🏙│ai-images

[bronze kayak]

https://tenor.com/view/sad-no-disappear-sad-noob-gif-24308846

[pale briar]

That looks so much better than the typical style

[brittle ocean]

sure

it will change now

[shadow bane]

"how many catgirls do You see in that picture?"

[bronze kayak]

https://tenor.com/view/catgirl-come-on-already-cat-gif-24598868

[shadow bane]

super broken, but the atmosphere is very nice xD

[shadow bane]

Sorry what’s the image/text in this image

[wet bramble]

:3

[shadow bane]

Oh yeah now I see it

It’s this

(The image in the TV)

[coral venture]

its my profile picture

[shadow bane]

O

Lol

I doubt anyone of us made an AI

I have made over 10k AI images... that would be a hard choice

[shell star]

gys what can i use for "imagine to ai imagine"

[bronze kayak]

these are all really nice! reminds me of Lynette from Genshin Impact

[shadow bane]

[bronze kayak]

even there where hair color changed she still has her little star on the cheek

[shadow bane]

yep, it normally wouldn't be strong enough, but in the last facedetailer/adetailer step I have the Lynette Lora on full power, so that is ensured

[calm dock]

nice bro how did you make it?

[shadow bane]

like this

[calm dock]

😭

[wet bramble]

What the fuu-

[shadow bane]

[autumn moat]

https://tenor.com/view/spiderman-crying-so-happy-i-cant-even-why-me-gif-19622708

[plucky crescent]

yo

[onyx relic]

Avion

#vion

Crea un avión

[calm dock]

thats so good bruhhhh

[scenic lodge]

studio shaft type head tilt

[calm dock]

no way thats ai dude

its getting too good

[tawny peak]

I'm still using SD 1.5 not even SDXL

[scenic lodge]

is sdxl even worth using?

[tawny peak]

I tried it and I was getting worse results from what I would get with 1.5

although I think it's because XL is still very underdeveloped

[shadow bane]

Just the two of them

[teal sphinx]

sdxl is not good for anime thats for sure

[pale briar]

At the risk of being labeled a heretic, not everything needs to be anime

[leaden oyster]

@pale briar has reached level 3.

[teal sphinx]

i've never said that

[shadow bane]

that's because most SD 1.5 models are based on the big leak from a professional Anime AI model, it was basically a walk in the park to make a SD 1.5 Anime model decent because of that

[shadow bane]

good SDXL models do just fine

[unborn moss]

@quaint tiger brotha someone wants to learn lora and stuff

[static cairn]

i did train a lora

i'm here to know how it can be better 🙂

[unborn moss]

probably better dataset

or database

@shadow bane is a god of sd too

[static cairn]

this is also what i think but if possible can i have a good dataset so i can know what is wrong in mine ?

[shadow bane]

Sadly didn't do training yet 😉
But most common things that can be done better:
tagging
imageset/dataset (not too much, not too few)
coherent vision for the model/lora (if the training end result didn't get your point, it will not work great)

[static cairn]

ya better tagging too

[shadow bane]

but that's superficial knowledge, the nobs and buttons of the apps/programs are out of my knowledge

[static cairn]

i use the dreambooth method to have a .safetensors at the end ..

[static cairn]

i do with colab/python so no UI

[shadow bane]

but it does still have variables right?

like steps/etc.

[static cairn]

ye

oh this ok mb

[shadow bane]

[quaint tiger]

I hand type image descriptions, which often gives me better results

[static cairn]

this is an exemple with vermeil from arknights

[quaint tiger]

Also train loras using Kohya-SS

[static cairn]

ok in koya there is a lot of method to do it

dreambooth is one of them

[quaint tiger]

You just use the lora tab

Dreambooth is a no go

[static cairn]

ok F

[quaint tiger]

Causes issues often

[static cairn]

so i need to find the diffusers to safetensors ...

[quaint tiger]

?

[static cairn]

since i don't use UI and have the other way of training a lora (with the diffusers) and i need to convert the .bin result to a .safetensors to use it everywhere

like a already have all of this

normal = diffusers
safetensors = dreambooth

[quaint tiger]

Odd, i just use webbased Kohya

[static cairn]

but when you don't need a UI

[leaden oyster]

@static cairn has reached level 2.

[quaint tiger]

Anyway

I personally would not recommend using Kohya_SS dreambooth tab

Shud deffo use lora tab

[static cairn]

so ok koya lora tab can do it so i will check the py file of that

[quaint tiger]

Aight

Now i gotta get back to work

XD

[static cairn]

i'm at work ... just scraping website is long

[shadow bane]

you go fluffer!

[quaint tiger]

Also

Fluffer is a word you might not wanna use

[shadow bane]

I meant more like flurrerl something something

the thing that you say in english when you see a fluffy animal

[static cairn]

oh i didn't found it first time because it's not named train_lora but train_network ...

[shadow bane]

ah yes AI is a wild west with many having 0 idea how to make coherent UX

[leaden oyster]

@shadow bane has reached level 17.

[shadow bane]

brilliant peeps all of them, but UX is a mess 😉

[static cairn]

also to tag images is it ok to use the deepdanbooru img2txt ??

[quaint tiger]

I hope you do realize what it actually means tho right?

[shadow bane]

yes the non cute version I can make an educated guess

[quaint tiger]

Bhahahahahaha

To those that dont know

You can just google

What is a fluffer

[static cairn]

an exemple
(i usually remove tags with : and () to avoid rating, character/anime tags)

[static cairn]

ok so let's go new colab file 🙂

[shadow bane]

https://tenor.com/view/aurora-borealis-steamed-hams-the-simpsons-simpsons-super-intendent-chalmers-gif-26950836

[shadow bane]

ok last chinese style image for now

[shadow bane]

would you guys help me in ai-art cuz im new and i dont know how to fix thing and make what i want

for example my prompt: white haired girl with a bloody katana in the assasin outfit,blurry_light_background
and my lora's:

and im using easy negative

[shadow bane]

weapons are horribly hard to do with AI atm, almost no models are trained for them, try to see if you can find a weapons/katana/etc. Lora or something to help with that
I sometimes use something like https://civitai.com/models/126751/xl-fantasy-warriors-by-hailoknight?modelVersionId=138640 to help with making weapons coherent

[ebon dome]

I swear there's something extra artifitial in my faces

the skin detail looks weird

like look at this beard, it's just a fuckin texture pattern lmao

[shadow bane]

[ebon dome]

maybe it's the VAE idk

[shadow bane]

are you using TI/Embeddings? Any Loras?

[ebon dome]

I tried offset lora on and off and also the CFG tonemapping on and off and nothing

they all produce this ugly pattern-like detail

[ebon dome]

I'm going back to my old images and these look 50x better

[shadow bane]

but what changed?

[ebon dome]

no idea, I messed around with my workflow too much

[shadow bane]

[ebon dome]

this one looks good though

Huh, I've brought up my denoising value again

[shadow bane]

I got weird patterns too when I used negative embeddings like
unaestheticXL_AYv1
unaestheticXLv13
they dipped to deeply into the noise

[ebon dome]

god dammit I don't get it

that beard looks fucking weird

[shadow bane]

any weights that are too high or anything like that?

[ebon dome]

significantly lowered CFG

still has that patterened look but it's softer now

[shadow bane]

does the pattern already appear in the base image or with high-res fix

[ebon dome]

both

actually lowering cfg made it look less worse in the base image

[shadow bane]

looks like it couldn't handle something in the instructions with that much guidance

so it either defaulted to patterns recognition, or was trying to do something it interpreted as a pattern command

how did you mention skin/beard in the prompts?

[ebon dome]

I did like skin detail in it and without or, sweating skin and shit

[shadow bane]

hm with sweating skin I could see how it would try to get droplets or something

[ebon dome]

lmao

this looks a bit better

[shadow bane]

[quaint tiger]

the word Dew for prompting might work

to get droplets

[shadow bane]

oh I know I tried lots of things when I tried to do that prompt challenge, sadly some models/seeds were not really intent on using (morning) dew and co on the fruit, so I tried droplets in different ways and got this funny pattern

[shadow bane]

DarkPizzaXLOrigin being randomly amazing yet again

[quaint tiger]

impressive stuff khazar

i really cant do stuff like that in a1111 i think

comfy ui is just too mind boggling for me lol

[uncut slate]

u'll get used it

[quaint tiger]

no ty

XD

[shadow bane]

[quaint tiger]

[uncut slate]

[forest ocean]

"want to see my head come off?"

[teal sphinx]

Made a new lora 😄

[calm dock]

make me images

now

[teal sphinx]

i will upload later more images to the cititAI page if you want

[calm dock]

the huh

ill just

save your ko-fi page

for later

[waxen hound]

It actually looks like a freddy, not a random teddy bear

[teal sphinx]

https://civitai.com/models/178284?modelVersionId=200143
it will be here in about a hour

[dark hollow]

martin gangsta king

[leaden oyster]

@dark hollow has reached level 1.

[scenic lodge]

https://media.discordapp.net/attachments/1165047354441879642/1167249122932314272/output.png?ex=654d7075&is=653afb75&hm=fe5c85a1e929fd149ea3cf43349a67c6be3d1fb44c16268bd78a6807c8f750fc&

[rustic raft]

https://canary.discord.com/channels/1159260121998827560/1160633319537520700

https://canary.discord.com/channels/1159260121998827560/1160633319537520700

https://canary.discord.com/channels/1159260121998827560/1160633319537520700

https://canary.discord.com/channels/1159260121998827560/1160633319537520700

the best channel

[static cairn]

man is having fun with # xdd

[unborn moss]

@uncut slate

Do this guy a favour

charkkkk

[brittle ocean]

-snipe

[unborn moss]

He deleted it

[brittle ocean]

ah

[unborn moss]

@uncut slate come fast bro

[uncut slate]

im here

[lucid lake]

🐀

[unborn moss]

charkkk said asian

N word

[lucid lake]

🐀

[uncut slate]

yeah i can see logs

[brittle ocean]

[shadow bane]

you don't need to repeat it, deleted is not deleted on server

[unborn moss]

@lucid lake Lil bud tryna escape

[unborn moss]

ik

I was just saying

[unborn moss]

I been a mod before in this server

[lucid lake]

get a job dude

[unborn moss]

I have it mate

[brittle ocean]

[unborn moss]

I have set my status to offline so I can see what someone says

[brittle ocean]

wydm?

[unborn moss]

I mean if your on phone

[brittle ocean]

wdym?*

[unborn moss]

You can see messages

[uncut slate]

lil bro ur the one being racist ya know

[lucid lake]

old discord mod ? 🤡 🫵 yeah i see now this is your job ig

[unborn moss]

I like your judgement skills

[uncut slate]

lil bro is so hopeless lmao

offed them

[unborn moss]

People getting way too comfortable nowadays

[uncut slate]

sounds like their problem

[unborn moss]

Just in general

Take Instagram for example

Filled with garbage

[uncut slate]

always has been lmap

[unborn moss]

Every social media is

[shadow bane]

too much offerings in general, people care less and less, cause their minds already overfilled with offerings
that's one of the reasons for the empathy crisis

[unborn moss]

True

[uncut slate]

[shadow bane]

@pale briar 24 hour catgirl tax!

[brittle ocean]

paris got beacon

[brittle ocean]

what checkpoint or lora did u used?

[shadow bane]

it's sadly an old model, but I really like it, one of my favorites from my early Newb days:
https://civitai.com/models/12903/gufeng

[brittle ocean]

line looks soo good

[shadow bane]

the creator updated his 2.5d model Guofeng to XL and even released a new version yesterday, so I'm hoping for an update at some point

[shadow bane]

I know right?

[brittle ocean]

[bleak dock]

Spiderman on 3d+2d

[shadow bane]

go online and make it! hell yeah let's goooooooo!

[static cairn]

is it good ??
first lora trained with caption.
~2h20
~14k steps
~200 image to train

it was for nian a character from arknights

[shadow bane]

picture is a little bit busy, so hard to tell if the Lora did it's job with the character, but looks cool 😉

[teal sphinx]

the composition is definitely good, but looks like the character lacks detail
whats the resolution and epochs trained? and repeats?

[shadow bane]

try to make a bland background with focus on the character, see what the results are

[bronze kayak]

could I use one of these as my new pfp please

[teal sphinx]

yeah ofc
check this, i uploaded more images in there
https://civitai.com/models/178284?modelVersionId=200143

[bronze kayak]

LMAO

[teal sphinx]

welp

[bronze kayak]

but thanks a lot

[shadow bane]

CivtAI burning atm

[teal sphinx]

the usual stuff

cant blame them tho

[shadow bane]

they probably get like 10k new users every day if not more

[teal sphinx]

its more like the unlimited unoptimized merges a lot of people make that takes so much storage and bandwidth

[static cairn]

512x512
10 epochs
and 10 repeat
total 15.6k steps

i should go to sdxl but i can't train it on colab ... there is not enough vram

[teal sphinx]

if you are able to train on 768,768 it would be much better, and make more epochs
also, use tensorboard to see if its not overtrained and whats the potimal weight

[teal sphinx]

well, if you want to train an anime character like above then sdxl wont make it better currently

[static cairn]

oh ... a lot of people said sdxl was better than normal sd because more resolution ...

[static cairn]

20 epochs ?
and the repeat are "good" ?

[shadow bane]

if you don't have the means for it, then it's just that, just use what you can use, SD1.5 is still the biggest userbase, and you can accomplish some higher resolutions with it than 512x

[teal sphinx]

repeats probably should be between 10-30, i would choose 20 usually
you train the amount of epochs until you see in tensorboard that it starts overtraining, and stops there, but for 200 images, you probably need much more than 10 epochs

[static cairn]

so i will use 768px image training + i will bring the epoch to 20 and see.
if there is a change i will also put the repeats to 20

tho you will see it tomorrow because it was 2h20 so with a x2 in the epoch it will be a looong time

[leaden oyster]

Ayo? @static cairn level 3 !!!

[shadow bane]

wish I could send you some 4090 power, but sadly not so easy

[teal sphinx]

how much vram do you have?

[static cairn]

i have a T4 with google

so 15GB vram

and 13gb RAM

[teal sphinx]

btw, i have them all upscaled if you are one of those who prefer high resolutions too 😄

[teal sphinx]

oh so you shouldnt have problems changing to 768

[bronze kayak]

I mean I don't know if that would make a big difference since I'm using it as pfp

[teal sphinx]

yeah np, whatever is good for you

[static cairn]

i was a 6/7GB with current training (because half precision)

i can't do full precision because google ..

[teal sphinx]

oh really? why is that? cuz of vram or ram?

[static cairn]

vram it need a little bit more

but once i have something good i might move it to runpod and train loras in serverless (with a http api call)

[teal sphinx]

vram is the problem? i train fp and save fp with 768 res and i have 11 gb vram

[static cairn]

fp16 or 32 ?

[teal sphinx]

oh i use the kohya and it didnt show me 32 precision option

[static cairn]

oh i do fp16 too

[teal sphinx]

i dont really know what bf16 is but i think fp16 is better probably

so i do that

[shadow bane]

https://www.linkedin.com/posts/furkangozukara_what-is-the-difference-between-fp16-and-bf16-activity-7095150263602229248-Ybup#:~:text=In this case%2C BF16 retains,more calculations per unit time.

[static cairn]

oh i just got a shit idea ...
i wonder if i can use accelerate to cluster 2 notebook ...

[shadow bane]

[static cairn]

should i try ?? to change it to bf ?

[shadow bane]

try right? not like it will hurt ya 😉

[teal sphinx]

welp, probably will try bf16 next time, tho it might show errors

[static cairn]

since it's a notebook i will be back in 15m when everything downloaded, installed, configured, to tell you how much time it will take xdd

maybe a little more i forgor to remove the auto tagging ... it will tag again the 200 pics noice

[shadow bane]

[static cairn]

also with 500/600$ what gpu should i take for AI

bruh

i don't want to deal with it now so pf16 is better for now

[tawny peak]

For that price probably a 30 series with at least 12GB of Vram

[static cairn]

i was thinking a 3060

[wet bramble]

That's definitely good enough

[static cairn]

i will also need to buy ram because one died not long time ago (from 32 to 24 gb)

[shadow bane]

a 3060 12gb is like 320-360euro here, what's with the other 200 bucks?

oh wow a quad channel ram user, that's a rare sight

[brittle ocean]

this checkpoint is crazy

[static cairn]

OOF

[leaden oyster]

Ayo? @static cairn level 4 !!!

[shadow bane]

[static cairn]

i wanted to go with 20 epoch but it will be more than 24h xdd so i can't due to colab

i can try with a smaller dataset tho

[teal sphinx]

that should totally be okay, as 200 is usually for a very detailed style or base model
for characters i think between 15 to 70 images should be optimal

[static cairn]

i don't want a base model tho xdd

i will reduce to like 100/120 images

[teal sphinx]

even less would be okay lol
the lora you see above is pretty flexible it trained on 20 images, just make sure you pick the best images the has the most features for the characters and different angles, poses, and are of a high resolution

[static cairn]

downloaded from safebooru so i have a lot of big ones

i made a downloader for it so it's easier to have images 🙂

[teal sphinx]

also, u probably did that already, but set the threshold for the captions for about 0.4 and add a prefix as the name of the character... that should help with the capacity of the model

[static cairn]

i did put a tag for the character that stay on all images and it's on 0.35 🙂

[tawny peak]

I have quad channel ram as well

[shadow bane]

[tawny peak]

And I have another computer that has Octo-Channel ram with 64GB of ram, because it's a server

[shadow bane]

well my gaming circle uses dual channel for at least 1 decade now, so its quite unusual for me
just recently I bought 2x16gb RAM sticks

[tawny peak]

Yeah, having more than 2 channels doesn't increase performance that much, it just puts less strain on the ram overtime

Hence why the server has 8 Memory slots, so it can run much longer

[shadow bane]

aye server is a totally different beast

most gaming CPUs are Dual-Channel optimized nowadays(not that I can remember for how long that has been the case), so that's kind of been the go-to thing for us

[tawny peak]

I believe it was around when DDR3 ram came out that they started optimizing it for dual channel, because up till then most budget computers were single stick systems with slow cpus and slow hard drives

[shadow bane]

yeah that sounds about right

[tawny peak]

Tried out your model, I like it, but it is a bit over-trained

[mystic garden]

which artificial intelligence?

[mystic garden]

bing ai?

[normal cradle]

Yep

[mystic garden]

Prompt?

[normal cradle]

Disney Pixar Movie Poster featuring an old grandma with the movie titled “times up!”

[mystic garden]

😁❣️

[ebon dome]

explicit??? 💀

hell nah

it's just mr bean with a hood on man

maybe if it's smaller

okay

here's the SSD-1B model

the distilled SDXL model

it's like SLIGHTLY faster

not SSD anymore

[leaden oyster]

Ayo? @ebon dome level 22 !!!

[ebon dome]

okay

okay I think I fell in love with low CFG again

[shadow bane]

The finger

[brittle ocean]

[shadow bane]

[tawny peak]

tried out bing ai, and it gets pretty good results with anime styles... when it doesn't block 90% of all attempts

[harsh ice]

someone make gore art with the ai, never saw one of these

[tawny peak]

uhh, try making it yourself, don't post it here because i'm fairly certain that's against the rules here

[scenic lodge]

not too bad

[harsh ice]

gtx 1650 would take centuries to do that

[scenic lodge]

what was the prompt

[harsh ice]

3 arms

[leaden oyster]

Ayo? @harsh ice level 1 !!!

[harsh ice]

[tawny peak]

Masterpiece, best quality, perfect lighting, 1girl, solo, dark brown hair, medium hair, puffy hair, messy hair, fluffy hair, amber eyes, ahoge, looking at viewer, techwear jacket, black gloves, black utility pants, utility backpack, goggles on head, utility belt, facing viewer, anime, sunet, mountain scene, snowing

[leaden oyster]

Ayo? @tawny peak level 5 !!!

[tawny peak]

The same as what you'd use with automatic1111

[shadow bane]

I didn’t realize at first but I made a bird Pokémon flip the bird gesture

Literal genius right here

[harsh ice]

bro thought saying masterpiece, best quality, perfect quality would make it good

[shadow bane]

That’s what they all do lol

[tawny peak]

it does have an effect with automatic1111, I just copy pasted the prompt

[harsh ice]

literally telling the machine to do it right fr fr

[scenic lodge]

surprised it wouldnt even be blocked off lmao

[shadow bane]

Mhm

[harsh ice]

like you have to tell it or the lighting will look bad

??

[tawny peak]

Yea, not with newer SD versions like SDXL

[harsh ice]

ai prompts are like r34 tags fr

🐀

[tawny peak]

Pretty much

[calm dock]

what the helllllll

[slate yarrow]

i mean thats why half of the results get blocked

[harsh ice]

yes

[tawny peak]

Content blocking even if your prompt is something 100% SFW is the main reason why Dalle-3 Sucks right now

[slate yarrow]

yup, everytime i have to emphasize the creation is male to get through

[tawny peak]

Huh, weird

Hopefully when Dalle-3 gets released under OpenAI's website it isn't completely butchered

[slate yarrow]

hopefully

i mean they were so confident to give bing the opportunity

[tawny peak]

It'd be cool to see what generative fill could do with this version because Dalle-2 was kinda meh

[slate yarrow]

generative fill is becoming paid

[tawny peak]

Well Dalle-3 is paid only right now unless you use bing, or I guess you could use Google which uses Imagen

[slate yarrow]

in photoshop they are planning to make it paid too

[tawny peak]

damn, photoshop is already expensive enough

[slate yarrow]

right. I wish to get my hands on premiere tho

[tawny peak]

pirating's an option

[slate yarrow]

ayo

[tawny peak]

nah i'm joking, i use open source software for most things

[slate yarrow]

true

Man I hate using Capcut

bad software for actula editing

but good for tiktoks or shorts edits

[tawny peak]

Kdenlive is what I use, It has a bit of a learning curve, but is pretty powerful

And it's 100% free

[slate yarrow]

nice

[leaden oyster]

Ayo? @slate yarrow level 3 !!!

[slate yarrow]

i thought capcut is the only one who's free

but not for mac tho..

[tawny peak]

I think you can get kdenlive on mac

[slate yarrow]

i'll try to

[tawny peak]

yeah it supports intel and m1 macs

[slate yarrow]

yooo i got it

nice

kinda hard to use

I'll try to master it

I gotta go now bye

[tawny peak]

definitely a different experience from capcut for sure

[shut cliff]

The woman face 💀

[tardy aspen]

The Lego movie

[foggy gorge]

Sonic

[frail pecan]

suddenly wanna see how the bing ai handles making a disney/pixar movie poster about my favorite game (Cookie Run)

it doesn't like me

wait

why can't I paste an image

https://media.discordapp.net/attachments/900802028450770955/1164027918758977576/image0.gif?ex=6541b87a&is=652f437a&hm=650d83f7899f796254716c27c8f942d18dcf17c87748f816dd83a7b4915ac0fa&

WAIT I DONT HAVE IMAGE PERMS?

screw whoever got the old server deleted I have to work to get to level 5 again

[leaden oyster]

Ayo? @frail pecan level 1 !!!

[frail pecan]

/nm

oh

wait now I have image perms lmao

[frail pecan]

anyways

[serene matrix]

https://www.youtube.com/watch?v=7jgVgCI-R8k

(Dall-E 3 image aside my music.)

[uncut robin]

So close...

[slate yarrow]

that's freaking lego

[wicked scroll]

[shadow bane]

blondie

[autumn moat]

its fortnite

[tawny peak]

nice

[shadow bane]

Snoooooooowwwwww

[brittle ocean]

this is UI difference

[normal cradle]

That’s the point

[shadow bane]

well and SDXL already starting with 1024, so more easy to stuff detail inside

or in this case 1344x768

[quaint tiger]

so did you make sumthin for spooky? owo

[shadow bane]

nope but today I have secret mission from squishy

[quaint tiger]

oh i know

im the one who recommended you :3

if its about what im guessing

uwu

[shadow bane]

indeed

[quaint tiger]

awesome

[tawny peak]

Wow

[brittle ocean]

at the very end😭

[shadow bane]

Updated

[ebon dome]

SDXL realyl likes this hand pose

it's so easy for it to get it right lol

MR BEAN WHAT THE HELL DID YOU JUST DO

[static cairn]

ummm i tried to use a civitai lora with the diffusers and ummm it's bad ... on the webui it works fine but not with the diffusers

this is the model/lora loading

gen = StableDiffusionPipeline.from_single_file(
    "./model.safetensors",
    torch_dtype=torch.float16
).to("cuda")

gen.scheduler = EulerAncestralDiscreteScheduler.from_config(gen.scheduler.config)

gen.safety_checker = None
gen.requires_safety_checker = False

gen.load_lora_weights(
    "./lora.safetensors"
)

did a forgot something ??

a civitai base model alone work

model without lora give that

[quaint tiger]

im sorry i have no clue tbh, im only good with webui specifically

[brittle ocean]

will there be support for amd gpus on windows in the comfy ui?

[uncut robin]

Now if only they were actually sitting in the seats lmao

[normal cradle]

This one came out meh

[shadow bane]

wow that's a busy broken background

[normal cradle]

SDXL

IMPOSTOR?

[quaint tiger]

1080p Background

[leaden oyster]

Ayo? @quaint tiger level 30 !!!

[shadow bane]

wow rare coherent gun holding

[quaint tiger]

1440p

[wet bramble]

Me when:

[hard cloud]

how do i make ai art

[shadow bane]

this pic goes so hard

[karmic ether]

good god it's hard to tell that's ai

[leaden oyster]

Ayo? @karmic ether level 1 !!!

[hard cloud]

but how do i make art

[leaden oyster]

Ayo? @hard cloud level 1 !!!

[karmic ether]

wdym

[ebon dome]

what type of art

[hard cloud]

how do i make ai art

[ebon dome]

by using stable diffusion

or midjourney

or dalle

[hard cloud]

free?

[karmic ether]

sd and bing dalle is

midjourney isn't

https://www.bing.com/create

that's free

[hard cloud]

thank you

[karmic ether]

you could also use stable diffusion web but it's a bit complicated

https://stable-diffusion-art.com/automatic1111/

[shadow bane]

What is his phd?

[normal cradle]

That looks crazy good

[ebon dome]

not my prompt 🔥

[shadow bane]

i can imagine that cat accosting a mentally ill man in a metro

[rustic raft]

@quaint tiger this shit hits hard as my lockscreen

absolute poggers

[rustic raft]

how'd u even make that?

loos like i could use a similar method to gen* myself some discord emotes

[quaint tiger]

its a style lora

[rustic raft]

omg

[rustic raft]

u got the lora?

[rustic raft]

[plain acorn]

imagine an LSD Dream Emulator Remake.

[leaden oyster]

Ayo? @plain acorn level 3 !!!

[magic fractal]

dream?!?!?! frome minceraft?!?!

[plain acorn]

haha

[hoary garnet]

mr bean in anime

[hoary garnet]

yall teach me how to use this

[raven rivet]

tutorial please

[thick thistle]

/imagine

[slate yarrow]

prompt?

[rustic raft]

idk ask Yume

[sudden vale]

What do you use for the ai images?

[shadow bane]

We need a pin for all the basic options for creating AI

[brittle ocean]

Or just make tutorial all in one with comfy , a1111 etc

[sudden vale]

Can anyone help me out with a photo

I'm trying to make a image and it's for a really dark story in a role-play we had
I can't just make it work I always get wierd stuff

I'm trying to get an image of a prisoner hanged from a castle balcony

But none of the generators give me good resoults

[brittle ocean]

Try to make it with Dalle

[sudden vale]

Ok lemme try

[ebon dome]

a prisoner "hanged"?

I'm not so sure that dalle will work with that

[brittle ocean]

It should work

[ebon dome]

oh damn

wouldn't have expected that

[sudden vale]

Yea like by the nack execution style

[sudden vale]

Can't use dalle
On open ai I need to confirm myself and their services are banned in my country

Anyways here is mickey mouse as a war general

[teal sphinx]

new lora 😄

[ebon dome]

I'm trying to do that with fucking controlnet but not even that works, it just keeps hallucinating

[sudden vale]

Alr tell me if u make any progress

[leaden oyster]

Ayo? @sudden vale level 3 !!!

[ebon dome]

nah I just gave up

I've made 50 attempts

fuck this

SDXL doesn't even know what the fuck a guillotine is

[bronze kayak]

Topaz

very nice

[teal sphinx]

Topass

[bronze kayak]

I'm kinda hesitating to pull for her

[leaden oyster]

Ayo? @teal sphinx level 11 !!!

[teal sphinx]

got E3 for the ice guy 😦

lost the 50/50 again

[brittle ocean]

let me try

[sudden vale]

Ok thanks

[sudden vale]

Lol thanks for the help

But I meant hanged by the neck

Like an execution

And wtf is that goofy ahh cartoon prisoner

🤣🤣🤣

[brittle ocean]

ah so this is not possible

[sudden vale]

Yea thanks for trying tho
Maybe I can edit that in myself

[brittle ocean]

try to make a more advanced prompt in SD

[ebon dome]

Don't even try with sdxl, no prompt will get you there, not even controlnet

It keeps doing hanging by hand or something

The only dalle 3 alternative we have is Deepfloyd IF

But that's hard to run and there's a lack of spaces, the official one is paused on huggingface

[brittle ocean]

i think it would be possible but you would have to create a fairly precise prompt

or just make lora

[quaint tiger]

:D

[naive sand]

wow

[leaden oyster]

Ayo? @naive sand level 1 !!!

[uncut slate]

hai yume

[quaint tiger]

hoi~~

[unborn moss]

Yume - lora god

Orhay1 - lora god

[uncut slate]

[quaint tiger]

pfff im not that much of a lora god XD

[uncut slate]

not with that attitude

[quaint tiger]

good

i now hand over my lora god status

[uncut slate]

to whom

[quaint tiger]

i shall never make a lora again!

[uncut slate]

why

[quaint tiger]

to the depths of hell >u>

[quaint tiger]

im just jokin hun

XD

[uncut slate]

i like it when u joke

[quaint tiger]

pfff i cant use stickers here ?!

FINE

https://media.discordapp.net/attachments/484048868674109452/1167941875240550600/YumeSmile.gif?ex=654ff5a2&is=653d80a2&hm=557cc3367154f7ded51029fa42450a84a4192676e11b1dc1fad68aedfa0f1d48&

ill do it like that then =u=

[uncut slate]

what?

[quaint tiger]

i cant use stickers in here

only the servers stickers

[uncut slate]

i dont even know

maybe a perm thingy i'll look into it

[quaint tiger]

i think so

[unborn moss]

Invite me

[brittle ocean]

i can use normally

[quaint tiger]

what

so boosters can?

[unborn moss]

Oh yeah

I can't use it too

[brittle ocean]

idk

[quaint tiger]

yea so boosters can use stickers

[uncut slate]

im not a booster

[quaint tiger]

ur mod

lol

[uncut slate]

ohh

[unborn moss]

[brittle ocean]

[uncut slate]

well boost al hub

[quaint tiger]

@odd lance i want my stickers perms ;-;

[odd lance]

wait wdym

[leaden oyster]

Ayo? @odd lance level 48 !!!

[uncut slate]

YEAH @odd lance NOW

[odd lance]

uh

[quaint tiger]

not allowed to use stickers outside server

[uncut slate]

booster perk it seems like

[odd lance]

try now

[odd lance]

hell no, i think it was a bug

[uncut slate]

[quaint tiger]

still cant

[uncut slate]

restart ur discord.

cache is bad

[odd lance]

tbh it can also be that his server lost the boost

[quaint tiger]

i did

[unborn moss]

Works for me

[quaint tiger]

but i can use them in verified

[unborn moss]

Works for me now

Oh nvm

It doesn't work

[quaint tiger]

.3.

[odd lance]

i dont get it

[odd lance]

theres literally no perms that is blocking this

[unborn moss]

Something to do with trusted member role it seems

[bronze kayak]

lemme look into it rq

[odd lance]

yes antasma do

thank

and good catch checkmate

[bronze kayak]

send a sticker in #✦│chat just to test

[unborn moss]

😈

[quaint tiger]

<w<

[bronze kayak]

I think I found it

try sending the same sticker in #🔥│model-maker-chat

or #🧬│ai-chat doesn't matter

[unborn moss]

Oh

[bronze kayak]

alright figured out

[unborn moss]

Model maker chat doesn't allow

[quaint tiger]

no access channel?

[bronze kayak]

oh

[leaden oyster]

Ayo? @bronze kayak level 38 !!!

[bronze kayak]

model makers chat

thought you were one :/

[quaint tiger]

XD

nope

:P

[unborn moss]

I can access it but not send

[quaint tiger]

:x

[unborn moss]

General chat also don't work

[bronze kayak]

@odd lance issue found: sending external stickers isn't granted anywhere for anyone except in #✦│chat for some reason

that's the only chat which allows it

[normal cradle]

https://tenor.com/view/who-pinged-me-ping-discord-up-opening-door-gif-20065356

[bronze kayak]

nobody

[unborn moss]

Actually thought of this, since when verified chat started the perms but I thought its not it

[normal cradle]

SD or Dalle-3

[bronze kayak]

@quaint tiger @unborn moss should be fixed

[unborn moss]

🙏

[bronze kayak]

added the perm for trusted members only

[unborn moss]

Thanks atmasn

[unborn moss]

Let's not clog this chat

[uncut slate]

sticking out ur gyat for the rizzler

[quaint tiger]

thanks

[uncut slate]

ur so skibidi

[brittle ocean]

[shadow bane]

Forgot to show this off last night but I generated most of this with AI and touched it up a bit

(This is for the credits of a collab I’m in btw)

[naive sand]

Masterpiece

[unborn moss]

@naive sand ur name is so familiar

Kya tum bhartiya ho

[naive sand]

yuppp

[unborn moss]

Oh that's why

Same here

[naive sand]

apka neme kya h

[unborn moss]

Don't talk here, this is for the ai image art stuff

[naive sand]

oops

[shadow bane]

catgirl tax

[ebon dome]

that seems way more likely

[shadow bane]

He seems chill

[ebon dome]

True

[shadow bane]

@brittle ocean btw, I got the GuFeng creator interested in a SDXL version for GuFeng

[sage jungle]

is there any way to use ai to turn a 1024x1024 image into a 16:9 aspect ratio image

[shadow bane]

so add width basically?

[sage jungle]

yes

[brittle ocean]

this model is great

[ebon dome]

felt cute, might delete later

[shadow bane]

Half life real

How’d you do this

[hybrid ingot]

half life but based

[ebon dome]

by doing it 🔥

https://civitai.com/models/135240/black-mesa-half-life-1 this is the lora btw

SDXL though

just saying it cause you guys use 1.5 iirc

https://civitai.com/models/58129/that-one-tf2-map-you-played-on-a-community-server-a-while-back holy fuck I love these TF2 map loras

the last 2 images with dr house were downscaled using nearest-neighbour to give it a aliased look

[tender aspen]

cr7 le este chupando el pie a messi

[shadow bane]

I really didn't need to know about your fetish

[shadow bane]

[ebon dome]

I do the same looking at all the SD loras

I wanna use those tf2 loras so badly with SDXL

I might train one on TF2 maps using google colab

Already made the dataset images for it

[shadow bane]

You can do that rn

[flint shell]

Hey I am new and what do you use to make that if you dont mind me asking

[brittle ocean]

automatic1111 or comfy ui

but comfy is more advanced but easy (at least for me)

[flint shell]

What would you say is good for a new user of ai art such as myself

[brittle ocean]

depends how u want to make it

spagetti vs prompt

left comfy ui right automatic1111

[flint shell]

How do i get the one on the right

@brittle ocean the left looks scary to use tbh

[ebon dome]

just tried the colab, the lora I made doesn't work lol

I might try again cause I am making custom captions rn

but still, not even the style itself works at all

it makes it looks like fortnite or some shit

the fuck is this

compare this to the half life 1 lora for SDXL, this one's amazing

holy fuck this guy looks dashin, even with those ridiculous ties

[brittle ocean]

Search automatic1111 on google

[iron oasis]

fortnite

[hearty cipher]

Where do yall make your AI ?

[naive harbor]

I just used stable diffusion automatic 1111.

[unreal scarab]

Asriel drip

My honest reaction

[shut mural]

he in Undertale

[lost rampart]

check this out

its a burger with no honey mustard

thoughts?

[ebon dome]

I'm devoid of thoughts because this burger looks too damn delicious and perfect

[uncut drift]

Are we going to eat burgers?

[unreal scarab]

hamburfer

[leaden oyster]

Ayo? @dire cobalt level 3 !!!

[subtle dock]

HOW DO YOU MAKE ONE OF THEESE?!!

[leaden oyster]

Ayo? @subtle dock level 1 !!!

[normal cradle]

Dall e 3 / Bing Image creator

[ebon dome]

by downloading the lora and using it

watch tutorials

[rustic raft]

https://discord.com/channels/1159260121998827560/1165042510142390404 ends <t:1698796800:R>

https://discord.com/channels/1159260121998827560/1165042510142390404 ends <t:1698796800:R>

https://discord.com/channels/1159260121998827560/1165042510142390404 ends <t:1698796800:R>

[uncut robin]

The one time I ask for a Coma-like Billie, you give me him having a good time - God Dam it Stable .-.

[ebon dome]

Ok 👍

the TF2 SDXL Lora is getting better

it still looks like shit though but it resembles the style lmfao

[neat nexus]

Got bored
Made goobers

[ebon dome]

holy fuck, real

[normal cradle]

New tf2 map looks sick

[uncut robin]

wow!

Can We port TF2 To PS1 guys?

[bronze kayak]

https://tenor.com/view/explosion-anime-gif-11253462

[robust hollow]

Venom AI Hub Chan

[hybrid ingot]

1

[ebon dome]

I trained it for 10 more epochs, needs even more imo

[normal cradle]

New tf2 map looks sick

[ebon dome]

but we're getting somewhere

[neat nexus]

Not quite TF2 tbh

[tawny peak]

tf2 source 2

[quaint tiger]

w>

<w<

:D

[uncut slate]

keep posting a lil more

[tawny peak]

wow it actually just looks like tf2 if anti aliasing was turned off

[ebon dome]

yeah I scaled down the image using nearest image scaling

to give it that look

[lusty phoenix]

holy shit thats actually nutty tho

[heavy tulip]

of fucking course it's a plane flying through two towers, because we needed 9/11 here to make it the comico 10/10

[unborn moss]

That is really cool

[uncut robin]

Now which song to cover for Halloween mmhmm...

Note: It must contain a lot of Darkness

[cold idol]

hi

[quaint tiger]

Hoi

[unborn moss]

That's coolio

What did you use

[shadow bane]

Bing

[uncut robin]

Dalle 3 looks to be the superior generator

sad to see its close sourced atm 😦

[unborn moss]

True i hate that they made it so restricted

Literally everything is copyright according to them now

One of the many reasons why Microsoft lacks behind

[robust hollow]

Some other Ai Hub Chan Venom

[fresh zodiac]

Al Chan is gonna eat all pizzas

[robust hollow]

Half Symbiote Ai Hub Chan

Ai Venom Hub

Last

[quaint tiger]

^w^

[quaint tiger]

Pretty o3o

[brittle ocean]

[normal crystal]

Chat where is the loras

[quaint tiger]

Theres isnt many loras released on this discord

But would be #1159496755218501692 if any

[neat nexus]

Happy Halloween

[quaint tiger]

honestly

i wanna see if i can generate stuff in the style of no game no life

[neat nexus]

Spuki

[leaden oyster]

Ayo? @neat nexus level 3 !!!

[quaint tiger]

@shadow bane did you submit something or? O_O

[normal cradle]

Did you train it more?

Looks good

What would happen if you try to generate a character?

[neat nexus]

I'm bored af

[fresh zodiac]

Kawaiii

[dire cobalt]

omg its the dude from fortnite

[fresh zodiac]

Dudeee

[ebon dome]

Yes, 10 more epochs, no.I didn't train it on characters though

[rocky knoll]

Guys, help me, ik what im gonna say is tracing, plagiarism, or anything you guys want to call it, but help me, i want to make pictures with any manga artstyle, is there any prompt to make it happen?

[brittle ocean]

download lora

https://civitai.com/models/16014/anime-lineart-manga-like-style

[rocky knoll]

do i need to download lora then the link or i can just download the link itself?

[brittle ocean]

u need to download it

file

[rocky knoll]

you mean the former or the latter one? please make it simple

[brittle ocean]

just click this

[rocky knoll]

Ok thanks

[rocky knoll]

and then what?

[brittle ocean]

u use automatic1111 right?

[rocky knoll]

i use microsoft bing ai image

[brittle ocean]

ah so nvm u cant use it on bing

just add to prompt lineart, monochrome

[rocky knoll]

can i use specific manga artstyle if i use this prompt?

[brittle ocean]

idk

[brittle ocean]

i got something likethis

[rocky knoll]

what was the prompt?

[brittle ocean]

school uniform, extremely detailed 8K wallpaper, masterpiece,best quality,1girl,solo,nice hand,good hand,lineart, monochrome

[rocky knoll]

ok thanks

[leaden oyster]

Ayo? @rocky knoll level 2 !!!

[surreal quarry]

Lmqi

[worthy dragon]

sci-fi city

[shadow bane]

Wise words

[slow vortex]

car

t

[robust hollow]

AI Hub Chan venom part 10000