#🏙│ai-images
1 messages · Page 10 of 1
@strong cosmos this u?
@autumn glacier this u?
real
truuue
(Amplify) This u?
Whoops I accidentally deleted the kalomaze one
(Ilaria)
lol
aint no way
Yes way
anime
That’s not Kalomaze that’s a low amaze
gigajax
Ayo? @vivid mango level 4 !!! 
caveman heavy
what are you using? also what are you sampling settings?
Reminds me of ila
this is actually fire, prompt?
checkpoint?
Who drew those
this was at the osaka univerisity of art gallery. they had a room dedicated to just teto
there was dozens of beautiful paintings featured. they were just having fun on the wall too
IPadapter
using Heihachi as input image lol
had to use high noise value and decreased strength to prevent the background from being just clouds
using Lei Wulong
Jin Kazama
also Kohya's Deep Shrink is really messy, half the time (if not more often) it makes distorted faces
Turned it off for now.
Here is Soldier Boy with IPadapter
so much more blurrier xd
another Jin Kazama image
it gets their likeness pretty well
Erosion Bird in SDXL
Smol
Ayo? @neat nexus level 5 !!! 
Where's mine at? You LL be hearing from my lawyer
ayo
yo best online way to do ai art??
3 easy ways:
Bing AI Dall-E
GoogleAI
https://civitai.com/
I think tensorart something something is also pretty good, but not like I have much experience with these platforms, since I create it locally
Explore thousands of high-quality Stable Diffusion models, share your AI-generated art, and engage with a vibrant community of creators
ty buddy
Sorry I forgot, I’ll do one
civitai looks good ngl
ya got any advice??
best settings??
if you go to the page of a model they are likely to tell you which settings they used, you can also see settings from generated pictures(if not hidden) when you open them:
these are just examples, just click the create button when you have a model you like @shell star
Here you go
ok really helpful thx
ASHJGAJKHSGBJKLNADGHJADHJGHJADGB
I LOVE IT SM
:)
He is not amused that he’s holding the British flag because that isn’t his country
this is AI?
This is Google‘s AI yes
😭😭😭
google has an AI??????????
THAT'S WHAT I WAS WONDERING
If you live in the US you can use it, just enable some Google Labs experiments and if you have some in your library you can use it
fucking US
Yeah I know
the VPS I have in america that I have set up as a VPN:
so i cant use it
Also I basically described how the character looked from a reference and gave it that as well as adding the style at the start
love it
I-I be poppin bottles
im not british
Ayo? @opal zephyr level 64 !!! 
(first design is amazing too)
True
silly mood :3
Can you try detective pikachu? Like the hyperreal version
Hm
oi fuck off
I mean it does generate
Sometimes the AI thinks it generated a bad image, but sometimes regenerating or trying a different prompt works
If you’re less descriptive it’ll give you realistic images
oo
one se
@shadow bane try "Detective Pikachu from the movie Pokémon: detective Pikachu. He's furry and wears a detective cap. Pikachu Pokémon, super realistic photograph"
@shadow bane mind sharing the link?
If you go to https://labs.google and click on the third button that says “Get Started” it should list the current running experiments if you sign in
I looked everywhere on that website, can't find it at all
Then once you enable it go to https://google.com/search/images
you mean this?
Yes
can't enable it since my account is not from the US (even though I'm in the us technically speaking)
Oh I forgot to mention there’s a limit lol
Sometimes if you change your account location to the US it works
interesting
I do have a couple Google accounts that have the feature available
you can do that?
No Category:
help Shows this message
Type /help command for more info on a command.
You can also type /help category for more info on a category.
Wow so aesthetic 
Ye :)
Ayo? @marsh silo level 1 !!! 
having a hard time getting homelandinger
IPadapter is sometimes ruined by kohya's deep shrink
hampter
friend of mine actually told me it looks extremely close to nintendo's artstyle
what's the generations limit btw?
Idk but I think it’s around 25-50 generations
Hey guys
Ayo? @polar moth level 1 !!! 
what do you put in here?
stuff you dont want for sure
extra hands
long next
extra eyes
get the jist?
With every purchase of Tayloo Sponges, every cent goes towards making a revolutionary tomorrow. Now why not join the Sponge revolution and buy the Tayloo Sponge? After all... you couldn't shake it off the mind of this product, even if you wanted to
Looking back into my old AI files and found this dated 11/15/22, it's one of the first ever generations I ever did. And it's using the Waifu Diffusion Model from the NovelAI leak. I think it somewhat still looks decent
do me! do me! 
in the morning i will not rn
@shadow bane check your messages I have dmed you about something
Squishy you are too adorable 😭😭
I tried lol this is the best I could do
can you do me too
Sure
🙏🙏
🙏
Kept blocking the images most of the time but this is the best I could do
I forgot who that character was
It's Basil
@strong cosmos lookie

how is that #🏙│ai-images 
Dataset 

looks like some pokemon trainer 
not basil :3
Thats kinda true
:3
Yeah
can you pls do a me? 🙂 (my full pfp for reference)

@strong cosmos this u?
nope
:3
i wish i was but nope
its not me :3
:3
he got da drip 

yay
@shadow bane can you do meeee
An alien chinchilla
I hope that's not too much to ask 😭
Why is the hair green
Isn’t their hair blue?
I couldn’t tell
Adorable
Ayo? @kindred viper level 48 !!! 
.<
WTF what is the name of the model
Ayo? @past moth level 1 !!! 
That one isnt ai generated
what the fuck is hands that then
The person is holding the other's wrist
really, I hadn't noticed
yo who gave u the permission to do ai art with my face??

AI is truly incredible
I can't send my ai images here :(
Chat a bit more, you just have to hit level 1
it's just an anti-bot thing
Ah k, thx:3
This deserves to be in a gallery
So Yeah the images I wanted to post here but can't are basically all about the idea:
"What if FNAF
But modern Disney parks and animatronics"
Drippy
Ayo? @dim beacon level 1 !!! 
spider-man on snot
In the city :)
I really hope you don’t spam these like you’re doing in the MattVidPro AI server

The only thing allowed to be spammed here is Tailork Swiftram 
Youd think with how ai art works we'd get a lot more of it smh
Spam your hearts out we'll stop you if you take it too far
I think it's the fact that it is not the main part of the server, both servers that I'm on that specialize in image generation had people that were sternly talked to for spam 
only every other month, but it does actually happen
Ty :)
I put more in the server I mentioned earlier
Meowmeowmeowmeowmeowshutthefuckupmeowmeowme-💥💥💥💥💥💥💥
Ayo? @hearty cipher level 1 !!! 
how to generate an AI picture?
maybe start by using the bing image generator
is it free?
yeah
thanks
ngghhgnngghmmgmggghhh
This goes so hard
Only if I could screenshot it....
yo what model did you use?
raetoonmix
I love smug mouth 
Ayo? @shadow bane level 22 !!! 
ty
Yes is free You get about 15 token per day for fast generations but you can still generate images even if you spend them all it'll just take longer
these dogs look pretty real
The double tail, weird tounge, pantigrade-like front paws:
Ever saw tails from sonic?
Can anybody give me some advice on training a Lora on an article of clothing?
Like how many images would I need for that and would I need to crop in so that the clothing is the main focus?
so you just want the clothes to generate in the image and nothing else
How to generate
/help
No Category:
help Shows this message
Type /help command for more info on a command.
You can also type /help category for more info on a category.
onm android idk if you can use SD, but you can use the Bing AI app to use dalle3
One message removed from a suspended account.
Anybody else knwos?
you can use all browser based things on android, as well as some android apps
google offers ai images, Bing with Dall-E does, you can go use SD on CivitAI.com, you can use tensorart for SD etc. etc. etc.
There exists one work in progress program for Android where it uses stable diffusion. Although it requires you to build the app from Android Studio and send the app over ADB to your phone.
The app is very limited in what it can do basically it’s just prompt, negative prompt, image size and select model,
Currently android is lacking in Neural Network cores unlike the iPhone where it uses it for Face id
imagine
Yes
she if from Sakurasou no Pet na Kanojo
Yes
it wouldn’t take too many images, just enough variation with the images should make it work
oh and yeah you’d need to crop
has anyone managed to recreate the artstyle of scott philgrim?
Scott Pilgrim vs. the World Style LoRA This model works best at a weight of around 1 to 0.5. Lower it if the output looks too overcooked. It's trai...
Neat, thanks
I used Civitai for the first time. I thought I could use this and combine another model. Which was Luz from Owl House.
The results were...
Bleh
They suck
yeah that does happen, the scott pilgrim model did seem a bit over trained, what weighting did you use for it?
At first I kept it at the standard weight. I had to mess around a by increasing and decreasing them to get slightly better results
Ayo? @surreal merlin level 4 !!! 
You can try using this one
I've been having better luck with this one https://civitai.com/models/196429/bryan-lee-omalley-scott-pilgrim-art-style-lora?modelVersionId=220849
yoo pretty good
@brittle ocean
can u tell me what im doing wrong here?
@tawny peak
im just super confused

you have an open parenthese somewhere in your prompt
it shouldn't matter it should still just generate
wait nvm i didnt saw this
your doing sd now? 
you may want to re organize your prompt
i just copied this from an image to start
oh, well then they should re-organize
Automatic likes it when Loras are called at the end of prompts and can freak out sometimes
c a t
hm
nah its not lora problem
this is the image im using
but it renders
Yeah i'm aware it can work, but sometimes it can skip over prompting data
what error is it giving you?
found it
no rerror
add ) at the end of realistic
yeee
should show up under the generation box or in the terminal window
if there was an error
Ah, you gotta download 4x
no you can change the upscaler
i just set it to None
isnt it better to do upscale in extras page?
Hi res fix can change the whole image if you do it while generating, while the extras one just upscales the resolution
ooooooooh
it's how I get the images I send here
it's just happening too slow for me
what is your gpu?
1070
how much vram?
do you have xformers enabled in your webui-user.bat
Hi
idk what that is
i mean this is possible to make something cool without it
i made janky comfy ui set so that it automatically does that xd
xformers is a speed booster in simple terms
i need this
to enable xformers: open your webui-user.bat file or whatever file you used to initally open the UI, and find this line: set COMMANDLINE_ARGS=
these are the arguments I use, although you may want to change them depending on your system config: --medvram --reinstall-torch --no-half-vae --xformers --opt-sdp-attention
you can actually forget the --reinstall-torch one
i just use it if another program destroys my torch install
Ayo? @tawny peak level 10 !!! 

ait
im too tired to figure this out today
anyone wanna help out with a little project? i just need 1 image 1920x1080
and yes it'll be clickbaity cuz that funny af
i can try but idk if 8gb of vram can handle it

@autumn moat shh
i saw nothing
u got the lora?
ye
i want them running, scared
that's it
imagine... temple run
but hd
u get me
?
xD
it's just that
so just environment jungle and on bridge ?
if u want, sure :Þ
yo wtf
i don't see why now, literally anything will do
even running in a dark room would be good too
dark neon running room?
idk
i can dm u what this is for
🔥
yea sure dw
@brittle ocean check dms
Wow can you believe those Swifties thought she was going to announce Rep TV? When she hasn't released her featurette on Creepin?
Daaaamn!!!
😎
Ayo? @brittle ocean level 24 !!! 
😳
rizzed up by missclick this is a tactic 
No Category:
help Shows this message
Type /help command for more info on a command.
You can also type /help category for more info on a category.
🤨
Na I took it
yoo
Ayo? @serene storm level 48 !!! 
why here
Can i have a image from there?
/help
No Category:
help Shows this message
Type /help command for more info on a command.
You can also type /help category for more info on a category.
What
these people seem lost
True
Ngl I’m guessing he’s underage cause he’s in half of every roblox server I’m in

What?
wait how tf this emote send
anime
/help
No Category:
help Shows this message
Type /help command for more info on a command.
You can also type /help category for more info on a category.
/help command
No command called "command" found.
/help command
No command called "command" found.
/help category
No command called "category" found.
@shadow bane #1159290666019471421
Does anyone know of a very realistic model, other than realisticVision?
and which collab do you train Lora in?
https://colab.research.google.com/github/hollowstrawberry/kohya-colab/blob/main/Lora_Trainer.ipynb
I use this for training, very simple
i swear ive watched the anime that this character is from
Sakurasou no Pet na Kanojo
I have watched it
Yeah me too
mega shiny gengar >>>
Not even close lol
this chimp wants to make a deal with you
Tempted to add some more
If it can give me some exclusive (Good seating) tickets to an Olivia Rodgrio concert for less than 2k xD
Anime
prompt?
Wait duh how u makin say a tutorial
the doom shredder
Aaliyah
yo i want the prompt for the gengar ong 
Most of the generations I’ve uploaded varied in the prompting, but the gist of it is “Digital art/simple flat drawing of a black colored fuzzy Gengar from Pokemon, orange eyes, wearing a black jacket, angrily happily smirking, crouching in a cool stance, pointing with both hands, orange and purple squiggly earthbound vaporwave background, blurred background, overcast, cool, cute, awesome, full color”
Some guy in another server iterated off of that and generated versions with Lucario, Snorlax, and Eevee
My 3D ones have the prompt of “Digital art/3D render of a black colored fuzzy Gengar from Pokemon, orange eyes, wearing a black jacket, angrily smirking, dancing, orange and purple squiggly… (etc)”
One simple change at the start of the prompt makes a lot of differences in the image
@glossy snow me when I click "learn more"
that i actual know u made with ai right? but if u made this with ai, i'd can say u made urself!
Please stop talking I can barely understand what you’re saying
alr sorry dude
||furry vore, ultra detailed, realistic||
You scared me for a second 😭
comfyui makin peak for me
/help command.
Took an AI image I made and then destroyed it in audacity
this looks AMAZING
How did you do this
Makima but doesn't look like makima
Ayo? @quartz sparrow level 10 !!! 

it does though
doctor who

rip hands
Big booba

What AI/Prompt...??
Bing Image Creator/Dall-E 3
The prompt was along the lines of “Digital art/Simple flat drawing of an anthropomorphic Lapras from pokemon, blue eyes, wearing a brown puffy jacket, grey jeans, smirking, blue and tan squiggly earthbound vaporwave background, blurred background, overcast, cool, cute, awesome, full color”
Sora from Kingdom Hearts as a Pokemon trainer
why is it so reflective LOL
That I have no idea
Part of the new logical processing of Stable XL Turbo? XD
I was going to use Good ol' Reliable Stable XL on hugging but it has runtime error x. x
Found a Stable XL host, ngl it has more options to play around with.
I got admit, I am falling in love with this new Stable XL Host...
i like making furry stuff with sd for some reason
I mean why not? ^
On a similiar note: I just finished up the song that this cover will be using ❤️
Miko Multiverse
the ai generated patreon logo lol
barely even looks AI besides the hands
i forgot to put it in the negative prompt
ikr
i finally found the perfect prompt
and style
do you mind sending it? also what AI are you using
i honestly can’t believe we’ve gone from DALL-E original to this
ai as in model
im using automatic1111
and fluffyrock
yeah
Ayo? @noble cedar level 2 !!! 
thanks!
and what program you use for AI images
automatic1111
you like Anime Illust Diffusion XL too huh? 
some of you have checked SVD ??
hey bois
omg furryyyyyy

dragon balls
@odd lance
is like me
thars why they sexy
poopyyyy
poop
furry x poop = ?


view on your own discretion, it's giant ?furry? poop 
looks more like a massive pile of brown cotton candy
well it is furry 
it doesn't even look that "horrible", but for some reason when I look at it for some time, it causes a gag reflex for me 
yum yum
Yeah good checkpoint
maybe I can help
imagine a cute version of the Grinch but light blue instead of green
@brittle ocean
middle of the screen 1x1 pfp
could be making a funny face
(i hate xmas but i guess itd make sense to atleast take some part ... uk)

that goes hard
@rustic raft
looks a little like a kappa yokai xD
just wish these were blue
I used this to create a Lora but the Lora was good but the image quality is like this
You created this Style, in this case it was you who created the style, or it is not Lora
if that person is the person you tried to train, then either the problem might be the parameters of the generation , or you overtrained the lora too much
I put 15 epochs, it was a lot, right?
Ayo? @past moth level 2 !!! 
Apparently it's not because of the epochs, I just tested one here with 10 and it did the same thing
hmm, that depends on so many parameters so there isnt a clear answer as to what could affect the training
i have trained more than 20 loras and its still difficult to understand how to train it better
In this collab, you can train style too, or isn't it blonde who trains style?
the detail on the left background... 
AlbedoBaseXL, my checkpoints are always in the name, so if you open it, it will show it 
why mika pikazo art style is so hard to train😭
Did you train as a model or Lora
lora
try training as a LoCon or LoHa, i’ve had better results with style models if I use it
i thought it only can train lora
im blind
A1111?
yes
im getting this :
RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check
haha what
i just read what must do
what GPU do you have?
let me try now
Radeon RX 570
im checking if it works now
i did --skip-torch-cuda-test
then it will not work, it will use the CPU and be super slow, old AMDs are not supported and even new ones are hard to get to work since they need ROCM which requires lots of steps to do by the user
you can maybe use it with SD Turbo, which is usable with just 1 step
why cant it just use AMD
Ayo? @vernal wyvern level 2 !!! 
whats that
cause CUDA is the driving force of Machine Learning(or currently known as AI)
and that is from NVIDIA, therefore AMD support is very miserable atm
ok
you can use tensorart and co if you want to create stuff
whats that
false
theres a fork of a1111 with directml support
works pretty well
oh that's great!
i used it on my igpu before switching to comfyui
comfyui is much more optimized than a1111 and i get the same speeds on cpu than on igpu with directml

made this using comfyui btw
yeah it can start, but use the link here to get the version that actually can use your GPU:
#🏙│ai-images message
yea no just use the fork i sent
I know, I hearted it way back then 
so i need to reinstall the whole thing?>

if you installed using a venv you can just use it with this repo
and you can copy hte models folder over
yeah models and co can be shared
yea
and it will work?
@autumn glacier did you try turbo and LCM yet with your CPU?
when ur done with setting it up just put it in the root dir of sd
bruh
nah
but it look like just same thing
no
the git link
it is the same base, with some new code for GPU communication
ok wait how did you install stable diffusion in the first place
Do i just put it in the files?
yes after you install stable diffusion from the repo i sent
having python and git and then in cmd i just paste this:
git clone https://github.com/lshqqytiger/stable-diffusion-webui-directml.git
now just put it there?
mhm
windows defender says i shouldnt use it
set COMMANDLINE_ARGS=--medvram --backend directml --no-half --precision full --opt-sub-quad-attention --opt-split-attention-v1 --disable-nan-check --theme dark --autolaunch add this to webui-user
ok
why medvram lmao
lowvram is best even on good gpus
make an exception, that stuff is open source, so any virus would be obvious to the thousands of devs that use it
also why specify the backend if youre already using a directml fork
i alr sent a p good bat for launching
it works so I dont see a problem
eeehhhh just sayin
i didnt know that he is using diff
fork
or
idk what this si called
wait
lol
i was watching tutorial how to set up this
and i got this command from tutorial
since this is for a radeon RX570, normal tutorial recommandations might not be optimal 😉
i have 580
i have vega 8...
well you have double the VRAM 
he can have 8gb
about? xD ok then medvram isn't the worst
you can play with it though and see which one works best for you
just sayin that lowvram is good even on highend gpus
well I don't know much about A1111 and how it handles resources, so I have to trust you on that 
which of these is best for training anime character
1st
ohhh is that the famous Anime-leak dataset?
so cosy!.... and the girl and the cat in the background 
im gonna try doing partial conditioning in comfyui
its where you can have different prompts for different areas of the image
did i earn sd creator yet
could I use it as my new pfp please
?
ai generated a new lang
sure, the cat in the background right? 
omg I hadn't noticed this 
this cat is uhhh
japorean 
lol
Finally I managed to make a "good Lora"
wow good job, looks very close to a real shot out of the anime 
idk ask @shadow bane
did i earn sd creator yet
😎


u gotta be honest... its for ppl who are knowledgeable
i wouldn't know, I'm not here at all
Tell me a patron artist so I can try to copy the style in LoCON (FOR TESTING)
@left carbon 5 image posts... and only today.............
cmon, I we need some more 
so I can better learn how to create styles and Loras and improve my knowledge
i kinda doubt a Patreon artist would be ok with that..
alr
it doesn't hurt to try
five..
lmfao
@left carbon ur amazing
thanks
sure 
joining that server for emotes
alr nvm
so many fuckign notifs
a lot of people in here with cartoon pfps lol
To create a style, do I need several different images of the style I want?
what about just adding the pack?
probably
is that out yet
well I'm not on that server, so for me it is
I have basic nitro though
ah
What is the average number of images for the style to look good?
idk
pretty normal really online and especially in here where art is what we look for 
neckbeards
the foot
some of them are femboys so they probably don't have beards 
no foot moment
yeah the troubles of AI art xD but hey the image looks very good, even the detail of the other hand being inside the sleeve and still being modeled
true
i was generating these before i got the style i wanted
@autumn glacier
this keeps jumpscaring me
what does?
the mod ping
i keep forgetting ur staff
lmfao
it acutally does look like something mia would use
wha
lol

im glad i finally found the style i want
aye that's half the journey 
half?!?!

Is 100 images for style enough?
Awesome!
shit
i’ve gotten styles working with 30 so yeah, be sure to use lycoris though
Lycoris
Ayo? @past moth level 3 !!! 

Yeah LoCon is Lycoris
how many epochs and what resolution do you think I should put
you can use the default settings you don’t have to change anything
not even the resolution in birme?, it's just 512x512 right?
if your dataset in mostly above 768 pixels on both width and height, training on 768 should be better
again the same model does one of the most heartwarming smiles, reproductionSDXL reall nails it.... ignore the broken background characters xD
Can someone do an ai pic of israeli solgers?

do that on your own time, you can do that easily online, thousands of online AI image generators available
@shadow bane 😦
bro
What?
Ayo? @subtle nebula level 1 !!! 
you're trolling
Why will i?
get lost
Holy hell you can do a lot with Pokemon in Google’s generator
(Click for rest of images)
So can someone make the solgers or not?
Idk how but why did you said to me get lost
read the 3rd note in my bio
Ok
And i dosen't bring war fucking shit into this i just whant to make it my pic
sure, my dms are open 
google AI, Tensor.art, CivitAI, go have fun
Tnx
i kinda wanna see what would happen if u play with this image
hm might be something interesting to play with, I will check it out tomorrow, already almost midnight here 

is there any way to optimize it on amd?
well things that go higher resolution will always take a long time, did you play with https://github.com/wcde/sd-webui-kohya-hiresfix before?
nah ill try
Meina Mix Model(My Style 8Epochs)
Meina Mix(My Style Final Epoch)
may want to tweak the weighting a little bit to get rid of the artifacting
good night peeps 
Real Saya?
LOL Sorry
booba

New OC again again
Why does he have a bulge in 1/3 of them
Maybe that’s why the filter is being tripped

Anyways
💀
It’s supposed to be funny
Looks as if the data set on the model didn't crop patreon watermarks 
ikr
most of the time it doesn't have a watermark
you can try combating that by putting (text:1.2), (signature:1.2), (watermark:1.2) in your negative prompt if it keeps generating them
ikik
Pee joke haha
sd creaotr role
@shadow bane try this
What model is this?
fluffyrock
yeah and it does styles really good
Don't remember what the artist's name is but it's very familiar
how do i get the sd creator role
Post a lil more and u get




bulge in 3 of them