silo]

@keen yewUse realistic vision 1.3

https://civitai.com/models/4201/realistic-vision-v13-fantasyai

[keen yew]

thanks man

look at #1042896447311454361

[near silo]

have a look in #🏞｜general-with-images to see what I have been able to make with realistic vision v1.3

[astral goblet]

experimenting with prompts is how. There's more too. With 1.x 2.x and potentially 3.x, they use different clip models for the pre training, and that gives the AI an entirely different dialect.

[astral goblet]

xformers always do give deterministic results, but they're not hardware agnostic. I think there's work done there to improve it, but the difference ebtween a seed with xformers and without xformers, i've never seen one that was a significant change

usually it's just a subtle difference and most of the composition is the same

[faint plume]

Guys what operating system do u use?

[vast ingot]

Hey people around here and other channels, there has been a certain wave of scam by PM recently. If you have been contacted about wining a cryptocurrency prize in a contest, don't hesitate to :
1/ report that person to the mod team through the #1010934719455707218 channel
2/ block them
Also of course, don't follow their instructions further

Lots of new users come around asking for this too, so if you see someone not sure, don't hesitate to direct them to this info please !

[fervent thunder]

win10

[astral goblet]

God damnit. Bing's chatgpt is more powerful than the base website i think. i'm getting wicked fast results out of it and just easily got it to write me a python script that looks for wallpaper sized images in a folder

this means i might have to start using bing

[rich plover]

where can we generate pics?

[surreal perch]

@rich plover

FAQ: How do I generate images?

Currently there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or running Stable Diffusion locally using your own system resources! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using the website or #🤝｜tech-support for any problems you encounter while installing it locally!

[inner tapir]

Is civit down for anyone else? Normally I see people freaking out when it goes out but nobody said anything

[vast ingot]

Got a PM asking the same, it seems to be

But like, it breaks a little everyday recently, so i see less and less panick

[mystic coral]

Hello 👋

[tight trench]

hi

[dusty pewter]

so for model merges like realistic vision 1.3 i notice it takes elements from other models. is it still useful to download some of the other seperate models too? like does it not contain all the date or do you only need that one? im a bit new to 3rd party models

[fervent thunder]

If you want it , download it

[vast ingot]

Model training don't make bigger models, it just specialise a model in a given way, making it learn things, but at the cost of forgetting other concepts, or maybe more get a little worse at a lot of other concepts.

Merging adds another layer to this. It takes multiple layers and merges them together, managing the weight of each, but still not making a bigger model. In the end, you have a little lower quality on each and every concept that was merged, compared to what you would have on the original dedicated model for that specific concept

[dusty pewter]

ah so its still worth downing a spacifc model in a merge or two if your interested in it? good to know

i wonder how long will i be till ai can keep the same charecter /clothes but change everythin else? now thatd be something lol

i think midjourny can kinda do that now but not quite sd

[little yoke]

Hi everyone. I’ve been looking for a good google colab for Textual Inversions. I have found a couple out there but with poor documentation and not much success getting them to run. Does anyone have a recommendation for one that you have successfully used?

[dusty pewter]

move over dazz3d! lol though daz3d is still useful for poses!

i think daz3d needs an update tho many of their charecters lately just arnt life like enough face wise lol

like the new victoria looks kinda meh

[wild steppe]

Try searching for it in the server or you can ask in #1011228442399883294

[hard osprey]

Any recommendations for a GPU for ~$300? Mine is just 3gb of vram. I am thinking an Nvidia 3060 or something, maybe the one that has 12gb

[dusty pewter]

the more vram the better for sure

i am happy i made a good choice on my pc 4 years ago and while its older it still is useful since it is the 1080 gtx ti and has 11 gb of ram 🙂

wise choice indeed there

wow

the 1080 gtx ti is still worth like 450 bucks lol even after all this time . crazy

[hard osprey]

I got mine a long time ago from my brother so I didn't actually pay for it xD specs are ok but the GPU is not

[woven sundial]

I can put SD on an external hard drive right?

[dusty pewter]

it helps if its on a faster drive i think tho

but i assume yes

[woven sundial]

Thx!

[dusty pewter]

man sd is great for me in my goals of one day starting a modeling agancy or at least getin more into that sort of photography even though i dont have the rl connections yet i can use it to generate the types of photos id like to do to show would be people interested exactly what id have in mind its great!

i used to use daz3d a bit for that but. eh its pricey for the figures and can take a good while to render stuff.

its amazng that u can use your own photos for imput . too :-p or what i like to sometiems do if i see a pic of a figure. i can photoshop em recompose the image then imput that into sd. and then generate something simular based on it but with my own framing 🙂

[ionic oak]

Hmm, I've been having a lot of trouble generating realistic animals for some reason.

[dusty pewter]

and the openpose addon rocks

[ionic oak]

Using the default model

[dusty pewter]

is there an animal based custom model i wonder? :-p

zoo diffusion!

[astral goblet]

that was top of the line for nvidia back then. Next step up were the Titan lines. they stuffed so much into the 1080ti. There was no 1090.

[dusty pewter]

i still am amazed we dont have a cat only/biased model yet

[astral goblet]

The flagship products tend to keep value for a few extra years

[dusty pewter]

all cats all the time!

wasnt the 1090 really just an OC 1070?

oooh wait

nm im thinking of somethin else

a 2080?

[astral goblet]

https://discord.com/channels/1002292111942635562/1056354736875507813 2.1 cats

no 2080 had rtx cores

[dusty pewter]

ah there we go

well what was the one that was really a reabranded 1070? i thought linus or someone said something about it

[astral goblet]

not sure. i had a vega 64 for the past 5 years

[dusty pewter]

back in my day we had 64 bits and we liked it!

[astral goblet]

maybe one of the 1660s

ngl, the branding on the vega 64 might've been part of why i got it

[dusty pewter]

ah that must be it

[astral goblet]

and i might've gone a little n64 kid on it when the box arrived

[dusty pewter]

so if you use 2.1 with a custom model i guess it doesnt matter if the dataset that comes with it sucks. tho i hear controlnet crashes with it more lately. ill stick with 1.5 for now

[astral goblet]

controlnet initial release was made for 1.5, and can't work on 2.1. but new models for cnet came out today that were trained for 2.1.

The dataset for 2.1 is mostly the same. people are just hyperbolic about the "censors" on it

the biggest difference with 2.1 is that it's structured differently and the openclip pretraining is trying to achieve a "flatter" latent space. I'm not sure what that means entirely, but it leads to lots of transitional difficulties with it

[dusty pewter]

but one guy showed how the images generated with 2.1 lack detail that 1.5 has

like for example a detailed statue and such

it seems like to me they took out a chunk of the old data and said screw it and thats what we have now lol

is it porn? is it legal? i dunno . chunk it!

the brain trust that is them :-p

[fathom cedar]

hey guys, If I want to train on clothing and specific male apparel, what's the recommended method of training, LORA or dreambooth model?

[astral goblet]

imo always default to lora first. non destructive and can be loaded with any refined version of the trained model

[dusty pewter]

we need more clothing based Loras. is it possible to just add more clothes? i dunno maybe not? lol thatd be cool.

[ionic oak]

I have seen some people generate pretty decent animals although most are very deformed

[dusty pewter]

lol username checks out

deformed kitties fund. donate to the deformed kittys of SD today!

[fathom cedar]

so going for Lora is better?

[fervent thunder]

It's not just clothes, SD has lots of problems with understanding some simple words and word combinations.

In most cases SD just don't know words I'm prompting , even tho they aren't like stupid uncommon.
In other cases I'd blame English language for being ambiguous...

[neat berry]

elo mates

[fathom cedar]

is there a place to guide me on how to start making a model/ lora for clothes? I've only seen guides for styles and characters

idk what category is clothing

[chrome spindle]

3080 probably, it was turned into 3070ti

wait, it was 4080 actually 😄

[gleaming lance]

What’s the workflow to create something like PhotoAI.io or deepagency.com? I saw some others ask this but can’t find the message. Is it like this?
thanks in advance for any tips.

1. First train your SD model with dreambooth using your images in different settings and poses and play around with the learning rate, steps and other numbers to get it right. Or is it different fine tunes models for different poses? The goal is to have the user upload their images and fine tune the model basesd on that

2. Download your .ckpt file that includes the weights and use controlnet on top of it?

3. Host it on replicate.com or huggingface to be able to connect to it via an API. (is replicate the best one?)

[lofty gate]

I wanna hop on the SD train again- should I just install from scratch and get all the popular add-ons? Or should I let automatic111 update it? I haven't updated SD since 2.0 released

[brisk bloom]

an update for the webui and extensions will be fine

no need for a fresh install

[lofty gate]

What extensions are worth getting? I heard loma let's you train on low vram cards and controlnet helps you pose subjects

[brisk bloom]

that depends on what you want to do or like

[lofty gate]

I just wanna know what's out there, any compilations ATM?

Appreciate your help :) btw

[brisk bloom]

controlnet is quite popular. using lora's you will need additional networks ext. open outpainting is quite good too

ultimate SD upscale ext is great for upscaling to large files without a huge amount of vram

[lofty gate]

I used automatic111 upscaler before and didn't have too much issue

[fervent thunder]

Quick question do you have to use a instance token and class token when making a lora with dreambooth???

[lofty gate]

I'll have to look into Lora's network thing. I thought it was just a trainer you could use yourself, darn!

[brisk bloom]

loras are like embeddings

[lofty gate]

As long as it's simpler than the Collab lol

[brisk bloom]

yeah local generation is much easier

[edgy heart]

I'm actually so bad at this what is the channel that can help?

Everything I do is a failed attempt

[wild steppe]

#1080946152318443610 or #🤝｜tech-support

[dark hemlock]

What happened to dreambot

[wild steppe]

Dreambot is taking some well earned rest #1047610792226340935

[dark hemlock]

Is it coming back?

[wild steppe]

Probably not. Although things are fluid and we have more planned.

[timid delta]

@thorny depot are there any folks in here thinking about social impact of ai. Economic and environmental ? aIm looking for people to talk about how ai is disrupting jobs and Data consumption and how to help solve for that

[lofty zealot]

Does anyone know where I can find Stable Diffusion's Terms of Service? Seems like the web page that had them is down currently.

[wild steppe]

https://stability.ai/terms-of-use

Could you tell me where you were redirected to the the broken page from?

[thorny depot]

Absolutely. As an artist myself, the thought has crossed my mind more than once.

[edgy heart]

I spammed the scam guy

Just for fun

[wild steppe]

https://tenor.com/view/uno-reverse-card-gif-26415214

[edgy heart]

Yeah exactly

Sad people that scam

Well I'm going to use stable diffusion and try to get better

[wild steppe]

Glad to have you here, we've got a nice community here #1080946152318443610, #📝｜prompting-help & #1072013871730131004 are very helpful for learning SD

[edgy heart]

Thanks man and hopefully we can remove any more scams and great ai it's good if you know how to use it I think

[warm tiger]

Anyone happen to know about an extention of some sort for wildcard prompts?

like it'll toss in a bunch of random prompts if you tell it to do wildcard ones?

seen a bot that does this but have not found a thing to do it myself

and is the Implement invisible watermark on automatic11111 just it adding the prompts list to the exif data? or does it add other things?

[unique patio]

QUICK WILL SOMEONE HELP? Need you to create a: Funny discord pfp for me! Please please

Or help me know the tools to use for it pls pls

I know stable has like a bunch of features and queries for things like size and stuf I think idrk but its useful right

[warm tiger]

funny in what way?

what sort of theme?

[unique patio]

Anything random

If youve been on discord servers with funny topics youll find funny pics

Anything's OK though I'd just like to see whats possible!

[warm tiger]

basically anything "is possible"

[slate dust]

I'd just take one from here haha https://discord.com/channels/1002292111942635562/1019361238234443776

[unique patio]

Ok ill check that out too lol thanks

[slate dust]

like this? https://github.com/adieyal/sd-dynamic-prompts

[quasi bough]

im new to stable diffusion...and i want to know if i can use a trained model with dreamstudio.ai api?

[unborn hedge]

The Constitution of the United States prohibits ex post facto accountability. Until a law is in place you can not transgress it.

[warm tiger]

i think that

's probably exactly what i'm looking for. thanks! ❤️

[warm tiger]

sadly they impliment laws and punish people for having broken them prior to the law being made constantly

[lament steeple]

That's the part people aren't taking into consideration. The law is already in place. They are determining if it ha been broken and it's not a criminal case. It's a civil case. Completely different animal

[warm tiger]

dusty is correct when it comes to the copyright thing.

[lament steeple]

And so am I when I say. Your but. Armor it or don't not my problem. My advise is to armor it

[warm tiger]

also true

might as well protect yourself in the event something does happen

what i see happening is the model creators being fine. but people that generate art that looks too much like a cerain persons art. and then potentially profiting off it would be a good target for it

[unborn hedge]

You're talking specifically about the scraping case.. In that instance the claim is that they violated rights as scraped a commercial site that licenses content and circumvented protection by removing watermarks.. We'll see how that plays out..

[lament steeple]

Then comes the question of whether or not using that artist name in your prompts violates their copyright. Yes it's a gray area, but one that can go black real quick and better to not be in the shadows if that happens

[warm tiger]

eeeeeh

[unborn hedge]

That is an incorrect assumption.. There is no protection for stylization, that is explicitly excluded from copyright. that's why Disney can't sue the endless people who've imitated their styles.

An artist like Koonz Trademarked his ballon dog that's the only reason why he can sue to protect it,

[warm tiger]

i didn't mention stylization. i've seem some AI generated images that basically looked like a fuzzy but nearly identical version of an actual art piece

with a character owned by said artist

[unborn hedge]

That's def theft..

100%

[lament steeple]

OK hear me out. You put an artist name in your prompt, you are tying their name to your work. You do that in isolation and you may be asking for problems

all I said was to use more than one and to hide your prompts if you do

[unborn hedge]

No an artist does not own the rights to their name unless they are celebrity and then only in limited capacity

[warm tiger]

is there any way to check someones prompt list on a photo if they remove exif data?

[lament steeple]

These are not unreasonable steps to take to protect yourself

[lament steeple]

no

[warm tiger]

they potentially could. alot of artists are pretty famous

[unborn hedge]

Famous is not celebrity and there is plenty of exceptions for fair use.

[warm tiger]

but they'd basically need to trademark the name and that's a whole other ballpark than copyright

[unborn hedge]

You'd be hard pressed to be able to trademark a name.. It has to be a brand.

[warm tiger]

well as the laws currently stand if a model creator that makes the model and ditributes it free of charge IS actually found guilty it'd be a gross misuse of the law

[unborn hedge]

And only in the context of the brand.

[warm tiger]

a lot of artists their name is their brand

[unborn hedge]

We don't get to just call something a brand.. it has to be defensible to prior claims.

[unborn hedge]

a trademark or service mark must be distinctive and not likely to be confused with any existing trademarks or service marks.

[warm tiger]

i mean if the person literally sells crap with the name on it. it's the name of their store, their company and their watermark on their artwork how would that not qualify?

[unborn hedge]

and it only applies to that specific domain.. So Porsche can't go after a person with the same name who makes sneakers.. It has to create confusion in the market..

[warm tiger]

which in this case it would

digital artwork and digital artwork XD

[unborn hedge]

Simply creating digital work is not unfair competition.. you have to prove that there are damages, as in lost revenue..

[warm tiger]

that'd be for copyright, yes.

there's also such a thing as "statutory damages" which can include up to something like 140,000 USD in payment without needing to prove any loss of income

[unborn hedge]

Trademarks have to individually registered..

[warm tiger]

i've been through multiple copyright lawsuits

[lament steeple]

I can't afford that

[unborn hedge]

I used to run a IP rights management platform, we constantly provided this for cases..

[warm tiger]

aaand in the event said person cant pay the ammount. you can have their waged garnished,,,, garnered? they take a chunk of each paycheck and it goes to you till they pay it off

[lament steeple]

So someone who is retired without a pension?

oh yeah. That's contempt of court..

[warm tiger]

they often have physical assets they could be forced to sell off

[lament steeple]

Not in my case and not in a non community property state

They get my laptop

[warm tiger]

i wouldn't know at that point. every case i was involved in they were able to pay

[unborn hedge]

what were the circumstances of these cases?

[lament steeple]

So forgive me if I don't take unnecessary risks

I do well enough without them anyways

[warm tiger]

due to this being a no NSFW chat i can't go into detail but that should tell you enough

[unborn hedge]

Generalize.. was it that images were being redistributed?

[warm tiger]

yes

keep in mind a lot of this started about 20 years ago and the last case was around 8 years ago

so way way WAY before DCMA takedown notices existed

[shy pumice]

I mean Digital Milenia

like, it's for this whole milenia

[unborn hedge]

So that is a totally different situation than what we have with AI Model.. that is literally copying and distributing..
These AI create statistical models of how pixels relate to words, they are in effect creating wholly new works..
These works BTW are completely unprotectable right now as copyright explicitly states it has to be human created.

[lament steeple]

That depends on how much human involvement there is after generation

[warm tiger]

i know. i was going into detail on you CAN still be charged and forced to pay money even if your actions dont cause loss of income to someone else

through statutury damages

and then gave a personal example

[unborn hedge]

True.. you have to show post production..

[lament steeple]

Why I keep all my files

[unborn hedge]

Perhaps but that case law has not been established which is what I called out in the very beginning of this conversation (yesterday)

[warm tiger]

true.

[unborn hedge]

Right now all we have is a bunch of people speculating on something that even the most brilliant legal scholars don't understand..

[warm tiger]

protecting against a potential lawsuit isn't a bad idea though. especially with various hot topics

[unborn hedge]

This is the point of our copyright system, it lags behind to keep from stifling innovation..

[warm tiger]

you have no way of knowing which way this will swing. and a lot of polititions are leaning towards censorship

[shy pumice]

what about the photography argument, you guys managed to square that?

[warm tiger]

what about it?

[shy pumice]

the minimum human involvement in it

[warm tiger]

photographs are copyright of the person taking them. though the items IN the photo may belong to others

generally unless explicitly stated otherwise the copyright remains with the one who created it

[lament steeple]

When you generate an ai image, where did you point your camera?

[shy pumice]

somewhere in latent space

I guess

[unborn hedge]

We have case law that actually says creature who took it.. After the monkey who took a picture that was used commercially.

[warm tiger]

love it

XD

stuff is much the same with various art comissions for furry art

weeee

[lament steeple]

You have to be able to separate the image from it's prompt, seed, and model

[shy pumice]

for what purpose?

[warm tiger]

but it gets a bit muddled when you pay an artist to create art for you and they attempt to keep the copyright of the artwork you paid them for

[lament steeple]

That's a contract law case

[warm tiger]

yup. a messy ones that i've never seen actuallly resolved

[lament steeple]

To show that you created it and not the machine

[shy pumice]

how does that show you created it?

[lament steeple]

because without the separation, I can recreate it without any further involvement on my behalf

[shy pumice]

Any further involvement other than replicating the process?

[warm tiger]

you could always remove the shadow of a doubt and provide heavy post processing. like Photoshop editing of a photo that requires you to manually work on it before posting it.

that'd eliminate the "but they didnt work on it/create it" argument

[shy pumice]

but what if you really didn't

[lament steeple]

Which if you are going to do any enlargement you are going to need to do

[shy pumice]

you'd still have to be minimally involved

[warm tiger]

you keep the source imaage as a backup and release a smaller version of the image

enlargment is still AI

you're just telling it todo so

same as clicking "generate"

[shy pumice]

it's algorithmic but it's not AI

well

unless it is

[warm tiger]

i'm talking about actual heavy editing done manually after the generation proccess is done

[warm tiger]

that's what AI is though

and if you noticed no other non-AI upscalling manages to eliminate pixelation like AI upscalling does

[shy pumice]

I guess what I mean is it's irrelevant whether or not it's AI

it'd be relevant if it was AGI I suppose

or maybe very advanced AI, not sure where the cutoff point would be

[lament steeple]

There really is a burden of proof as to how it was created. What is required is that a creature holds the copyright

[warm tiger]

most models you can download also directly stipulate nothing made with them can even have a copyright and that you agree to that before even using it

[shy pumice]

human creature

[warm tiger]

at least most of the ones i've seen so far

[shy pumice]

can that possibly hold up in court

[warm tiger]

some checkpoint merges wont say that but if it's using X and Y models that DO say that, same thing

[warm tiger]

i'd assume it could. i mean i doubt it's legally binding but i mean you ARE generating content off the backs of alot of other artists.

[shy pumice]

yeah, I'm doing a lot of stuff off the backs of other people I guess

don't think it's relevant

[lament steeple]

Copyright has been granted to a monkey

[shy pumice]

no

it only applies to humans

[warm tiger]

and i mean, unless it's a model you yourself trained or your doing a LOT of manual work on it after generation someone else could create an almost identical piece

[shy pumice]

for now

[warm tiger]

that'd be incorrect

[shy pumice]

really?

[warm tiger]

happened in 2015

[shy pumice]

seems unlikely to me

I don't think non-human creatures are intelligent enough

not any that we are aware of anyway

[warm tiger]

it appears that they did not rule in favor of the monkey

i dug into it a bit and finding the conclusion of the case was kind of a pita

but yeah. they ruled against the human but also said that a monkey, being a non-human couldn't hold the copyright.

@shy pumice @lament steeple @unborn hedge

[hallow hatch]

@queen karmahsage deleted@

[warm tiger]

?

[unborn hedge]

Right that was the case! But not so different from Generative AI..

[hallow hatch]

@ivory lynxl the poowowher of my. himmmmmaaaaaaaaaaatch iiiiiick@

[warm tiger]

Are you okay?

[hallow hatch]

ima let you finnish silence, but right quick i like que ‘ce; bae en ‘ce had by far! the funniest music video thi’ce year. aright tayr you ‘ca fini’ce

[unborn hedge]

Must be a LLM.. they produce gibberish when you exceed the vector space..

[hallow hatch]

heir roar core wrecked eng

sorry guys i’m entropy based algorithms engballs right now

is this the line for printstring?

“helluva world”

[hallow hatch]

de_ddos at dix. yamm daddu. sexierchild like. ur b& from all reminders for the rest of the night. donut make me remind u;

[forest apex]

man, you need to upgrade your brain matter

[rigid ember]

Question: How do I take a photo and keep it like completely the same but change the actor's pose?

I've learned how to detect pose in the pose editor, fiddle with it, and export to controlnet, but I'm having a hard time getting it to cooperate. I think I was once able to get an arm to reposition using that technique, but it was very minor and the change kept the arm largely in the same place, but change the relative pose.

I'm having a hard time with, say, taking someone's arm that's across their body and moving it to their side.

I'll inpaint where I want the arms to no longer be and where I want their new placement to be (guided buy the pose in controlnet, but it ends up changing a lot of the image, when really I'm just trying to get arms to be moved/repositioned.

And I'm also wondering to what degree my failure is coming from certain settings not being adjusted like denoise and what they should be set to with this kind of goal in mind. Or what other controlnet I should be using besides openpose.

Keep in mind, I'm trying to do this with img2img with stuff just taken from online, not produced by the ai.

[hallow hatch]

want.exe. scam for virii first

[warm tiger]

i'd say run img2img with the same prompts but shift them slightly to mention the pose you want

maybe turn up or down CFG scale a bit or ( ) or { } as needed

[rigid ember]

And what does CFG scale do exactly in a basic sense? Just one of the hundreds of things I'm working on getting answers to. lol

[warm tiger]

higher the number the more stictly it adheres to all your prompts. lower gives it more ability to ignore them

going too far in either direction generated gibberish

i tend to turn it up or down 0.5 at a time then test with the same prompts

[rigid ember]

So how is that different from denoise? Because that's how I understood denoise - the higher (0.9+) the less freedom the ai is given/more strict with what you want?

[waxen pasture]

denoising adds noise to an image giving the ai more freedom, higher denoise = more noise, so a denoising of 1 wouldn't cohere to your image at all and a denoise of 0 would just produce your image with nothing else

idk what this conversation is about but I hope that helps.

[unique knot]

Hey guys. After making image i wanted in 512x512 resolution i want to generate same one but 1024x1024 but it gets some more stuff there like other people. How to fix that?

[lament steeple]

other way around

The lower your denoise, the more strictly it stays to the original picture

[lament steeple]

Send it to extras and set the upscale to 2. pick an upscaler and click generate. that will just super rez your picture

[rigid ember]

I gotcha. I'll play with that.

How do I set up a "save" folder for outputs in colab? The folder image/first icon doesn't open anything up so I can't select a path (or I'm just not seeing something) and the "save" button just forces the image presently displayed to be explicitly selected (after like 2-3 seconds) so that you can manually send it to your browser downloads driectory by clicking download. But I'm not sure how to set up a saves folder in colab where when I click save, it sends a copy of it to a specific save directory in gdrive instead of everything (good and bad) that auto saves to the general outputs directory

[astral goblet]

higher denoising adds more noise to the image for the diffusion process to remove, so more of the original is destroyed

[hard osprey]

I was looking at graphics cards and some of them on eBay like the Quadro 5000 have 16gb vram but are much cheaper than other cards that have even less vram. What am I missing here?

[uneven heath]

Oh hey, Happy Holi to Indian bros out there 🥳 Nice day to controlnet shit out of holi pictures xD

[astral goblet]

it's ancient and doesn't have as fast cuda cores and doesn't have any tensor cores

[hard osprey]

Oh, so it's an exception to the general rule of high vram

Because everything else is bad on it

[astral goblet]

older professional end cards had loads of vram in their time. those were made for cinema work and vfx usually

[hard osprey]

I'm thinking of something in the $300-350 range. I think most likely a 30xx with 12gb vram is the best option?

[astral goblet]

30 series for a $350 is quality grab. 12gb isn't great but it's a solid score still

i paid $1600 (CAD) for for a 4080. not a great score. i had money to blow and didn't want to compromise for once

gpu prices are hard to look at now a days

[hard osprey]

I see, it is pretty fast though.

Do you think a 20xx series with 12gb vram can generate a 2048x2048 image? It's cheaper but still not as slow as the Quadro I think

[fast jungle]

im just a bit confused, how are models like deliberate made?

like, do they train more images?

[astral goblet]

deliberate i think is a merge. they take other models that they like, and merge their own. sometimes these guys dreambooth refine one they like, and merge it a bunch of times again too. Merging sort of mashes all the knowledge into one consistent soup though, i've found. they all have similar faces and super similar outputs

[unique knot]

Isnt it better to generate bigger one instead of upscaling?

[fallow ferry]

Hi can I ask, my laptop is too old so I tried run stable diffusion on google collab, it's that same stable diffusion with local?

[lament steeple]

i's about 50/50. other option though I'm not one that uses it because it doesn't give me as much control is to use the hires fix with the same seed. original resolution 512x512 and output 1024x1024

[astral goblet]

sd ultimate upscale script is a nice option too. a little more tweaky to use though

[unique knot]

Ok thx

[unique knot]

Maybe i can use img2img from 512 to 1024 and it will make me same image but with more details?

[lament steeple]

I have done that

[subtle glen]

making my first model rn, should b done in 35 min

ant tips ?

[unique knot]

What are u learning it to?

[wicked sandal]

Random Question:

Is it possible to host a Fine Tuned Stable Diffusion Model on a website?

fined

[unique knot]

Not so random question:
Is it possible to combine two or more images that i have as files and not text prompts?

[rigid ember]

Am I able to train a LORA in automatic1111? Or do I need local hardware to do that? I am trying to see if I can train my own loras/keywords that can be used freely between models and are not tied down to a specific model.

And if I am able to, anyone have a link to an article or youtube video that walks me through it? Everything I found seems to be local hardware based or has the limitation of needing to be model tethered

[signal dome]

Do people realise this is not help chat

[unique knot]

Not really

[vast ingot]

You may find better help in either #🤝｜tech-support or #🔧｜finetune

[deft bronze]

yeah, photoshop

[deft bronze]

does this even work? that sounds cool if it does

[vast ingot]

It does work, it's called full fine tuning, you train on so many tokens you try to update the whole model, change all the weights to push all results towards what you want.
But it's rarely done on LoRA from what I know, since it tries to go for the full model, when LoRA intentionally limits how much weights it updates in the model.

You can find multiple models that came from full fine tune, even if those are a lot more rare than classic dreambooth or others. They require quite a lot bigger dataset, with a good captioning system, and a lot of training, like a lot a lot

[unique knot]

That's 400iq solution 👍🏿

[exotic latch]

Highly recommend this AI forecast webinar on how AI will change in the next 1 - 3 years and how to strategize around these changes.
When: this Thursday, March 9th, at 10:00 a.m. PST
Where: 🫴 https://lu.ma/AIfuture

[fervent thunder]

Question so I made a very crappy/good lora with dreambooth and it only renders semi decent results if I use highres but gives pretty good ones without anybody know as to why this would happen or if there is any fixes??

[vast ingot]

Did you try to change the upscaler used by the high-res fix ? It saved me on some models

[fervent thunder]

No I haven't Ive always used latent? any other recommendations?

[fervent thunder]

do you have any recommendations??

[vast ingot]

Outside of that one I gave, I'm really not sure what's creating that phenomenon... Logically, it would mean that our dataset didn't incorporate enough close ups of our subjects, and that the latent upscale doesn't know what details to add. But the few times I had that problem i couldn't correlate this logic with my dataset, and didn't manage to fix it correctly yet

[fervent thunder]

Got it There's always the possibility its just my pc being garbage as well thanks for the advice regardless

[bleak matrix]

Good morning, everyone. How are we all today?

[cold sage]

@crude star is dry DMing people in here with spam, I've never even chatted in here before so they're just going through the user list. https://i.imgur.com/hdgtH5J.png

[bleak matrix]

N o t anymore. 🙂

Thanks!

If you ever seen any spam like that, just file a #1010934719455707218

Ticket with a screenshot and the full username

So, what's everyone up to?

[lament steeple]

making pictures from song lyrics and thinking about sleep

[woeful patio]

Hey guys! Is there any good AI drawing group on telegram?

[bleak matrix]

That's always fun! You can get some super creative stuff!!

[zenith thunder]

well i broke my sd

[bleak matrix]

Oh no! What happened?

[vast ingot]

hey sunny ❤️ happy to see you ! hope you're doing ok today !

[sleek osprey]

whats this about sd xl

[vast ingot]

[sleek osprey]

is there an article or something

[bleak matrix]

Hey Guis! I woke up earlier than usual and couldn't go back to sleep but it's all good, ig.

[vast ingot]

I think there are only some tease here and there for now, i didn't read or see anything official on it yet

[bleak matrix]

I hope you are doing well! It's good to see you!

Hm?

[vast ingot]

I'm doing quite well ! Even if a little anxious about some time some things can take

[bleak matrix]

Good to hear! Rip time

Well, you're here, we're all here, and that's what matters!

[vast ingot]

Yep, rip my time, i just bought monster hunter rise+sunbreak

(monster model incoming in some weeks)

[bleak matrix]

It's gaming time

She says as she devotes almost all of her time to SD

I have so many games to play

MH is great

[vast ingot]

I loved MH world

I spent around 600h in it lol

First monster hunter i played

[alpine bear]

[flint harbor]

🍟

[bleak matrix]

🍟

I have played a few of them

But yeah, you can sink some serious hours into them lmfao

[vast ingot]

Damn, the dog in that monster hunter makes my dog react

I love how they let you put your pets in it with you though

[bleak matrix]

laughs it's always funny how animals react to a sounds or things on a screen

[upbeat tangle]

hello how i can select vae manual?

[visual oracle]

Hello all!
Stable Diffusion was running very fine!
But suddenly - since today - if I want to "create" images - my grahpic card doens not anymore run the full speed core clock.
By creating images with stable diffusion - GPU was running with maximum core clock at 1902 MHz - now it only runs with half of them: 855 MHz.
All other 3D - application running very good.
I did not change anything in SD.
What is wrong? I absolutly don't know what I can do.
Can please someone help?

[amber cairn]

how to report people that send invites to a different server in DMs?

https://i.imgur.com/aF0wpXs.png

[brazen linden]

I just came here to thank Emad again, He knew when he was doing when he made SD open source. @wise stratus , the seed you planted will grow forever even if all models 2.0 + are crap, the community models and the community in general is what it what it should be. Everything is in ints right place,

[amber cairn]

@bleak matrix ^^ if this is the wrong channel, which channel do I need to be in to report... this guy?

[vast ingot]

You can block them first of all. If you believe what they send can be dangerous, you can contact the mod team with those informations on the #1010934719455707218 channel

And last thing, you can change settings in discord to refuse PM from people you don't add first. But there isn't a lot we can do to prevent people contacting people i think, it's a problem at times depending on what gets send by PM

[amber cairn]

oh yeah, it was a DM request, so it didn't show up at first
and I have no idea if I am the only one that got an DM, there might be more people, and maybe they did enter that server shown in the image

[vast ingot]

There were some crypto scam running on PMs yesterday... I advise anyone to stay alert, but since we don't allow discord invites on the discord itself, people tend to send those by pm. I'm not sure that one is directly threatening, what happens on that discord there may be.
So yeah, stay safe in case of doubt. I know I managed to fell for one about 10 days ago... Not proud but i can just invite people to stay alert

[amber cairn]

I have no idea why they are XD
they never chatted in here either..
I just wanted to let mods know, what they do with it, idc XD

[vast ingot]

Go for the #1010934719455707218 then, best way to achieve that 👍

[amber cairn]

yeah, already made a ticket

[crystal patrol]

Anyone got any experience training on a 7900xtx?

[halcyon granite]

any tips of rendering at high Resolutions (now I have a 3090)? For example does higher steps help?

[warm junco]

You want to use highres fix for higher resolution

[wild steppe]

Looking to get started with Stable Diffusion?

Find the most suitable method for yourself by visiting our #1080946152318443610 channel! We have a curated set of the most popular ways to use SD - whether you're looking for a fast online service or a customizable local installation.

If you have any further questions during your installation or regarding any of the sections, please visit the channels linked within that category, browse the #1072229020520947753 or ask the community here in #🤝｜tech-support!

[quasi leaf]

is there an invite link to the discord anywhere? i cant find it

[wild steppe]

https://discord.gg/stablediffusion

[quasi leaf]

thank you

[chrome spindle]

Why are models on civitai missing yaml files?

[frosty saffron]

anybody got tips for making lora outputs less digitally "crunchy"?

[tacit bronze]

@bleak matrix
@deep crescent is bot sending PMs

[bleak matrix]

You're fine! Just use #1010934719455707218 with a screenshot and the full name.

[amber cairn]

already done :)

[bleak matrix]

Okay! If you could send a screenshot with a ticket, that would be great!

[amber cairn]

it has been dealt with

[bleak matrix]

❤️

[stable pawn]

I started to test prompts and tags to see how I could improve my prompts, I started with single tags and expanded to simple prompts, but I need what can be seen as a "reference prompt", so is there and prompts that are seen as "Classical AI Image generation prompts" or some famous prompt that can be used as test and reference?

[wet basin]

Home

[chrome spindle]

Is there a guide to upscaling?

[woeful basin]

does thispersondoesnotexist is no more

[crystal patrol]

Anyone got any experience training on the rx 7900 series?

[proud nova]

It does exist but under other URL

Just Google it

[dusty pewter]

whats a good upscaller for images made with realsitic vision? i tried a few but they all seem to remove some of the finer detail in the image

[proud nova]

Latent (Nearest-exact)

Change denoising strength to something higher if it does artifacts

[dusty pewter]

gotcha

[woeful basin]

https://this-person-does-not-exist.com/en this is not usefull as that one, are you talking about this

[proud nova]

Ig

[chrome spindle]

Do i understand correctly that you can use controlnet with any model?

[thorn hare]

is there something like NovelAI text/story generation for stable diffusion, or something else for free ?

[alpine bear]

@frank escarp sending server invite links to random people is the online version of solicitation and it is straight frowned upon.

[vast ingot]

not yet to my knowledge

I hope soon

[vast ingot]

more than frowned upon, going to lengths in order to reach the most people, circumventing the limits set by the bot here (no discord invite), is a ban-able offence. There has been a lot of report about this lately. I invite anyone getting random server invites to report those to #1010934719455707218 so we can take action

(and yeah, I'm back baby!)

[alpine bear]

PIIIIIIINK

[vast ingot]

YEAHHHHHH

just got it back, I got to say, the security at S.AI is quite the thing 🙂 but I'm thankfull I got cleared again

[pliant iris]

Does anybody has working colab? There seem to be update and it broke mine

[fervent thunder]

was using this one yesterday
https://colab.research.google.com/github/sagiodev/stablediffusion_webui/blob/master/StableDiffusionUI_ngrok_sagiodev.ipynb#scrollTo=gJBtadxGOK7b

There are some more
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Online-Services

[jade tide]

all dreambots ch are close, right?

[fervent thunder]

Kaggle banned me for some reason...no explanation , not answering email for 2 days...great

[chrome spindle]

How do i purge vram after generating huge batches of images?

[crystal patrol]

Anyone got any experience training on the rx 7900 series?

[fervent thunder]

https://civitai.com/models/16768/civitai-helper-sd-webui-civitai-extension
That's interesting

[fervent thunder]

idk , I just resstart console thing if I feel VRAM leak (always happens after controlnet model swap, probably caching unnessesarry shit)

[chrome spindle]

i tried that before and it didnt help

[dusty pewter]

i saw there is a realistic vision 1.4 is that only on the paid website? lol

[dapper gate]

Isn’t it just here? https://huggingface.co/SG161222/Realistic_Vision_V1.4_Fantasy.ai/tree/main

[normal coral]

does anyone have any experience with training Loras?

[alpine bear]

I do
Just trained my first LoRA last night!

[normal coral]

were you training an object?

or face/style

cuz my question is about objects

[alpine bear]

I trained a hairstyle, or rather a specific texture to be applied to any hair in an image.

[lofty gate]

Does Lora require a separate installation or can it be done in an extension?

[alpine bear]

Dreambooth extension can do it, but the dedicated Kohya webui is highly recommended. Check out Aitrepreneur's video on it.

[lofty gate]

Thanks yo 😊

[normal coral]

@alpine bear sec i just stumbled upon something that might answer my question but ill probably have some for u after

[lofty gate]

I have that installed and was hoping to use the extension instead of the separate thing but if it works it works!

[alpine bear]

I haven't personally trained lora with dreambooth extension, and I didn't bother trying cause I've heard it scarcely works properly. There's better documentation for Kohya's lora webui, so that's the route I took.

[rancid trellis]

hi

i need help

[alpine bear]

Hello! I may be able to help :V

[rancid trellis]

:0 really?

[alpine bear]

Potentially :V

[vast ingot]

:V

[rancid trellis]

I have a problem starting automatic1111 in google colab

[alpine bear]

Aaaaand I have no skill with collab - Guiz

[rancid trellis]

oh 😭

[vast ingot]

you may find the best help for that in #🤝｜tech-support usually. We don't want to spam tech talk here, and there is no image sharing in the current channel

but yeah, there will be help there, or in the #1011228442399883294 channel too (less populated). See you in Tech 😉

[normal coral]

@alpine bear how did you caption your hairstyle lora?

[rancid trellis]

thankss

[normal coral]

from what ive seen the best way to train stuff like that is using a unique name in front of the keyword stable diffusion is already aware of?

[alpine bear]

I used the WD1.4 tagger extension, then did some hand-tuned prompt edits in the captioned files to make sure they were accurate.

[normal coral]

so im guessing you did something along the lines of Trexdel hairstyle on every .txt?

ah

https://www.reddit.com/r/StableDiffusion/comments/11b3i30/5_training_methods_compared_with_a_clear_winner/ cuz im looking at this

[alpine bear]

My unique keyword in every tagged image was 'Doritos Hair'

[normal coral]

reasonable

going by that post thats pretty much the only thing it had to be

from what i understand

[alpine bear]

If we head over to #🤝｜tech-support I can share the training data I used - should give you a better idea of what the captioning should look like

[normal coral]

sure

[copper saffron]

Hi newbie here, may I ask how can I start understanding what is the difference between Controlnet and Stable Diffusion? Also can we install Controlnet in our workstation?

[wintry stream]

Take a look over at #1080946152318443610! Tons of awesome information there to get started!

[daring acorn]

does anyone know how to fix this in colab automatic1111 anyone know how to fix this File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/webui.py", line 7, in <module>
from fastapi import FastAPI
ModuleNotFoundError: No module named 'fastapi' in
in colab automatic1111 ?

[astral goblet]

controlnet is a tool alongside stable diffusion. think of it like a guidance layer

[vast ingot]

it's complementary, and it puts constraints/guidance on what the model is trying to make. It's very potent to control what will compose your picture, how, where, in what posture, ...

I did an example a little earlier here https://discordapp.com/channels/1002292111942635562/1011634831467221033/1083124919220584478

[dusty pewter]

im using sd to make a mock modeling agency website to inspire me to hopefully make one for real one day. using some of the loras/texture inversions as "models" lol fun times 🙂

[copper saffron]

Thanks! I think your example and explanation really helps too. ☺️

[fervent thunder]

Just try another notebook, there are plenty of them

and make sure you are using gpu.
Sounds dumb , but just incase, won't hurt to doublecheck

[daring acorn]

Tnx will try

[lament dune]

Anyone receiving the spam DM quoted below should block the sender and open a ticket in #1010934719455707218 to report the spam. Don't forget to include a screenshot of the DM.

I noticed that you're interested in AI art, so I’d like to invite you to the Maze Guru ! Maze Guru is a free AI art generating tool. There will be new styles arrive every week, and support api acess,bring you new creative ideas.You can use it with Stable to make your artwork unique! Feel free to join, no pressure！
(Just sending this for an invite campaign, I hope this doesn’t annoy you)
[Invite Redacted]

[vast ingot]

TOS things will be tossed

[dusty pewter]

is there a more realsitic model then realistic visoin? and is the 1.4 only on their paid website lol?

[vast ingot]

lol it seems so yes, they are trying to put a licence on it (even though it seems to be a mix of lots of models they don't own the rights of)

[dusty pewter]

lol bad them lol

thank goodness i not need it commerically :-p

[vast ingot]

not sure about more realistic yet, sorry no recommendations from me :/

[ebon tartan]

hey guys, my stable diffusion started to be very slow and doesn't load things, it hangs, and a grey window appears when I upload the image to inpaint, it was working normally, would anyone know what it could be?

[vast ingot]

maybe try in #🤝｜tech-support , you'll find the best people for this there usually

[ebon tartan]

thank you

[vast ingot]

reddit seems to be loving the drama https://www.reddit.com/r/StableDiffusion/comments/11m280v/fantasyai_claims_exclusive_rights_to_models_that/

[wraith night]

Kobaldai is likely the only one that is free, the requirements for text are pretty high, you can only use the small models on normal cards (I have a 3090 and that fits a 3gb model

[alpine bear]

hmm, wrong reply?

[halcyon granite]

anyone know if grouping words has the same effect as individual?

e.g. [house], [tree] Vs [house, tree]

trying to decide if I should bracket entire wildcards groups (in prompt) or individual words (inside the txt file)

[fervent thunder]

should be the same, but better to try with x\y\z plot

[steep mortar]

Has anyone come across an extension that helps you to generate solid color backgrounds for the images you create? I'm trying to create some characters without some landscape, cities, etc in the background

[vestal dew]

try "yearbook photo"

it then puts up a background like you'd see in a yearbook

Or, do img2img with a solid color feeder img. Set denoise to 1. It will place whatever you prompt on the solid color

add the color to the prompt. If black, for example, add "black background" to prompt

[steep mortar]

Thnx for the info

[vestal dew]

sure

[copper saffron]

Hi Community, not sure if this is a noob qn.. is there a way to train SD to only have your own artwork and then allow the ai models to make it nicer? That means i don't want any remnants of other artists works.

[glad yoke]

I need some help with --api in my webui. I have it in there but when I load Automatic 1111 it always say that I don't have it in there. What is the problem?

[lilac pine]

Guys how do i train a LoRA with different trigger keywords?

[hard osprey]

I was curious if anyone knew of people doing graphic novels or comics with stable diffusion? I would love any links or anything to peoples work.

[lilac pine]

Training SD completely on your own style of art is really impossible, as SD is already trained on thousands of other artists’ works. The closest thing you can do is to make a LoRA, Textual Inversion, Hypernetwork, or Dreambooth. They give you the ability to make the image it generates fit to your art style.

Personally, I recommend you to use LoRA.

[copper saffron]

Thanks for the clarification ☺️

[lilac pine]

You wouldn’t want to train SD from scratch man 💀
Anyway you’re welcome

[paper prawn]

hi

[vivid isle]

Hey, I got a question, is there a free AI voice deepfake out there where I can use clips to make my own voices, etc? Is there one like Stable Diffusion where I can download it myself?

[fervent thunder]

there is...but I don't remember how it's called =.=

oh...download for youself probably not, I've seen service which can do it

[rigid ember]

I'm also new here/to this, but I think it's because colab is down and needs an update? Does anyone know if it's still down? Someone told me they had been working on it all day, but I just tried to hop in and I still get the same error across dreambooth and automatic1111

[fervent thunder]

collab I'm using works just fine

[ocean mango]

The Cr[AI]tive Revolution - The Future of Art

[foggy bone]

what notebook are you using?

[ripe walrus]

Guys just checking I am using RTX 3070 on a laptop, when generating image with chilloutmix, i am running at around 2-3 it/s, and the GPU utilization rate is only at less than 5%. is the it/s and utilization rate normal?

[jovial maple]

Sup

[polar osprey]

This @heady tree person is spamming members of this server with invite links

[stable pawn]

I did not get my invite from that person, but I got (what I think is same) invite, they can be sent from hacked accounts.

[copper saffron]

May I ask when we start SD, do you guys usually go to the webui user batch file to run everytime?

[vast ingot]

you can put a shortcut to it but yes, basically

[copper saffron]

Thanks!

[bleak matrix]

Good morrow, people of the c h a t. How are we?

[bleak matrix]

Please send in a ticket with a screenshot

[fervent thunder]

Is there a way to flush the Torch active/reserved cache?

[near silo]

does SD work on m1/2 macs?

it appears it does, amazing

[devout hull]

Hello, i have an issue reaching my stable diffusion from outside my home network... ports are open and firewall set up... I am able connecting to InvokeAI but not to duffusion.. same settings beside the different port.

[fervent thunder]

Hey loduo

[sleek osprey]

Has anyone tried 8 bit quantisation on stable diffusion?

Howis the quality?

[leaden mural]

I can help ya, what up? We'll discuss in #🤝｜tech-support .

[leaden mural]

W E A R E F I N E

[merry thicket]

is anyone successfully running control net on auto1111 on a mac m1?

[dusty tree]

anyone know ultra realistic models?

[stable pawn]

Depends on what you think of when you say "ultra realistic", there is many models for realism, you can brows and see examples at https://civitai.com/

[rich trout]

Deliberate

[fervent thunder]

Wha do you use to enlarge images?
I tried the built-in enlarger in Automatic1111, Topaz AI and PhotoMagicAI
The last one seems best so far

[solid prairie]

Hi guys, anyone else having problems with? fast_stable_diffusion_AUTOMATIC1111.ipynb
I have this error, does anyone know how to fix it?
ModuleNotFoundError: No module named 'fastapi'

[fervent thunder]

Seems you don't have a module fastapi. Try pip install fastapi

[solid prairie]

It is in the google colab link, how do I do that?

[fervent thunder]

Not sure about Google Colab. I use Automatic directly on my PC

[obsidian geode]

Same error as well, its probably some update with automatic and it broke everything. I hope they fix it soon

[wind egret]

do you guys share your prompts? i am super new to this and would like to know your inputs?

[obsidian geode]

I would suggest to do something like !git checkout 3715ece

but I dont know the command for a gdrive setup

[fervent thunder]

You can get inputs on civitai.com from sample images

[stable pawn]

A good way it to look at civitai.com for all images with a small (!) have a public prompt, but when you use them you hasll try remove thing like "best quality" and so to see what they do for the image, then you will learn and get your own way to make prompt. It not that hard and there is no seekrets

When you do a prompt, just write what you want in the image, do not add what quality you want, that is up to upscaling and post edit, like "landscape hills clouds flowers rural farm, rain", then add words.

[wind egret]

thank you for your useful feedback! I have been using PromptPerfect to make prompts, but I am not sure if the generated prompt there is too long or something. OR maybe I haven't found the perfect model to use yet.

[astral goblet]

"best quality" is actually one of those magic tokens and it comes from NovelAI. They did a lot of prework on their image set for training, and the ones with the highest aesthetic scores were given the tag "best quality". Any merge that uses the leaked NAI model in it will benefit greatly from that token. So basically ANY merge.

i dont think there are any merged models at this point that haven't incorporated the leaked NAI model at some point in their lineage

[stable pawn]

The problem with things as "best quality" is that best and quality is different from person to person and subject to subject, what is quality when we make an image?

[bitter mauve]

is it possible to use Control Net from the command line?

[astral goblet]

https://github.com/lllyasviel/ControlNet collection of original controlnet scripts are here. These are what the extension developer uses to create the webui version

they should all be command line scripts

[astral goblet]

You and me might look at "best quality" with different subjectiveness yes. But i had just explained that the image set NAI built for their model uses "best quality" in a very significant and particular way

[bitter mauve]

Thanks!

[stable pawn]

Do you have ant practical example for how "best quality" can improve the quality of a prompt?

[astral goblet]

Anything V3 sees a lot of success there. It's not a subjective belief. It's literally how they trained their model

[crimson veldt]

@thorny depot oh hey i see you are the mod of this server

can you tell me where can i post this ||https://rentry.org/niakonichan || so, many people can see it?
this is a rentry page created by a user @Niaka#8480

the page has all the big updates, tutorial etc for peoeple, this is also updated frequently
it even contains guides for newbies

i thought it would be beneficial if i shared it with the sd community but i didnt knew any other place than their official discord. therefore i ask the mods support in making this rentry reach the mass.
the user has no plans on dropping this project so it will be updated for a long time.

well

i hope you do something about this

[thorny depot]

Hiya. Just a sec....

[stable pawn]

They trained it on "best quality"? How could they decide what was "best" and what is "quality", I shall download that model and test then I just now have made test and experiments on this subject.

[astral goblet]

That guide reminds me of the high quality text guides you'd get out of "gamefaqs.com" back before clickait ad laden guides and youtube tutorials that don't get to any point for 30% of the video swept the scene and changed everything.

it's good to see real grass roots efforts that aren't manufactured around the concept of marketing

a ton of good guides are showing up on rentry lately

[astral goblet]

i told you. an aesthetic classifier did it

[crimson veldt]

this one also includes all guides and stuff in one place

[thorny depot]

So far as I'm aware, you're fine to post it here.

[astral goblet]

looks like a high quality resource

[crimson veldt]

ah thanks

btw i am asking for like maybe pinning this on a channel or something so it reaches many people.

[wild steppe]

As long as your link follows #✍🏼｜rules-and-tos, you're good to share it with the community here

[stable pawn]

So it was a routine that come up with what is "best" and what is "quality", but those words are subjective and can change between individuals.

[crimson veldt]

i see but i need further support

[wild steppe]

Please open a ticket in #1010934719455707218 & we can continue there!

[crimson veldt]

i dont know any bigger community sd related community

[stable pawn]

Hey! @astral goblet I looked up "Anything V3" and it is anime based, for me Anime can never be "best" or "quality"... sorry, that was a bit of trolling, but , I am not a big Anime fan.

[fervent thunder]

hie

what's the difference between train and dreambooth tabs in A1111?

[astral goblet]

that's fine, but it was just an example of how novelai's initial model has now merged into everything that is popular now a days. "best quality" may not have the effect that you want, and you might not think it has an effect, but every model that has NAI in it's merge lineage will know "best quality" in a particular way. Whether or not you agree with it being "best" isn't relevant at all. It's what the model knows and it has an effect towards the final image.

a lot of photo realism models merge in anything v3 or NAI and those tokens can have give an anime edge to images due to the training

[stable pawn]

I am testing just now to see how good it is on "best quality" and if the quality tags give better quality.

[astral goblet]

it's not that it makes it actual best quality, or good, or nicer looking somehow. The high aesthetic rating anime knowledge when "best quality" is used is the reason people use it. It's a specific gear that has been designed into the knowledge set

[stable pawn]

Well, if it work when used in a promtp we will be able to see it, so I doing a test

[obsidian geode]

Its a modifier, just like supreme, sublime

if best quality was taken as literal you would get A+ answer from the engine

however theres a limit to sd

[astral goblet]

you subjectively don't like anime so your biases will determine your test fails. your approach to this is all wrong and you're misunderstanding me. i'm not arguing with you in order to convince you to use it. i'm just elaborating on why people use it. The goal isn't to make an image "best quality" and instead it's geared towards high aesthetic rating anime images. It's good to understand why people use these tokens instead of just accepting them as literal language

[stable pawn]

@astral goblet If a model is trained to be able to do "best quality" according to some aesthetic anime rating, and the prompt work, we should see a difference when those tags are used and folk should agree on that "best quality " look best in an "aesthetic" way, if not we can that it do not work as expected.

[astral goblet]

When it comes to ai models and how they take data in and spit data out, throw away all expectations

[stable pawn]

Yes, I did not expert "best quality" to work .

[astral goblet]

keep in mind that the original model, NAI, trained it specifically for anime, but over time as it was merged into more and more models, those references got blurred and are more generalized. You'll see their biggest effect with closer to the source NAI models, like Av3

[astral goblet]

That's because your expectations of it are literal. I've explained this 100 different ways though. It's not trying to make things the "best quality" but instead thats just a token to reference the trained knowledge.

Maybe others can benefit from our back and forth but i feel like i'm just writing a post instead of having a conversation here. Gtg. have a great day.

[sudden terrace]

/start

[stable pawn]

I do keep that in mind, but as long as I can not see any difference when the tag "best quality" I have no evidence for that it work, and folk that swear this tag work will never show examples that can be repeated.

[astral goblet]

I often see that kind of language when people don't understand how ML works. "People will never show examples" ... ugh. You're demanding work out of people instead of trying to come to an understanding. I've explained this so many ways that it should be easy to grasp, but instead you're acting childish now and saying this kind of nonsense. Lame. Very lame. This wasn't intended to be an argument, but you took it here while i've been agreeing with you most of the time. "best quality" doesn't mean litereally "best quality" and searching for examples of that is going to be impossible, since you've admitted that you subjectively hate the results it's producing and disagree that it's "best", while that's entirely beside the point

Like i said, i hope other people can benefit from our back and forht. I feel like you're just here to manufacture an argument though and won't benefit at all.

Here's what you've done. You have admitted you hate anime and therefore any result from the tag you will hate. Then you demanded people show examples so that you may judge them, while already knowing that you will hate all the examples. You're leading people down a path of goalless arguing. Don't do that.

[stable pawn]

@astral goblet All I asked was if you can show it work so I can see it. But how shall I be able to know if it work if I do not know what I shall look for?

[astral goblet]

"i'm just asking questions" is one of the oldest troll games in the book. Oh, btw, you just lost the game

[stable pawn]

So you do the blaming game now?

[astral goblet]

"blame" is manufactured drama. you're still doing it. Try having conversations instead in the future

[stable pawn]

You said "best quality" work as a tag, why is it wrong to ask "can you show so I can learn"`?

@astral goblet can you do a summary what this "drama" is about and how it started?

[astral goblet]

backpeddling and rehashing events is dumb too. another old troll game. bro. stop being so obvious. It was more than that. I expalined it 100 ways since that initial question, while you were trying to argue every step of the way

there was no effort of trying to understand

Stop gaslighting

[stable pawn]

So you blame me to be a troll now? For we disagree over a tag and if it work as folk expect?

I not been rude to you @astral goblet , I not called your name or said your where wrong, what I did was to ask if you could help me see how this tag work.

[dusty tree]

is it possible to get viruses from civitai?

[obsidian geode]

there is a moderator team on the site, so we can suppose it wasn't hacked, however it is filled with porn and such sites have many viruses because of the people they attract

[dusty tree]

ok

[north meteor]

can someone teach me how to make nfsw made by ai

[surreal perch]

@north meteor

FAQ: What are the server rules?

The server rules can be viewed at any time in #✍🏼｜rules-and-tos

[dusty tree]

what's the conclusion, yes or no?

[stable pawn]

Not from the site, but there is a risk with some models, but the risky models are removed rather fast and old models then been downloaded often are safer, and there is SafeTensor that is seen as sequer.

[north meteor]

im not talking only about this ai

[dusty tree]

ok

[north meteor]

what is nai diffusion

[dusty tree]

is Cinematic Diffusion safe? it looks sick

https://civitai.com/models/3091/cinematic-diffusion

[wintry stream]

This sever does not permit NSFW content

[north meteor]

did i said im gonna send it here?

or what

[proud nova]

That's a topic on a thin ice

[stable pawn]

It been up long time and downloaded over 2000 times, I would see it as okay..

[north meteor]

huh?

[warm junco]

Just take an anime model type 1girl, and the stuff you want to see

[proud nova]

Meaning that it's quite sensitive. But yeah. As CS1o says

[north meteor]

u cant

[dusty tree]

how do i know if a model is a virus, is there a way to find it out?

[proud nova]

Who denied you to write stuff of this kind into prompt?

[warm junco]

You can

[north meteor]

stable diffusion rules says u cannot make nfsw content

[dusty tree]

not automatic 1111 though

just use another model

[obsidian geode]

if there is a download link select .safetensors file name other than than that look the reviews

[proud nova]

Exactly

[north meteor]

so what do i do

[dusty tree]

just look up a model that supports it

[north meteor]

none does

[dusty tree]

bruh, like half of the models in civitai are nsfw

[north meteor]

yo fr

?

[dusty tree]

yea

[north meteor]

send link

[proud nova]

Man just google it

[north meteor]

😈

[north meteor]

i did

found one but it sucks

[proud nova]

I don't feel comfortable talking about this stuff

[north meteor]

ok

[dusty tree]

https://civitai.com

[proud nova]

To show NSFW content and models you must be signed in.

[dusty tree]

yup

[proud nova]

That's all I'm gonna say

[north meteor]

how is that a problem

[dusty tree]

he's just saying that so you know

welp, i hope my computer isn't hacked lol

[north meteor]

wdym

[proud nova]

What's going on?

[dusty tree]

if models can have viruses

[north meteor]

they cant

[dusty tree]

´wdym

[proud nova]

Man models are the same like downloading unknown apps and running them

I am using CoLab so my laptop is safe ig

[dusty tree]

yea, most are safe tensor though

[stable pawn]

To ask how to make NSFW content with AI before you know what a checkpoint is, that is a bit like to ask how to drift with a semitruck before you can drive.

[proud nova]

Lol. That's true

[fervent thunder]

https://colab.research.google.com/github/sagiodev/stablediffusion_webui/blob/master/StableDiffusionUI_ngrok_sagiodev.ipynb#scrollTo=gJBtadxGOK7b

[quaint tapir]

Anyone using Dreamlike? And if so what sampler works best?

[dusty tree]

just go to civitai

[north meteor]

how to generate images there

[dusty tree]

download one of the models and put it in the models folder

[north meteor]

bruh

too hard blud

can someone teach me how to make nfsw made by ai

[dusty tree]

just press download latest

[north meteor]

huh

[dusty tree]

chose the model you want and press download latest

[quaint tapir]

Well tbh you gotta be good at creating prompts, and negative prompts, I’m a dummy so I use ChatGPT but that won’t make nsfw prompts

[dusty tree]

do you have automatic 1111?

[north meteor]

what is that

[dusty tree]

bruh

the program needed to make ai art...

[waxen swallow]

Everyone starts at the beginning

[dusty tree]

you can also use others

[quaint tapir]

Bro is lvl 1

[north meteor]

can someone teach me how to make nfsw made by ai

[dusty tree]

like invoke

[dusty tree]

i can

[north meteor]

?

[dusty tree]

but you'll need automatic 1111 to make the art

or invoke

[dusty tree]

you'll need to get git and python first

[fervent thunder]

Just go on https://civitai.com/ , look for a model you like , read through guidelines of this model, look how other people propmt and do same thing , but jut based on what you want, nsfw isn't that different from normal prompts , just subject is different

[north meteor]

bruhhh

never thought it can be that hard

[stable pawn]

in Img2Img, if you have that, there is often a function called "Interrogate CLIP" (I think), you can upload a image and then use that to get a AI based prompt suggestion.

[dusty tree]

https://www.python.org/downloads/release/python-3109/

https://git-scm.com/download/win

[north meteor]

what after

[dusty tree]

?

[north meteor]

when i dowload it

what next

[quaint tapir]

You can’t just put like “Miley Cyrus showing her (blank)” it’ll probably look whack

[dusty tree]

did you get them both

[north meteor]

no

but when i do

[dusty tree]

ok

[north meteor]

what they even for

[dusty tree]

to run automatic 1111

it runs on python

[fervent thunder]

it's not hard , it takes like 5 mins
https://github.com/AUTOMATIC1111/stable-diffusion-webui#installation-and-running

[dusty tree]

yes

depends on your specs though

[north meteor]

what it does

[dusty tree]

a program that you can make ai art with

[quaint tapir]

If he don’t have a good video card than run on Google Colab and there’s another one I can’t think of the name

[modern stirrup]

Is there no way to combine a textual embedding (->character) with a style LoRA??

[dusty tree]

btw what specs do you have

[north meteor]

?

idk

[dusty tree]

on your computer

[north meteor]

lol

[dusty tree]

is your pc good?

[north meteor]

not best

but kinda good

[modern stirrup]

Technically a1111 is a user interface 🤓

[obsidian geode]

You need nvdia gpu

[dusty tree]

okay

[quaint tapir]

Don’t you need like a $1000 video card?

[dusty tree]

do you have python and git now?

now

[north meteor]

yes

what next