#🏙│ai-images
1 messages · Page 48 of 1
That weird pattern you got there cinai says the model is still trying to resolve a thing it didn't figure out from the prompt
which one? i'll just say its a feature not a bug 😄
that one ^.^ it's not a bug so much as the model can't understand something in the prompt that might effect 'style' parameters
what patterns do you see?
where?
you don't see the pattern it drew the whole thing in? it might be easiest to see here on the middle woman's left side back.... where that pattern inthe purple is actually across the entire piece pretty much
in the image the pattern is purple, but extends over the grass, and over the water using that grass, etc
Dam sharp eyes
I've played with a few images^.^
But that's a recent thing with the new gpt model... it just kinda gives up when it can't undersatnd a certain part of the prompt and for some reason it just does that
Hey thanks for the review, the NYC one I did right before had nothing, felt completely dead, no emotion at all, so I just started throwing paint at it basically. Funny enough what you caught in the purple pattern is exactly that same instinct, the model found a texture it understood and just kept applying it everywhere like a painter who found their mark and couldn't stop.
post
Well, it's less of a review and more of a moment to try and share a fix. if you go back to that image and 'edit it', with something like 'continue to resolve what is still not resolved in this image.'
Sometimes it takes a time or two iterations - if you ever want to salvage one of those that turn out like that
wasn't judging ya ^.^
The NYC one is suffering the same thing... it simply ran out of effort tokens before it had to fire... understanding the cause of it - a confused part of the prompt - is the most important thing Im trying to share rather than judge on 🙂
Appreciate the fix but I like it unhinged, that's the painting. Her hands are the only thing bothering me, everything else I stand behind. Artistic intent carries it, meant to look like someone actually touched it. Post it though, good or bad, judgment is the whole point, that's how museum pieces get made. Just put it up.
Thalessa?
How many Ocs you have
one
Emilia
I tried to make another girl
I saw her as an android, like a sex doll you know, with soft skin, blue lips and eyes, short hair hime cut.
results were okay, looked cute, but inconsistent. when I tried to make a lora I got a reminder that I SUCK at loras so... I left her
lol thx I won’t have seen it u till the poke
So skill issue 😭
sometimes I feel like I exist just to make things wrong yet somehow sometimes get decent-ish results and make smart dudes angry by that XD
One day that skill issue will become a issue skill 😂
Idk what the hell I’m saying
yeah, I'm fine with my foxy
I love her
That’s the thrill
lmao
nice one
Looks like king made a visit today
😂
nah, I don't see Gnom here
Gnom who
Oh
Fairs his the king of generation
kittyyy
@steel haven just joking my dude
you are beyound existance something whatever
ass cool
also thank you for breathing and drinking water
AND FOR RAM CHAAAAAN

hmmm
Well said
Alr gn
gn~
gngn
Sup y’all
hi
Hmmm
A movie poster
It can be whatever but turned as a movie poster
Like a title and all that
thats a format not an idea
How about the most realistic city with realistic scenery but it have to be night
sounds like a good idea
night city is light physics, how neon actually bleeds on wet pavement, how a streetlamp creates a cone and then nothing, how glass buildings absorb and reflect simultaneously.
that's a prompt
as you say lol
night city is light physics, how neon actually bleeds on wet pavement, how a streetlamp creates a cone and then nothing, how glass buildings absorb and reflect simultaneously.
I love doing silly things. such as "using 100% wrong prompts in illustrious" XD
HEY CUTE SD MODEL, CAN YOU PLEASE GENERATE ME A BEAUTIFUL NIGHT CITY, WITH NEON LIGHTS, RAIN AND REFLECTIONS, NO PEOPLE AND FEELING OF LONELINESS BUT STRANGE PEACE INSIDE?
I know it's wrong
but I don't care. It's fun to me to play like that. It's stupid but it works
everything is a prompt
heck i threatened to throw this guy's mum into an Ai like she was a prompt...
if it fits, it ships!
no i said his mum ^.^
Thats a good job
And cool pfp
I think that’s an actual anime show




gn all
Gn
This is what ambition looks like when it loses the plot. The city stopped being for people somewhere in the middle of building itself and just kept going, roads crossing roads that cross nothing, buildings mirroring into the sky because expansion was the only language anyone remembered how to speak. The sepia makes it feel like a found document, evidence of a civilization that got exactly what it wanted and couldn't stop once it arrived. There's no human scale anywhere in this frame. That's not an accident, that's the confession.
lets prmpt bros
I got school
are u in hs?
Guess
idk
anyone know best ai to make artwork ?? alongside using my own artwork skills, so for example if i make something in photoshop what ai software is good to import that into that will make it BETTER?
canva?
it's got the same 'can't complete prompt' look i was telling you about
look at the lettering on the building
yours said dry gods, was that intentional?
if you have to draw letters, it isnt art
@muted dust tried to focus more on lettering
you should just pm me your prompt ^.^
i'm trying to teach you how to stop stressing the model so it can actually type in readable words...
i use natural language. and i try to make it look like a painting.
Yo can I ask you a technical question
I'm not home so not directly in front of everything, but what's up?
So is it possible to make a character with a similar face or same face as a manhwa character
Because the art style is totally diff from one another
You see the character is on my pfp
Gotta get a lora I think
How would I make character from the image
You'd need several images of that character, up to 60, ideally in the same target style you're looking for
You can try using Nano Banana to modify images for the dataset as well
Let’s say I create a Lora in the original style and want to use it for a different style would that work
the more outfits and backgrounds the better
If you just made the character yes, it will work with different styles but it would be hard to say... Get realism out of an anime character
Alr thx so much for the feedback
But you could probably take the same character and wrench him into DC Comics style
that's not too far of a leap
Any time dude, if you have more questions just pop in my server in the #1159289934407020586 section, someone will definitely guide you to the promised land
Alr thx
Also your pink alchemy anima is 🔥🔥🔥
Just need to improve my skills
Thanks dude!
Just make the lora for the model architecture you're targeting
ruhroh
these are with yours
very nice
@astral ridge Lock your seed number first, that's your character's DNA, don't change it. Then build her with two or three words that hit her core tension, whatever makes her react, that's your trigger architecture. Every prompt after keeps the same seed and just adjusts the trigger words slightly, she stays the same person but the scene moves. The seed holds the skeleton, your words add the muscle. Simple, repeatable, same character every time.
Thanks for the feedback really appreciate it.
tfw out of credits but want to prompt paintings
I wasted 2 credits because i wrote pikatchu in the prompt it wouldnt generate due to copyright
i do locally
give me the prompt
i haven't seen you do a painting, just make it look like a museum masterpiece
alr
The magic is in the words man, I just type whatever I associate, no SEO no predetermined anything, everyone's different but when it comes direct from mind to medium that's where it lives.
so i create painting without an prompt. i mean my own prompt?
alr wait
i can't draw but i know how it should look 😄
tried
Here’s a strong, cool original painting prompt:
In a breathtaking epic painting, a lone silver-haired swordswoman stands atop a crumbling ancient temple floating high above the clouds at twilight. She holds a glowing ethereal katana pointed toward the heavens as massive aurora lights and shooting stars streak across the deep indigo sky. Her long flowing hair and tattered black-and-crimson coat billow dramatically in the wind. Cracks of golden energy run across the broken temple stones beneath her feet while distant floating islands drift in the background. The entire scene is filled with a sense of power, solitude, and otherworldly beauty, rendered in rich dramatic oil painting style with visible, expressive brushstrokes and cinematic god-ray lighting.
Did you come up with that yourself bro? It reads a little processed, the words are colorful and the scene lands but it feels constructed, like each element was placed deliberately rather than felt. Double the natural language, let the feeling breathe before the description kicks in.
Then try feeding it raw feeling words before the scene, twilight, solitude, crumbling, ethereal, ancient, shattered, radiant, windswept, luminous, desolate, transcendent, brooding, crimson, indigo, golden, tattered, billowing, colossal, silent, mythic, stormy, glowing, weathered, majestic, haunting, turbulent, vast, charged, fierce, serene, raw, cosmic, painterly, expressive, textured, layered, atmospheric, melancholic, powerful, timeless, alive. Let the model absorb the mood before it touches the composition, the feeling first then the frame around it.
I’ll be honest I didn’t
Cus i tired my version but then i was way more horror to be uploaded
Original plan was sun rising in the distant and new born baby lifting with 2 hands
Sounds horror, that's good, just type 50 to 100 words yourself, whatever you associate, don't think just go, and see what emerges. It will look better than you expect every single time because the model reads the energy underneath the words not just the words themselves. Your chaos is the prompt.
I create locally on comfyui
But I’ll take your advice as it will help improve my prompting
Comfy UI is insane once it clicks, connecting 50 different nodes and watching it actually work, each little tweak changing something downstream, the whole thing humming, that's the fun part, when the system starts feeling alive and you built the nervous system yourself.
Ikr
Ik
I almost sent you a friend request but then I was like...
I'm shy.
they did the same to me
no answer as to why
sounds like a bot
true
Y’all
yes mr Batman?
yo
Come up with your own, what you generate from someone else's idea looks like someone else's idea.
Beautiful 👌🏻
The artist lives in the tension between order and collapse, mapping a city actively consuming itself, lantern warmth against storm cold, control on the left, apocalypse on the right. The train means progress that doesn't ask permission, industrial age making time linear and scheduled, every town synchronize or get left behind. The dirigibles are empire floating above consequence, refusing to read the weather. The dome is institutional permanence, built to outlast the men who commissioned it. The Crownfall sign is the tell, empires create wreckage and someone always profits from it. The map is the colonial imagination made physical, to draw it is to claim it. The split sky is the exact moment civilization realizes the price of its own momentum. Everything here is a culture at peak belief in itself and the storm rolling in from the right is the part nobody in the frame has turned around to see yet.
I may have spoken too soon. that model ends up putting the pattern on ice cubes for some reason. So stressing it is only part of the problem... there's some sort of bug in that system making that pattern....
it's not rendering some things correctly and nobody knows why
thanks... i finally got a web surface that has half a style up as a demo...
it's not nearly as cool as my GUI but it's a start >.<
Thanks for the feedback, didn't catch any mistakes on your end either. Zoomed out that water reads as a flat silver-blue mirror strip with no ripple variation, which is exactly why the model keeps hallucinating ice cubes — its texture classifier fires on the luminance and reflectance signature and calls it ice because the painter never broke the surface up enough to read as liquid at distance, so the bug is upstream, the pattern trigger is reading surface features not object identity. Artistically that's a legit debate because an Impressionist would say still water IS a mirror and you paint the light not the water itself, while a Renaissance painter maps every ripple geometrically, so the artist could be making a conscious choice to flatten it and let the sky dominate the reflection, or the "manic artistic" prompting just pushed expressive mark-making everywhere except the water where the model defaulted to the path of least resistance — intentional tension or lazy output, hard to tell from the outside.
nice progress. the web surface will become cool eventually
The one you want to dude.
What is the best ___ is just an ignorant question to ask
what is the best author with cleanest NSWF art on civit? 
Rodzilla 4 sho
New rodzilla
Rodzilla was just Batman's alter ego
oh that one is choice work bud
Damn Zerg... you let that Ai breath or you just drilling it till it gives it's last breath?
lookng good tho!
ty. "still life" fruit bowl but then went somewhere else lol
that was my... idk, last 30 mins?
thanks :3
oh wait
lie
I got ~30 pics
so ~ an hour
🙂
Almost got the web demo up... but look at the prompt closely, not every single row of the presets is meant to be activated at once.
The buttons are easy mode, fast bundles that you only need 1-3 lines of honesty.
But there are so many possible combos, that I'd be here trying to cram presets into every nook and cranny there might be on the web page.
The actual app lets you save states and will have further developed tools to help people navigate their way through the known universe of prompt descriptors.
At least for the chatgpt model.
I need to do styles particular to other models, too...
The web surface I'll share in some short time has a lot of these easy buttons... but look closely at the prompt... it's easy to pick 'too many lines' and get some repeated words in the activated levers...
that sort of stuff gets patched out later, it doesn't really hurt the overall understanding of the prompt or conflict with it...
But when I drop this, the easy mode of just the buttons is probably going to be the 'make or break' if you 'like or understand' the project
How do you post pics?
you have to stick around and chat for a while before that get's automatically enabled
Oh ic thanks
ooooh this webpage look so basic but it soo good
wait i really like all 3 for some reason, it just tickles my brain in ways i cant describe but i like
@daring lichen not sure who to ping for this but this is a scam i think
i swear ive been to this place before but without that basket of fruit
makes me think of the model "Plant Milk"
ooo thats nice
woah ok wait AI is gettding too good
i say that but then look at all of this 😭
Thank you, I deleted the message, I've already told that guy not to do that
probably a bot tbh
No image, probably joined 5 days ago
hmm
Most likely a bot
sounds like a bot to me
If the user has no pfp I assume it's a bot
Kinda sucks yea

yeah that's the problem when you push a model right to it's edge sometimes ^.^
this is an image of the world i'm building
Because so far most people know if it's good it was probably mine :-p
(completely kidding)
no i got a banner i use on my like really styled stuff
Dang. It’s actually crazy how good AI is at pixel art
I’m a bit conflicted how I feel about AI art.
I don’t see it is actual art, BUT I think it’s a great tool for learning the technical elements of art as an educational tool.
Inversely, I think AI art should list what references were used/what artists they learned from in creating the piece
AI generating art is definitely one area of AI I do find personally more ethically concerning in terms of plagiarism, but I’m also not sold on the concept that it’s like actually stealing/plagiarizing directly either
yeah but it's also the least 'artsy' so i figured it would be a good one out of my 120 styles to just let bleed into the world
I think AI art doesn’t plagiarize so much as it synthesizes patterns to reproduce certain styles
I think where AI struggles most is if you try to make it actually create art with its own voice/style because it has to learn from what it sees rather than what’s within
no you're posting nature images in an ai art thing and links are bad
We're going to have to discipline
yeah it has nothing of it's own style really... it's just going to try and give you what it understands as your preferred style
I guess eventually it will be able to have more its own creative stance and vision. Which is a bit scary cuz art may rapidly advance in a way artists aren’t ready for
AI could explore aesthetic possibilities at inhuman speeds, it could dramatically alter our relationship with art
that's what makes my prompt forge so cool.. you can whip out 10 prompts faster, dynamically than an llm will give you 2 or 3 static ones
what's up peeps ¯_(ツ)_/¯
I made my name at with the chatgpt boys - by developing an art style never seen before
That's the new direction any new artist is going to have to find one day
@dapper hollow@muted dust@narrow pelicanHey great pics looking good, all these are art — what's hitting is the compression, time stacked so thick the images feel geological before they feel visual, centuries refusing to dissolve, the oldest sediment always pressing upward through everything layered on top, light arriving like it traveled further than it should have, shadows so long they stretch across eras not just surfaces, atmosphere that doesn't sit on the image but lives inside it, the kind of depth that pulls you through the frame into somewhere that existed before the image did, civilizations bleeding into each other without resolution, nothing clean, nothing finished, every piece carrying the particular gravity of things that were built to outlast the people who built them and somehow did.
Does anyone recommend the best AI tool to create legit looking graphic design cover art for music?
What I've tried so far was incredibly basic and the font was like comic sans despite asking for specific font...
I'll attach image for reference of styles you can see previews of the other ones on the left
next time mention what you've used.
Yeah makes sense I legit forgot lol!
It was propper random
All good
You probably want to use google or ChatGPT's image generator tbh
They're both EXTREMELY good and customizable with 100% ledgable text
You think? The free versions? Better than Runway, Midjourney etc?
ChatGPT all the way, can't go wrong with a cover, and the secret is imperfection — a cover that's too clean looks generated, a cover that breathes, has weight, has one thing slightly off, that's the one that hits, good enough is the whole philosophy, overfine it and it loses the mouth, loses the body, loses the thing that made you want it in the first place.
Midjourney is a question, but yes.
I'll take this one and phonk-inspire it
love dis one
too much boob (
=_=
don't get me wrong
I like your stuff
I'm just afraid that big dudes will be worried
Cool image
Blue image
ew religion
Why aren't you prompting anything?
not as interested in ai images as I used to be back in 2021
I prefer the voice ai stuff, much more interesting
It's come a long way since the Will Smith eating spaghetti days, oh well, anyways yeah sounds good — prompting is just fun at this point, vids, words, stories, images, it all works and it's all fair game.
ai videos are fun, I just prefer to draw stuff myself tbh
I wouldn't consider myself an artist but I know how to draw
thx wormy :D
yur welcome Grim :D
This is a postmodern building, sky playground, cat utopia — signifies longevity, wildness, free spirit, and what it says about life is that the structure gets built with logic, every ramp and platform connected with actual rules, and then everything alive moves through it sideways, finds the gap, takes the tunnel that wasn't designed for that, sits on the edge of the platform that was meant for something else, life doesn't follow the architecture it just occupies it, longevity isn't compliance it's adaptation, wildness isn't the absence of structure it's the refusal to be defined by it, and the ginger cat half out of frame on the right isn't lost, it just found the part of the system that the system didn't plan for.
snakessss
pattern is tamed power, dangerous thing made loyal, the snake is always bigger than her and always coiled around her like protection not threat, someone is working out a relationship with something overwhelming that turned into safety
The one with 40k followers who never acknowledges chat directly but reads everything, runs her own schedule, doesn't do collabs unless she wants to, and the audience stays because indifference from her feels like a reward when it breaks.
Looks good, totally feels like a game — but when this is actually playable and you're standing inside it, torch light hitting wet stone, something moving in the corridor you didn't render, what does that feel like?
Well this isn’t to eliminate actually making the game. At least not yet
For AI image generation and game making I see it less as something to do your work for you and more to get inspired and get stylistic inspiration or some pointers here and there for how to render something
Technically competent art is what the internet is using AI for right now, the output looks correct, lighting atmospheric, genre cues present, but you feel nothing because the person who made it never went inside it, they assembled a format and called it a world, and a format is not a world, a world comes from someone generating out of where they are actually standing, from what they feel in the dark right now, and that is the only thing that pulls anyone through a screen.
I don’t think AI should be used to create complete readymade art. I think it should be used as a tool. Or at least a human should have an extensive hand in directing it. I think art will shift to needing stronger skill sets in directing and taste over raw technical skill
it's already happening, and not only for art
now your tech knowledge isn't as important as before. but how you use AI and what you can do with it is important. So we are coming closer to the era where your softskills and vision is much more important than hardskills I guess.
I know it’s happening, I just think there’s a limit you can’t cross for authorship if you want to actually make something artistic and expressive
A director directing a beautifully rendered stage he didn’t make, a script he didn’t write, and only guiding the actors can make something special. But if even the director isn’t actually doing anything he has not created art
At that point you’re slapping your name on a title and taking credit for doing nothing
you still have to make a banger and make it more profitable than loosing money
the problem with placing a border is, you know, the Ship of Theseus problem
at which exact point it becomes not art but a ai slop? who decides? why?
questions we will never be able to answer
(probably)
There is no clearly defined line, but that doesn’t mean a spectrum of authorship doesn’t exist. Ideally all art should innately possess a very high degree of authorship, because that’s what art ultimately is
I don't know how computer
approximately yes
nah that would be pretty accurate I'd say
I WAS WAITING FOR SO LONG
first victim :D
Now hear me out
everyone likes DINKS Donuts
Like the donuts are okay
But everyone wants to see the IRL anime waifu at the counter
It's weird, she's 3d but 2d at the same time!
uh, you can just order those from japan now too
which ones?
old dolls are kinda meh
new robit gurls are still in progress 
yeah most of those robots would crush a fat man tho
they aren't that light weight 😄
yeah, plus they are, well, just not processing well lol
so nobody tried out my bare bones pixel art webpage?
Looks good, the figures remind me of Michelangelo, the bodies carry real weight and that is already rare, and Michelangelo never wrote the word God between Adam's fingers, he put a gap there and let the entire weight of human longing live in that inch of empty air, and Dali never labeled his clock with the word illusion, he just melted it over a branch in a silent desert and let your mind come apart quietly trying to make sense of it, both men hid the meaning so deep inside the image that finding it feels like something happened to you rather than something you were told, and that is the move available to you here, trust the figures you already have, let them carry the feeling without the words explaining what the feeling is.
Damn
he already posted it
&?
it wasn't good the first time
Nothing you make strikes any chords either
All I see is low resolution blur
@dapper linden after two and a half years I'm beginning to really hate what the community has turned into
Just a bunch of dickheads who turn on chat GPT, punch in 3 words and say "mine's better"
With exactly ZERO effort
And it's been happening everywhere too, and these same people have probably never seen a LR graph in their lives
hard to take the craft speech seriously when the feed is full of neon girls and physics-defying proportions, that's the strip mall you built
Cinai, last I heard art is subjective.
Do you know what this is?
posting softcore and calling it culture isn't
Yes or no
Yes and I've seen enough AI "Still life" to clog a warehouse full of toilets
So I'm just going to say, you use chatgpt, I build shit from scratch
We are not the same
You can't train a model, don't know what a quant is, don't have any idea what a lora is, right?
So you're basically fingerpainting
yea I know what that is, it means you can connect nodes and read directions, good job, thanks, now let's just prompt
Can't handle a little discussion outside of ripping on someone else's shit?
Kinda sad
youre right
wat?
THAT'S THE REASON I STARTED MAKING THE PROMPT FORGE... for people who type 3 words...
got so sick of looking at slop its like here' i'll write the other 30-50 words you couldn't figure out
LOLOL
How'd you get this picture of me 
this one ?
Ye
bradar what is this
fluffy fat black cat?
Light holding dark, the angel is not afraid of the black cat and the black cat is not afraid of the angel, and whatever you think needs wings to be sacred and whatever you think is cursed to be unloved are the same thing held in the same arms, that is the whole theology in one image.
you were talking to me like I type three words but you're the one teaching your model to output gothic girl red eyes on a loop, and building a forge to add thirty words of algorithmic noise around the same prompt isn't craft, it's a vending machine with extra steps.
i did no such thing on either end of your specific deduction.
Got it, you're the cartographer, everyone else is printing the same map.
alogorythmic noise? There are 23 different descriptor types that affect images in the ways that a Photoshop program would or the way a director of a scene would.
i don't know if you realized yet, but i've been working with the original image model that all the other image models are copies off, for years...
and i'm one of the few that were successful at pushing the image model past know barriers without any extra training data, consistently.
in fact, i didn't speak to you as if you had 3 word prompts, i specifically asked you to share your prompts so that i could help you look at where you were conflicting your prompt and making the model give up to save effort for the render... which takes a lot more than 3 words could ever do
algorythmic noise... your hubris and projections are noise
neon ramen vendor, guiding a lantern boat through reeds, with a collapsed bridge far behind them, pixel art, sprite-readable structure, arcade gameplay clarity, authorial tilework, loosely observed sprite depiction, rich surface work, selective depth emphasis, dense layered detail, ultra-clean finish, layered scenario storytelling, noticeable retro wear, suggestive motif use, strong gameplay tension, strong scale awe, light playful charm, soft daylight, crisp sprite contrast, restrained palette color, aspect ratio 1:1
point to the noise.
i won't be holding my breath... go touch grass
kid sees saved presets for a potential of 4.3 million supporting prompt outcomes that are controlled by the user for fidelity
says 'algorhytmic noice'
.... don't chump bleeding edge work just because you don't understand what is in front of you.
23 descriptors is just rating consciousness on a scale, you dressed up a checklist and called it a system, and sure okay make maps, I genuinely thought we were building something together but turns out you're another anime producer pulling whole prompts off civitai and rebranding them as methodology, and the forge you built is just a fancier copy machine pointed at the same source material everyone else is using, and if you're really working on the original model that all the others copy from then why does your own hypothesis and your own test data point to GPT as the top performer, because that tells me everything about where your work actually lands, and all the architectural grammar and directorial logic in the world still produces lower resolution output that you run through an upscaler anyway, so the noise you're accusing me of is just your pipeline described honestly, good luck with that.
The eyes are the whole painting, half closed in the specific register that is not sleepiness but sovereignty, the look of a creature that was offered the full complexity of existence and considered it carefully and chose the chair and the popcorn and the salt air instead, and the museum visitor who stands in front of this work long enough will feel not pity but a specific and uncomfortable envy, because the cat is not checked out, the cat checked in to something smaller and warmer and more honest than the thing everyone else is still chasing, and the popcorn is almost gone and he is not thinking about that either.
awww my man
I think you are little burnt out lately
maybe even not a little
if you wanna talk about anything I'm always ready (well, except those anytimes when I'm at werk)
and I still have a feeling that you guys are speaking to ai in this chat 





Sky is making his auralis anima btw
soooo
maybeeee
I'll try to make merge between your guys stuff
(if you share how to merge this shit ofc lol)
I'm still a lazy noob
...I should finetune.
Hii Zerg, congrats on moving to full mod status btw :3
gratz to you too :3
Hii
I'll wait for you 
Why am I bloo?
u what
blue
Ah it must have been Vijay
THE ramthrust

Promoting someone to mod whose content is algorithm bait is putting the vending machine in charge of the menu, the standard for what belongs in the space is now set by someone whose curatorial instinct peaks at visible areolas, and every community that starts with genuine energy ends up here eventually, one promotion away from the strip mall running the door, the bait feeding the algorithm, and the culture reduced to whatever gets the most reactions before the mods look away.
so far I've only seen not quite objective criticism and kinda (as for me at least) mediocre stuff from your side. If you have something more objective to discuss we are open to it, but I haven't seen any areolas here or other nonsenses you were talking about
Mods aren't quite in charge what is allowed and what isn't - it's for "high" mods and admins.
We are here to keep things clean and to protect community from filth
wow you really are mental. all of those 23 descriptors and the 14 others that can join ARE OPTIONAL.
it's giving you the entire pallette with paint on it seperated into colors, but only a fool throws the whole mess at their work
i'm sorry but you're simply too stupid to understand what i've built and frankly too stupid to understand that by sharing your prompt on those images you could have discovered that i really did know what i was talking about.
but you want to call 'model failure' a feature.
as if the world doesn't already know were that pattern comes from.
just stop talking to me, your so overly judgmental and literally mental that i don't have time to deal with it
go produce some software you understand, like snake, or maybe an angry birds knockoff
sorry bro this guy's nonsense is on my last nerve... i'll chill if you need me to.
i put up a simple version for one style out of 120 that i have to offer and the guy thinks he knows all my stacks are defunct.
i'll be.. nicer.
all I could ask, ty
yeah, I don't understand him as well
not like I'm a big pro or something. but at least I was trying to understand when my model makes a fart and why it does. this guy doesn't try...
each model's different... they all have the same core training for the most part... but each model has grown accustomed to different types of words to be used - in a nutshell
but you already know that I think yea?
sure
can u say the same to civitai so it doesn't die every time i try to load image pretty please
XD
nah civit is just dead
always
i don't know if i asked you or maybe i missed it but what models do you use? I can look into their specific semantic maps, or build some if they're not availbe yet.
I use illustrious. my merges, if you want to look exxactly at them - https:/ /civitai.red/user/zerg212/models
before that I was using RAM's pink alchemy and anti-nova models (illu as well). they are gorgeous
the exact merge I'm using for last 1k pics at least is "dark_demon" one
oh lol
civit is really dead right now XD
i got your thing bookmarked, but it says zero models... i do get to see all your image posts now ^.^
=_=
you don't see everything
half of my stuff is NSFW
maybe models aren't visible because of that as well. you may need to login to see -_-
oh yeah XD
welp
RIP :D
oh yeah that might be a feature rather than a bug... i'll make an account
got it. Next time i'm out of coding credits i'll look into those model bases and what they respond to ^.^
tyvm
Dark Demon, Cyan Cloud, Rose Raspberry, Black Bloom, Blue Berry, Anti Aura, Pinky Promise, Purple Fog, Pink Auralis, these are not painting titles these are Sephora products, and the grid is ten variations of neon woman against cosmic background with the color palette rotated one click to the left each time, technically he trained ten checkpoints and called them a range but it is one image with ten haircuts, and the man who told you that you were too stupid to understand his 23 descriptor system built his entire catalog on the aesthetic equivalent of changing the filter on the same photograph, and Cyan Cloud has 281 downloads which is the market confirming that the vending machine is working exactly as designed.
nothing dude, I don't hide anything 🙂 if someone wants to look into my stuff - why not
well, if you can judge checkpoints just by looking at 14x19 pixels screenshot through two downscales and only by preview images - touche
well eventually i have to learn the semantic maps for every single model out there because i'm that ocd about my software lol
but it's not like your opinion is valuable for me, sorry
It's good stuff bro I'm just saying it sounds like lipstick or nail polish brand.
why it shouldn't?
moreover, you have been pulling that criticism even not knowing the fact how these names were desighned
so... idk
bro you're still coming at me with 2% understanding of a 110% project and talking like you understand it.
just because i offer the entire semantic map doesn't mean somone should unleash it every single image
it just puts people in the directors chair so they can focus on the artistic side of things instead of typing out a million things or waiting on a single static prompt to come out of an LLM
me like ur stuff 👉 👈
and i'm rich and can't even close it
zerg likes rainbows contrasting against really dark colors and women...
not anything wrong with that ^.^
hahahah so classic. this bug is pissing me off for last few days
the whole operation is a better way to make prettier wallpaper
in 5 hours my checkpoint will be on generator btw, so it may be used if someone is using civitai generator
as video models get better and better, having the best images to begin with becomes extremely important.... it was never about the still images ^.^
Wallpaper that nobody went inside is still wallpaper no matter how many descriptors built it, the question was never whether it looks good, it was whether it means anything, and they answered it without knowing they did.
(●'◡'●)
CC is good. it's pretty clean. I like how it looks, probably my cleanest merge
but I use dark demon now because of how wild it is and how it likes to make more complex scenes with simple prompts. plus it's artistics and color
yeah and it also works really good and smooth with almost every lora
it was created only based on best dudes models :3
Why so pessimistic?
You might like to know that I've been selling prints of my images because they've been a cut above what the internet has to offer for almost a year now.
Maybe you don't do anyting with your images but they can be cut into web assets, app assets, prints, graphic design, and about 50 other commercial lanes.
You're just so driven to convince us that things are worthless when literally the entire internet runs on text and images.
lol we use AI in our company to make commercial ads, "professional" portraits for employees and many other stuff
come on, even logging system is partially integrated with an ai
I'm not even saying how much money it generating for us just with being on our 1st line as a chat bot in our help chat for customers when they need help. because we have shit ton of customers from all around the world
AI is very, very useful. the problem is that most of people don't know how to use it and in which cases it's a good idea to use it and in which cases it's not
Artists suffer when they sell out, show me the maps next time not the pron, out of credits anyway so GPT tomorrow, peace.
gulp
I've been an artist for 25 years and you call me a sellout?
I applied my artist spirit to the Ai and created art styles never seen on the planet before.
i did show you a map, but you didn't take enough time to study it and i haven't shown you pron.
maybe you come back tomorrow and not be such a jerk to someone who tried to share experience with you man.
no to ai images
lgbt
wdym?
wh40k
ohh cool
@timid orbitwhat the hell is that name... help @dapper linden?
reminds me a little of overwatch
lol cinai you're intentionally pushing into that pattern now... perhaps hoping for some comment other than the fact that it's obvious you're intentionally trying to incorporate it :-p
No u
gosh that show was so funny to me when i was a kid
gif wasn't deleted
aw;(
yeah sometimes bug may happen and message survives
or I'm allowed to smash gifs by someone...
we will find out next time XD
yuck
Agree
i gotta drop it.. google ai made it.. one prompt.. its for my new tool. Hilarious so good they got. I did feed it Tux.
did they implement brain thought sensing already ? it was a very short prompt ehe
it was waaay cooler when it first came out... but then, diablo 1 is way cooler than it is now.... ^.^
Hey, any of you guys have any kind of tips on getting the best one shot image, um, to output in the highest quality possible? Like, at the moment, I'm just using ChatGPT for ChatGPT two point o. Um, and then my goal is to use that to make a video out of it. Um, not as a reference, but just as, like, a sort of animated version of the picture itself. Do you guys know what the best workflow for that would be?
best you can do is just try. again and again
you see, the point is - you didn't describe what you need at all. like, AT ALL. what style, what concept, what is on picture, what framing, what subject, what lighting - nothing. plus, you said it's for the 1st frame of video - what video? again, nothing.
And not like it's your fault or anything. Anyway no matter how hard you try - you won't describe it good enough to make us see it just like you do. So, the best way to do what you need here - is just to go and try over and over again
if that part about the best workflow is about what I suspect it to be - man, no one would be able to say what you need with this level of description
(and don't even try to ask me about all of that. I'm pretty noob, especially with big things like gpt images. I really don't know how to play with it)
idk, maybe Windy has some advises for that, but I suppose he will need quite more info as well
ive been having claude slightly alter my image prompt and then i feed that image to gpt, it's getting close to what i want - i just wonder how to get the model to output the highest possible resolution
or am i stuck with upscalers?
gpt outputs are quite sharp as is. Plus, the best one to ask about gpt images is gpt itself lol. well, sometimes
if you got pretty lucky already - worth trying just to explain what you want to gpt and ask how to make it better
<@&1159293140440723499> Hacked alert.
What they hide is simpler than the image suggests and more devastating, they hide that they believe the chain is necessary, that without the weight of constraint the offering would not be taken seriously, that a free hand extending a flower is somehow less than a bound hand doing the same, and the painting is the confession of someone who has convinced themselves that their own limitation is the proof of their sincerity, and the blue rose glowing in the dark is everything they have ever wanted to give that has never quite made it past their own wrists.

Auru Omega on her Ligthing Dragoon armor
Dude, just take a hike.
You clearly don't like anyone in here and you go out of your way to just bash people's stuff. Like what value do you get out of it?
your reaction
I guess so but it's just not helping anybody having a chunk of rotting flesh in the punchbowl.
what if he just has a good mood by ragebaiting?
I know pretty many people like that
(and you too. you hang out in civit discord dude, how it may surprise you at all?)
Maybe

Maybe I'm just used to people being clever and not just whacking people over the head and saying "u bad"
NOW I NEED MY 0.5L OF COFFEE
that's why our place is divine
ty for that
Even @odd lance at the end of the day is funny :3
aw, that was cute of you
Anyone got a good model for ponies for comfy ui?
I got good anthro creatures outta furrynovaxl
@steel haven is an expert in these things as far as I know
What, Pony? Pony's super dead
It's essentially pointless, v6 or v7
that's not about model I'm afraid
sorry 👉 👈
nobody in here is actually prompting, you're all friends who share a love of the same civitai folder and pointing at me is easier than showing what you've got, so let's see it, drop something, anything, I'll wait
nice bait, almost got me
painting, painterly, ghost in busy tavern
forgot the mastuhpiece
angry goth woman lora
beautiful!
it's ur CC so yeah :D
ah, resrgan, I see.
I don't like it a little. can't say why tbh.
I miss lines sharpness a little. I think if you will increase denoise here to 0.2 it'll be little sharper
ah goddam, it's USDU
what gpu u have?
cute little 5090
why USDU then in first place
it's the only way if gpu is weak. but with decent one it's a waste in my opinion.
But I don't insist

dat what I use (and RAM as well, but with 2 steps, extra 2 is just for even more detailing)
this is only partly true now
well, if you just model upscale and do regen with illu or how else you do it - it's still much better than usdu lol
sorta. Anima needs stuff and things for upscale
best upscaler for anima is illustrious
i did that thing back when i had 3060 and there is almost no change since i can throw u this thing if u know how to make it better i would like to know (●'◡'●)
try this setup I shown
you may start with just 2 steps
the only tweaks is needed sometimes - lowering denoise a little on close-ups or with photoreal/semireal stuff, and raising it on large wide shots.
as upscale model I use "2xNomosUni_esrgan_multijpg"
the most popular (I think?) choise is 4x-UltraSharp or remacri_original. I just tested a bit and nomosuni looked almost same but little better to me
1100seconds? damn, something wrong XD
dunno it's 4 steps
220 looks pretty okay. It's little longer than I do with 4steps on my 5080, but still looks okay
220 with USDU
strange
you should do faster with usdu
and MUCH faster with iterative
could you show me your settings on upscaler?
weird
and the bottleneck is the upscale step?
just saying - my upscale factor is 2.3
what batch size btw u usually do ?
I suspect you are getting OOMed somewhere, and when VRAM is at 100% your PC just stutters
that may happen due to various reasons, but shouldn't happen on 5090 with properly built workflow
¯_(ツ)_/¯
I am getting OOMed sometimes
but I have just 32RAM and 16 vRAM, and my wf is a mess, so idk...
my wf is a mess and i also do not endeed know
hmm
hmmmm

it's strange
I suppose you got weird one
idk, should not be happening XD
yes, 100s looks much more convenient
i would say so :D
my poor gpu would be happy to hear that :D
some colors are lost
but it wasn't a locked seed, so can't really tell how much we lost here
i did exactly the same seed tho
anyway, best way to find a thing that will work best for you - to try more other setups. other schedulers and samplers, other upscale models - everything can work. Or not work 
that's weird
it's not the upscale difference - it's totally different, from the base
some things looks locked, but everything else is moved, diff posture, clothes - everything. strange. idk, maybe USDU was changing it too much, but it shouldn't with 0.16 denoise
well i did tried a lot when i was building this, and it was a long time ago, and it was just working :D
can u tell where remacri and where ultrasharp?
ofc not
I didn't use both for ages XD
:D
both are good actually
just different
ur nomosUni
well, lemme try 5 then XD
oh wait
it's amount of seconds for 5 pics?
that u are showing
yeah
:D
I'M DOING 120-150 SECONDS FOR EVERY PICTUREEEE
damn
I blinked
what was that
XD
so here ur thingy
pretty close in comparison
1st (left of two) and this one last are my favs
well, you can just take the fastest one and don't care much
difference isn't that big to take care
plus it may vary seed to seed between models
so, idk, can't tell
i mean between this 2 it's 50/50 for me, i don't like how eyes look on the second, but the little details i like more then on first one
last one is overall better imo
lol
it's 100% the opposite for me XD
:D
but hey, it's just taste I think
overall lines sharpness is good on all three I think
and it's way higher that with old usdu setup
I hope it will make you happy :3
BLINKED AGAINNNNN

how u do this
XD
nobody knows :D
noice
if i take the whole thing somthing around this now
for one?
whole wf with batch of 5
ah
yeah, seems good
so if you will raise upscale factor to 2.3 I think we will be somewhere close
but you don't have to. it's kinda excessive
I just want my pics to be 4k lol
eh
i did just that and it's kinda works itself
it's actually a bad idea
if possible you actually want to use one of these
for base gen
:D
because illustrious is trained on 1mp pics with specific ratios (I guess, may be wrong here)
but every single smart dude was saying to me that it's best to use 1216x832 or vice versa or 1024x1024 or so. so, ratio should be like that
it will make much less artifacts in pictures, and overall result will be better
anyway, if it works for you as is - do as you feel, for sure
true
ah, here we go, found it
it is might be better but for me at least i didn't see much difference tbh, perhaps from 100 pics 4-5 was a bit better
most of my pics just that ¯_(ツ)_/¯
yeah, sure
if it's fine for you just do it
I actually know some cool big dudes that doesn't comply to it as well, and they produce masterpieces
so... anyway
OOOOOOOOOO YOU REMINDED ME
OOOOO HELL YEAH
GOD
DAMN
I altabbed for 3 seconds again
idk
you are a wizzard XD
if i were a wizard i would've summond rtx6000 :D
a couple
i can't show my wide images i get bonked :D
some... bups? XD
yeah pretty much no so some :D
XD
my DMs are not
ah, okay, now it's time to get some real life and grass touching
anima base is really good
hello! i have card game with alot of symbols, animating, like fire golem symbol etc, they have few animations, like win, idle, etc, i would like to expend and create them with ai, is it possible? i seei its mostly used for nsfw and girls. currently i do spritesheets, or spine, but spine i know is impossible, so static images would be ok, but i have issues with it 🙁
lol, it's just we like doing girls XD
try to explain what you want to gpt/gemini/grok, they are quite good with tasks like this
but for end build it's probably not good enough, just for concepting
welp yea, but few issues:
aspect ratios are different than what i achieve, its strugling to get sizes right
transparency is an issue, but its ok, i took handle of that
girls have 100000 hours worked on them by ai, i swear they download porn for training data
symbols and abstract fantazy stuff does not have training, i wanted to have comfyui workflow for that, but its hard af,
codex and other image gens are struggling as well...
spritesheets are quite possible
even animations are possible nowadays with AI. even 3d ones with blender models
3d i know, convincable spritesheets.... i have to see it to belive it
but you have to dig some info for it. I definitely saw some videos about it on youtube, so not a secret tech
i know some people gen videos, and transparent background, after that they take frames from vid and put it into ahnimation
hmmm
I don't know about that. I mean, I just wasn't really searched into that, don't have any experience here, can't tell
fr?
right in front of a mod?
but nevermind with animations, simple static placeholders is what i try to do now, but its strugling with sizing and creation
so, what's the problem? I don't quite get it. aspect ratios? what ratio do you need?
Pink alchemy the best

@steel haven why all your gurls are angry last 3 days? 
me sad
you sad me sad
I made a few loras
They're not mad enough
What kind of loras
Also bro is saying like it’s piece of cake 🍰
depends on who is talking XD
for some guys it is not something unreal
we got even few lora lords here actually
ya, character loras are super easy to make
they are though 
once you get the right parameters, it just becomes second nature
still finding the perfect parameters for anima, illustrious I already got down
...
Illustrious for me was:
10 Repeats at 10 Epochs save every 1
3e-4 Unet Learning Rate, 6e-5 text encoder learning rate
cosine with restarts
16 network dim & 16 network alpha
Prodigy optimizer
turn off text encoder training
It mentions that in the circlestone labs whitepaper
like in all caps
:O
Doesn't mean it won't work
but..........
If the author says something I just do it
I never even looked at the docs
I meant specifically for anima (I too am animaz)
I said fuck it we ball
I still find it funny that Anima is actually faster than my Illustrious workflow, went from 70 seconds for illustrious to 45 seconds for anima.
Looks cool
It's made for memeing
and that's pretty much the goal for now but I'll end up making styles.
Yh I can feel the meme from the face
So everyones coming to anima
nah dude, hide it XD
Yh we can see the but
I can see the damn... whatever, these words are forbidden here XD
How? I’m struggling getting a dataset ready
I see what you see
It’s gone
ya i nuked it
So how do u create dataset for the character
As I’m using gpt to create different shot of the character
As I get consistent result
But only make 4-5 of them
Before hit the limit
collect a bunch of images for the character, label them in numerical order starting from 1. Then alongside it you need a txt file also named the same thing (e.g. 1.png & 1.txt, 2.png & 2.txt) and so on
I tried banana but it was inconsistent
I get the genera idea of the setup for the dataset
then for tagging you can either use AI to tag it or tag the image yourself
It’s the actual images of the character
And tag means that you mention things that will be changed in future where as the things that define the character will not be tagged
Right ?
basically just describe what the character is wearing in as much detail as you can as well as expressions and poses
The more detail you put into tagging the original outfit the more customizability you have later on if you want to change it to something else
This is for anima right. Also do u add tags or just natural language
yeah yeah
and after that you dudes smash some wizzard potion on your PC
I know
Tags & Natural language will work, though in my practice tags work best
Bro don’t worry I will travel this treacherous path for you bro
But shouldn’t natural work as it going to use anima
Also another question bro, how do you create the image for your dataset for character
yes it should work
i don't I find actual images
Oh dam
very rarely will I use ai generated images in a dataset
Well thank you very much bro. You are a legend and your tips will be used. 🫡❤️
What about OCs
I just create them using the base model, and tune the prompts around to get repeatable results
no loras needed
no loras for example:
The moon one is peak
@dapper linden I don't wanna go back to Illustrious :<
It's been a year of it and much success it's just time to move on to a new architecture
Yh he should jump towards anima
who knows the best ai site that generates greate animated photos and videos ?
seriously wtf am i doing
@dapper hollow sorry man, I deleted one of yours. bikinis are a no-no here









