#🍥|anime
1 messages · Page 44 of 1
Suspended in a "bed" shaped to her exact form and upside down. The bed being held up by like a bunch of organic "cables"
the organic cables help her see far across the world
and well the bed and chamber too
or holding a katana in a wide legged side stance, the katana in the hand furthest back and resting on the other hand , the background being an arena with a crowd
The first one sounds really crazy, 2nd one sounds kinda standard :o
They do?
I feel like they don't look bad, but they are missing something 
Also they're like a shrine/temple maiden :o
lora or model for the style?
oh gosh I've just closed SD xD
can you send me the metadata?
Yeah I think so! I've always been a fan of less complicated designs. If you want to add something more though, maybe like wings or a tail?
Lemme check it on smaller img
coz this one is upscale
Fair fair, and I was planning to add a tail, but I think I'd need to do like a meaty tail LoRA, cause I think the normal demon tails with like the lil shaped tip are kinda basic :o
Steps: 24, Sampler: DPM++ 2S a Karras, CFG scale: 7, Seed: 2062924846, Size: 512x768, Model hash: 099e07547a, Model: Anime_darkSushiMixMix_brighterPruned, Denoising strength: 0.64, Clip skip: 2, Hires upscale: 2, Hires steps: 30, Hires upscaler: RealESRGAN_x2plus
lora:2bNierAutomataLora_v2b:0.8, lora:mochizukiKeiArtStyle_v10:0.6
I'll put you in a jar
and prompt?
absurdres, (((masterpiece))), (great composition), (dynamic angle:1.2), (half_body:1.2), extremely delicate, beautiful face, 1girl, solo, lora:2bNierAutomataLora_v2b:0.8, lora:mochizukiKeiArtStyle_v10:0.6, simple background, yorha no. 2 type b, solo, white hair, puffy sleeves, short hair, mole under mouth, long sleeves, juliet sleeves, black hairband, parted lips, black gloves, cleavage, turtleneck, bangs, feather trim, feather-trimmed sleeves
thx

what's absurdres supposed to do?
make it absurdres
High res but bigger
ever tried a comparison?
make it looks nice 😄
better quality
i like max detail personally
kek
Upscale until you mindbreak it
Upscale it until your PC burns
better

I mean, I sent a 8k x 12k pic
But it mindbroke discord
XD
O.o
am i the only one, because this is totally my type of girls XD
You like depressed anime girls? 
yas and the hair + skin
the "vibe"
Fair xD
Whats wrong with it? 
it looks different and im not digging it 
It looks the same to me 
preferences 😄
cyote
Oooh I like that

She hasn't had enough sleep
like me
😢
Same
guys, zoom in.. it looks soo good
oh look its my sleep schedule in a pic
Same 
I bet she's trying to get perfect hands in AI
(she hasn't slept in 3 days)
aint perf but its nice
dang was hoping it look more like me since i used my selfie in contrlnet but thats rly pretty
lol
?
agree
she has a neck like an owl
if u dont mind the lack of example pics u can go model browsing on hugging, found a few nice ones that i needed to test to find out
@untold glacier someone made an updated patch file for SAG with the new auto1111 update. Just replace the patch file in the extension directory with this one and run the command from the original extension repo to apply the patch.
https://gist.github.com/CamWiseOwl/eb13ba59e8750717f20db3af7f8364db
i hope what i'm doing isn't an heresy
i'm using ft mse vae on aurora instead of the advised kl f8
she's tall
(and giving off the middle finger cuz fuck you)
The main problem with mse is that it usually struggles to get decent looking anime eyes. It can also struggle more with fingers in especially flat art styles. Aside from that it's quite good. Probably the best available as far as colors.
i use animevae
for anime
mse for non anime
pastelwaifu sometimes if i feel adventurous
i think the creators of aurora or someone of that level could easily make a record setting vae
Luna is nice
yis
Problem is that nobody really trains VAEs. We still only have 5 total aside from ones that people have applied tweaks to contrast and brightness settings and a couple that were merged and don't look very good. There's the WD team's anime and anime2, Trinart team's anime2, Stability's ema and mse, and the redacted VAE. Basically all of the ones out there that are named after models are just WD Anime2 or redacted vae that's been renamed so it can be loaded automatically by WebUI, maybe with brightness or contrast adjustments added (even that is uncommon though)
anime2 vs mse
I don't even know how to begin to train one
Last time I looked at it, the trainer needed at least 24GB vram.
damn
I can't find anyone that wants to sell a 3090 for less than 10k used and I risk buying a ex-miner card
which if what I remember is true, uses similar executions that will affect the memory speed and quality etc
as to SD
ha
civit 
why isnt a Vrchat AI(gpt3.5) bot existing?
depends on what u mean vrc bot lol
I mean... it uses vram. But miners also undervolt and underclock their cards so that they aren't spending all their profits in electricity or damaging the card. Mining cards likely to be in better shape than cards owned by an average gamer.
like a model that is autonomous running gpt's llm within vrc?
unlikely cos tupper hates innovation
🙃
i mean user made
would require someone to script it within the world but chances are if someone actually manages to do that chances are wont stay on long cos previous statement of tupper hating innovation. killed the modding scene in one swift action
I think someone made a test on that and it turns out they're around 5-10% worse than the average second hand card
they are punished and they do something with them to force signals to just go through the same path
rather than be an all round card
LTT has done a few videos on it and found no meaningful difference. Fans might be more prone to failure due to being run for many hours.
this.... is not a thing.
it's not?
are you high
yes
makes allot of sense
no
I also haven't slept well in a week, I'm still in a lot of pain
but I'm not gonna use that as an excuse for being potentially wrong
lol
they want their vram to be as full as possible so that they are making as much money as possible. Why on earth would they try to force only some of the vram to be used?
no that's not...of course they access it all. I didn't say they selectively pick a vram
I meant they bypass a lot of other features and nonsense on the way
to optimize the information flow
or whatever you call it
This is what I remember reading about when I was trying to decide if I wanted to buy a miners card or not
could be bullshit, I could be misremembering an important detail
I could be right for at least one subset of miners
I dunno
Wow! I'd like to see other versions of this hottie
some kind of bios mod
apparently
and this is one of the reasons they undervolt
so they don't fry the cards
just did a few tests of my model before taking a break
oh wait this was with kandinsky nvm i remember
They sometimes use a moded bios to turn off things like video output and cores that just aren't used at all in mining to save even more power. That has nothing to do with why they undervolt the cards and isn't damaging to the card. The only reason it would be a concern for a future buyer is that the stock bios would need to be reflashed before the card can be used normally again.
Dead inside and wandering a dark looking forest
Sounds relatable in a mentally taxing way
I love girls with bags under eyes
no
understandable
thanks
which then?
Adjusting the weight can change the degree of dark circles. Staying up all night training lora gave me dark circles.
those "comics" people are putting together are awesome
any anime pros here
i look for a long time now for a anime i watched long time ago
maybe someone here recognize it: a girl lost her family and she once saw a package full of cats on the street, as she came back the cats were gone which was so traumatizing to her... i dont realy remember on much more but maybe someone does
It was Majic Mix Fantasy 2.0 - "majicmixFantasy_v20"
yes
I like something about it
could be a dope wallpapar
Our vibrant communities consist of experts, leaders and partners across the globe. They are developing cutting-edge open AI models for Image, Language, Audio, Video, 3D and Biology.. AI by the people, for the people. Learn more here 
thanks
was this edited?
you too
I feel like im boutta steal it fr and use as a wallpaper 💀
you can if you want, the prompt should also be in the pic
lol
thanks xd
no
what model are you using?
aurora
is aurora that good?
I like this one
well I use ti's and LoRA's aswell so that can be biased, but I like the model
with what model can i do this?
Fantastic news. Thanks!
midjourney
mm i finally did an update today. so far it's going ok. 🙂

working on getting hairstyles generated and then picking the best, doing the same with eyes. Gonna read that guide I was recommended in a few days
I want to be able to actually type in the name of a hairstyle, however made up and it shows up
And with stable diffusion?
Pirates in the Eldritch Realm ride on ships like these in an eternal war across the ever-expanding and collapsing sea. If they don't sail forward fast enough, they'll fall down to the hungry elderly Eldritch.
idk go check civitai or something
SwinIR upscale is missing from my list after updating 😦 😦
it's a sign that it wants you to use multidiffusion
Well, to me thats clearly something that took more effort than just generating, pretty sure it is SD, but its clearly heavily edited
This image looks like it was composited to me. I don't think it was made with just text2img. I suspect the background and character were generated separately and that the sword and bird were added in post with a pass through img2img at low denoising.
and now I realize that @magic wave said basically the same thing as I did.
should have prompt right?
I just informed about image being stolen



i'm trying to make original stuff out of SD, seems like it has endless possibilities !!
True
I also made this using MJ then SD
omg there is LORE?
Yes XD
@elfin geode Proof me trying to upload 8k x 12k res, I went too far, it broke discord
woah there!
forever loading.. 
Yeap, and the page doesn't load even if you click original XD
nubby's cow submission was the largest image i've seen posted here #1087493722645725184 message
That's what you get for trying to outdo my 12k x 4k submission 👿
Tru, what a tragedy 

haha, its kind of macabre, but seeing @magic wave struggle do give me the inspiration
A old style I am fixing up.
Fair Fair, you did say my suffering inspired you XD
facts
so pwetty
shes blind she dosnt even know shes pretty
Tru
but we know
Any recommendations for relatively simple styles? Like basic cel shaded stuff.
windy
So windy it blew part of her hair off 😂
I use the thicklines lora for stuff like that
Thanks. I use 3d renders and use them for img2img but don't look 2d enough unless denoise is high and that changes things too much... hopefully thicklines Lora will help
I get the best results using it on a more realistic model at like... .4 to .6
I'll get an example in a sec
Without vs with (.6)
Closeup images look better ofc
Wow, that's really good. If you use it with a model that's more stylized, what happens?
I haven't actually tried that yet, hmmmmm
It would be good if there are ways of making things more stylized/style change without having to rely on denoising, for the consistency factor. I'll DL it and give it a try
Have you tried just adding anime screencap to your prompt? It's a booru tag so it woks pretty well on many many models.
Here's what it looks like on colorbox... pretty weird lol
Never heard of that. Writing down these techniques in breathless anticipation...
dude's a poet
Oooh that does look nice yeah
Probably so; it doesn't work well on some that I tried and kind of just messes some stuff up.
Yeah
Honestly that's one of the most fun things about AI art though, just mixing things and seeing what works/what you like
When you get something that sticks it's really rewarding
It's like recipes
Yep!
Anyone tried the Stable Diffusion plug-in for Photoshop?
wake up new smol sus just dropped

Btw, are you making new version of your bra?
yesss
In the works now I assumed?

kinda taking a break rn
man im making like 3k a month from this lol

not too bad
good for my unemployed ass
I wana jump my whole self in entirely

Do it!
Anyway, remember to take breaks, don’t get too burn out on this! It’s only gonna get better and better 
time to buy a 4090 and COPE
maybe you should, but if your gpu is decent, maybe wait a year?
I'm sure all GPU companies are making the next series of GPU with more vram for AI
due to the rise of AI
mine is a 3060 sadge
12gb?
tbh, im not sure
do they release new series card every year or 2?
regular, then Ti and a Titan, then next series
haven't tried the photoshop plugin but I use the Krita plugin frequently.
so like every 2 - 3 years?
There's several photoshop plugins
for nvidia yeah every 2 years or so depending on the tech front, stocks, world situation etc
I used to work with this plugin everytime I boot up sd in photoshop https://github.com/isekaidev/stable.art, but it has been discontinued.
autoshop is the best one that's free
This one is still updated https://github.com/AbdullahAlfaraj/Auto-Photoshop-StableDiffusion-Plugin, but I have hard time interfacing with it
TBH I haven't had any interest in the ps plugins for 2 reasons. 1. I refuse to buy into adobe's subscription BS so I still use my CS6 master suite which I assumed isn't compatible. 2. I'd rather use Krita for illustrated art anyways.
yes....buy
Hmm... That's a new way to roast a marshmallow. 
I should sleep, first day home from hospital and first day I can actually sleep uninterrupted
but first I have to promptcraft just one
yeah I'm addicted to SD
it's like crack and meth with heroin and weed in a legal and safe way that harms nobody. Except the people in my prompts. So it's nothing like them...but it is addictive
evenin'
hello there
Nightin
what's up?
Does it work with Vlad?
not sure, I still use AUTOMATIC
welcome home 🙂
quiet tonight
wizard
trying some new techniques too that uses the looping on max tokens
wizard couldn't make it to the study cellar, so her hat went
cloned herself to be more efficient and then realized she is now two students competing for the number 1 spot
cozy, nice polish and boredom
not sure why they are sitting inside the tables
oo that shine
I will have to ask them about that
I always feel like cartoonish/webtoonish styles are better at making wackier or just more interesting hairstyles
Yeah I'm digging these styles lately.
got a wide one generating
takes a lil' while on my 3060 🙂
candles needed some help to see, "I'm reading....It's not upside down", "how use wand?", "how do I become moe?!", "MOE MOE!", "welcome to my study weeb, I sense you have many questions. Which is why I will give you this 12 episode anime with no conclusion so you have to buy my light novels"
the smol brigade @lyric zealot
Caught recording herself lip-syncing lol
ok modified prompt a bit
let's see what it churns
`young small moe moe wizard, wearing an oversized purple robe, oversized wizards hat, studying in a dimly lit cellar, large white wax candles consolidated and fused together, messy unorganized book case night, cozy, flickering, wood, dim lighting, soft lighting, candle lighting, 3d digital illustration, digital art, anime art style, dark fantasy art style infusion, high contrast, low saturation, HDR, Vivid, Intense colors, extreme detail
(studying in a dimly lit cellar:1.175), (candles consolidated and fused together:1.15), (messy unorganized book case:1.125), (3d digital illustration:1.2), (anime art style:1.225), (dark fantasy art style infusion:1.175), (low saturation:1.125), (young small moe moe wizard:1.1), (wizard hat on floor burning up like a candle:1.075)
[bored], [shocked]
(acid burn aesthetic:1.15), (32bit colors:1.075)
(Multicolored penguin yellow_black_red with devil like flippers and an insidious look:1.05)`
I'm testing this prompt method
I got my acid burn!
"I-i-i-i-i-it's n-n-nnnot smut"
okay prompt works now
I'm the prompt peasant as always, cuz I have like no prompts
I've been writing prompts daily since september last year
so if I was still putting out shit, that would be rather concerning
"a moe moe's journey by Chibi Haggis"
good night everyone
Gn!
"hi mom this is my new friend. yeah he's a penguin but he's housecl---oh"
I leave you with that
and the prompt if you want to try it, my model is on civit and is used in the images https://civitai.com/models/6209/dark-gemini-v3
`young small moe moe wizard, wearing an oversized purple robe, oversized wizards hat, studying in a dimly lit cellar, large white wax candles consolidated and fused together, messy unorganized book case night, cozy, flickering, wood, dim lighting, soft lighting, candle lighting, 3d digital illustration, digital art, anime art style, dark fantasy art style infusion, high contrast, low saturation, HDR, Vivid, Intense colors, extreme detail
(studying in a dimly lit cellar:1.175), (candles consolidated and fused together:1.15), (messy unorganized book case:1.125), (3d digital illustration:1.2), (anime art style:1.225), (dark fantasy art style infusion:1.175), (low saturation:1.125), (young small moe moe wizard:1.1), (wizard hat on floor burning up like a candle:1.075)
[bored], [shocked]
(acid burn aesthetic:1.15), (32bit colors:1.075)
(Multicolored penguin yellow_black_red with devil like flippers and an insidious look:1.05)`
Neg: low quality, deformation, disfigured, deformed, bad anatomy, melted face, extra digit, extra limb, cropped, spaghetti body, contortion,conjoined twins, conjoined body parts, unnatural pose, text, title, description, logo, watermark, artifacts, header, footer, game name, game logo, sponsor, name, letters, big breast, massive breast, major breast, big tits, massive tits, major tits, hentai, eroge, nukige, pornographic, porn, XXX, erotica, nude, nudity, naked, bad_hands, burnt edges, burned colors, smeared colors, washed colors, color bleeding, saturated, oversaturated, saturation burn, sitting inside table, clipping, clipped through, clipping table, clipping floor, body inside floor, game graphics, bland anime
Steps: 30, Sampler: DPM++ 2S a, CFG scale: 7, Seed: 435062053, Size: 1024x640, Model hash: 021972d7ff, Model: DG3_FP32_Full, VAE: kl-f8-anime2, Clip skip: 1
Ooh that watercolor is very nice
Thanks 😄 Had to generate a bunch to get one I liked.
How do you get such a nice glowing effect? Is it a specific model?
mixed 526 with other 3 models https://civitai.com/models/44596/526s-cusp-of-serenity
Have anyone used Counterfeit V3.0 yet?
didnt know there was a new one lol
yeah the author just uploaded it 5 days ago
Tried to make hand positions for a 3d model deliberately difficult to generate added depth map controlnet. Also tried making the eyelashes match the hair color because models don't usually do it. 3D Input image:
Output:
Yes. Its good if you like that style I guess. Didn't do much with it except use it with img2img with 3D models. That's pretty much the only thing I do with stable diffusion these days
A tip for generating from 3D is, position character hands in places where it is more easily editable in a 2d application, if you use it. So here, it would be better to make her left hand a little higher and more towards the tshirt. Because the edit can be colored around it with plain white to match the tshirt portion
Instead of having to redraw the belt, skirt, wall, etc
I used it to make random comic shit last time
I didnt found noticeable improvement from the last version
Yeah me too, if there is anything then I guess composition is a little bit better
Christ who he cutting with that
I asked ChatGPT for a character description and it suggested pixel print clothes for her. I did not expect that to work at all
nice
I did use cutoff at 0.75 for it. Not sure if it would work without it.
Me: "Why the fuck are my gens taking so long."
Me (4 batches later): "Oh ffs... I fat fingered my steps up to 150 again."
The joys of using Auto1111 on mobile.
what mobile you got? How long does it take normally
Love this style
It's a Pixel 6 Pro. SD is running on my PC though, I just use the --listen argument so I can access the WebUI from my phone while I'm in bed.
oh ok
I don't even want to know how long a phone would take to render batch size 2, 512x768, with Hires fix doing 2x upscaling, at 150 steps. 
Anyone know if theres a way to train a characters outfit without describing it?
Like i can just prompt one word and it knows right away?
you can do that yes, but you cant do it with 1 image, you need at least like 20-30
Covered
So how do i do it
make txt for each image and just enter like skirt, or dress
So basically the most general of tags?
1st tag is the activation token, anything after that is what ever
They all have that already
But ive never gotten a lora to trigger an outfit by using the first tag
it should trigget if its unique name
What if i just remove all other tags and just put the one unique one
5_dress
10_character
5_boots
epochs like 12
Yeah ive never gotten an outfit to trigger just based on the folder names
Idk why
In all my tests it never seems like the names are doing anything
Im on mobile rn, what is that
download it, its settings for lora with example
HUUUUUUGEEEEEEE
Ill get it later, thx tho
this is so good script, it should be a movie soon. And great execution, this is fantastic
haha, the script comes from @magic wave
Cute prompt, thanks for sharing!!
What the tool called that takes wireframes for pose inputs, control net?
yes it's called controlnet

Hey I'm new to the server and want to generate anime girls, I'm kinda lost how do I download the software you guys use?
@echo timber
Currently, there is no public bot on the server that generates images. There is an experimental bot available for early server members & Stable Society. You can obtain the Stable Society role by winning our weekly events #⭐|pow-info #1087493421209485393! However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝|tech-support for any problems you encounter while installing it locally!
I tried the website but its not even close to what you guys post here
most of us use automatic 1111 locally on our computers, but you will need at least 4gb of vram on your gpu
Yea my GPU is decent enough, I just need the software
Also gotta find a good repetoir of models and embeds, defaults are, well, defaults #1047197565365538826
Which software does chewtoy use?
There's many links and downloads I'm kinda lost
Thanks
there's plenty of youtube vids if you need visual guide and not just text
Also need gucci vae(s).
Vae?
All I know is that it helps color processing ¯_(ツ)_/¯
I know very little and just followed directions and google
you will need to download models and a vae after you get the installation of the webui sorted out
What's the pricing on this
free
All free
Oh really
the power of opensource my friend
I tried websites like soulgen and stuff but wasnt really content with it so I'm trying some alternatives
This definitely looks good
But complex
I'll invest some time into it
It's 100% worth it
Problem is I got 0 idea about it other than entering a creative prompt
oh one thing to note, I think you can download and use python 3.10.9, It recently got updated
very interesting style. what model is this?
It's a blend of dreamshaper, animepasteldream, and darksushi
It's very detailed so I use the thickline lora to give everything a cel shaded look
ah thats very cool. so detailed as well. want to try this on my characters as well now 😄
"highly detailed background" in prompt is my best friend lol
(((highly detailed background:1.4))) 
Hello there everyone
hey nerf 👋
thank you, I don't know why but my mind seems to connect the image I see in my head into a prompt but intuitively because I pause to explain why I chose to do certain things
good day everyone
hey chew what model is this if i may ask?
GM
holy shit that's badass as fuck
Thank you
gm?
good morning
why is your pfp naked

though you're the one who's naked?
if you zoom in you can see a tiny piece of lums bikini okay! soo who's naked now?! huh
didn't you notice anything weird when you cropped it? lol
i dont know what she is holding, bit its being held
squints
Yep that's definitely a thing she's holding
One could say that
A pen with a bag on the end maybe? Idk lol. I've got a few images with mystery objects too. Need to go through and pull out the funniest/weirdest ones
Almost a paintbrush but the end has a bag over it
Experimenting with facial expressions. Input is 3d model with an angry face with img2img
might be a big chinese calligraphy brush
Output:
Can't seem to make any facial expressions according to the input images...
I'm surprised how well those gears came out
Have you tried using this? https://civitai.com/models/5891/crazy-expressions
I got this expression for my sticker using "sad" as a prompt with crazy expressions at... .4 iirc?
Actually I think it was like "sad" and "tears"
oi bois remember me?
I haven't. I wanted to use only ways that would work on single characters that have multiple characters in an image.
Ah gotcha!
Does regional prompter work with img2img?
want to share a drink with me?
is it snozzleberry flavoured
or... a subtle alcholism advert? 😮
made a small change to my lyrics -> llm -> keywords and it's a lot better now at getting visual queues
it still likes darkness a lot though, or i just listen to dark lyrics...
you should do something by will wood in terms of lyrics
his music should make nice images, White Noise is my recommendation
lots of words
lol
oo i hit the realism/anime wall
where my model cant decide wether its anime or realism
with context size limits, having a lot of words is an actual downside 😦
aw hec
You fill your head with thoughts you find you can't even feel Try to make room in your skull but it's full of them All of the things that you think and think about thinking I know it's hard But they're not who you are They're white noise
too wordy?
White noise full lyrics:
Person listening to music in a room with speakers, abstract artwork on wall, empty chairs, bright light source
but next gen is :
A man sitting alone in his studio listening to music while surrounded by empty bottles of alcohol. He has a laptop open next to him displaying notes about the track he is working on. There is also a guitar leaned against the wall behind him. In the distance, there is a window overlooking a city street at night.
niceeeeee
all them JD bottles
if you want depression lyric inspriration Will Wood has plenty lmfao great singer
oof. bigtime alcoholism
the very wealthy kind
what model is that?
hello there @wispy canopy
hey man, glad to have you back!
they got the whole tumor and 1/4 of my small intestine
DAAANNGG ❤️ who wants the apple... it's not cursed, i swear
girl holding an apple against a dark background, symbolizing temptation and knowledge
oof, your intestines did already get so much punishment
Me if i had money
yeah but I'd rather just lose some organs than die
hmmmm
When the hentai model doesnt do all the hentai as it was created for
actually v1.1 was actually mixed so it could do other things better again
v1.0 was a total drunk accident lol
It was a glorious tale
today or tomorrow I'm going to generate like 500 images of hair styles for men and 500 for women, pick the best 150 of each, read up on how to use mbw with loras
Give eface some alcohol and supermerger. Miracles happen
lol, trying to keep alcohol in check tho -> living alone is a dangerous recipe for alcoholism
Fucking felt that
that should make it easier to add the hairstyle you want
The only thing keeping me from drinking anymore this year is saving up for 3060 for ad
Sd
Lmfao
SD saving people from alcoholism, only to make them get another addiction
Which then powered by alcoholism becomes too powerful
unfortunately, i'm too well payed, i could afford both a used 3090 and then still a couple of litres of liquor and still have money left over...
(not that rich, but certainly not in a bad spot)
Ah yea im disabled so gotta make every penny count unfortunately
yeah, that f'n sucks.
Esp with canadian inflation rising every year
although, i've been most of my work from home lately
May the 4th be with you all

oh yeah. it's that day.
Had this fucking locked and loaded
if the last movies weren't crap it would be ok... but what the fuck was that?
Star trek wars
I can't drink alcohol anymore as it could affect my stomach sack's surgical scarring from my 2021 surgery
I love how my LLM prompter goes for a single theme in the song and then goes with that, a single song can have multiple outcomes
Cant wait to use llm on gpu
same song (within temptation - paradise)
yeah, cpu is mighty expensive
but to get the best out of llm prompts you'll need 13b models at least
the 7b models are ok, but they can't do lyrics/SD prompting very well
Wont be messing till oct when i get 3060, chances are nee models will be out by then
So am waiting
Bestfriend got me wow for my bday this month. Ill be occupied easily make time go quicker lol
🍎🍎
reinstalling torch and xformers 😭
i'm updating them basically
i got a warning in console
with the args to launch and reinstall
when LLM model stucks on keywords, it just keeps repeating words it associated with the prompt:
it was going to put cataclysm next.
for sure
girl holding a picture frame, person inside the photo, abstract artwork, sad expression, blue hues, melancholic atmosphere
it just had catastrophe, and usually doesn't repeat words
Ah insee now ok
girl in white dress walking through a field of flowers, sunset in the distance, hopeful expression -> Sawano Hiroyuki - Avid
the actual album art: lol
ez 10/10 points
I just get portraits 😔
probably picture frame + sad expression that does it
Looks beautiful
cursed stuff = good stuff 😄
I just realized my model can't do dragons very well, or I'm doing it wrong
never tried doing dragons on mine
I'll see what i get
very cool actually!
i like that
my model can't actually do anything specific well lol
this
i like it's instability, it gets creative output 😄
I'm gonna train a dragon lora (or find a very good one) and merge it into my model
thats fair, the choatic randomness adds a sort of artistic flair
oooh... porcupine tree - Anesthesize
Man sitting alone at night, city lights, streetlights, empty road, car driving by, neon signs, darkness, sadness, loneliness
I remember seeing a dragon lora on civitai but I would always get some pretty weird shit whenever i tried it
once i get my gpu i wanna do a WoW lora lol, ima take ss's of each character creation screen option and caption them myself to not get tagger giving them weird tags
i have no idea how to get stuff in and or out specifically
i do know what i want removed, but no way to do it i suppose
not terrible
trial and error, you can utilize the discard weights or my approach is i make allot of different version merges with the weights altered and i test
my model has like 100+ iterations rn cos im never sastified with it lol
nice, may I have the prompt?
dragon
just dragon?
yep
no neg prompt?
bahaha
hmmm?
neg I'm using this. Don't know if the neg embeddings of aurora still change anything since I already merged so many shit together already. I've been using this for a long time alrd, never tried changing it
(EasyNegative, ng_deepnegative_v1_75t, KHFB, Aurora:1.1), (extra fingers, deformed hands, polydactyl:1.5), sketch, ugly, huge eyes, (text, logo, signature, artist signature:1.2), monochrome, worst face, (worst quality:1.4), (low quality:1.4), (bad hands), (extra digit and hands and fingers and legs and arms:1.4), (deformed fingers:1.2), bad hand, (holding:1.6)
lmao
uuuh, what's the easynegative etc? Embeds?
oh. my model does 1 thing VERY well: accept loras. Chibi dragon suit using that dragon lora:
what is easynegative? an embed?
yeah, easynegative is an embed
it promotes ecchi though
that's fine, what embeds is in his neg?
Having a hentai model promotes ecchi 
i wonder if people realize DPM++ 2M Karras has unnoticeable differences from 20 steps to 15 steps
shhhh.... it's a creative model 😄
i do everything on 15 steps
lol. my model begs to differ
it even has differences on 35 vs 50
the other 2 you can get from aurora page
SDE++ is the best conceptual sampler imo
Ah creative that's what we're calling those things on her chest now 😂
functional.. ehh... armor. yeah. armor.
thicc
oof... doing that without the lora tho 😦 (something something creative model taking creative 'liberties')
wait. my model does this without lora?!
thought it was wearing thighhighs for a sec xD
aurora my beloved
i can't get aurora to work...
my prompting style doesn't seem to work with aurora
ah
have you copied the base prompt
probably not
what's kfhb and aurora?
aurora neg embeddings
aurora = model by one of the luna`s
ty
with your neg prompt, the embeds and "dragon"
I need to find out what those easy embeds do, cause they seem to give the dragons human features
yeah
is there a negative embedding that doesn't change the style? (apart from deepneg)
Doesn't really look like a dragon but still pretty cool
switching to my regular deep neg
(image 2: hah)
regular is dark fantasy dragon
nothing else
gonna start adding things
Nerf's negative embedding was trained to not affect the style much https://civitai.com/models/56519/negativehand-negative-embedding
I heard there's a way to use mwb to train such things without affecting anything else
bad_prompt_version2 and negative_hand both from Nergun3 are pretty good about not changing the style much.
angry kaiju fanfic XD
giant sopp by accident...
oh no, the usopp is back
this looks awesome
arachne
as prompt
just that?
ye
small comparison, using aurora negatives
sad fangirl noises
WOW
oh and optimus primate because i seemingly downloaded that one but never put it in my models folder, so it was saved from THE PURGE (accidental deletion of models folder)
these dragons look cool can it do night furies?
ever since i realized how good tiled vae is, i've been making images with 1024 in one of the dimensions
my model has issues with correct dragon anatomy, where they have a larger heads
they either do the head right
or the body
what's that?
optimusprimate does it best? 😮 i think aurora doesn't like this prompt as much
what effects do drawn and illustration have?
please don't ask how i found this old ass message
imma be honest, i have no idea how it affects the pic
illustration is conceptual, the composition

It was one of my old prompt and was just reusing the same format
or cariacature
I basically just recycle the quality tags that I used to use on NAI
Not sure of what illustration does either
but I still use it sometimes xD
❤️
same same, but sometimes i like to try genning without some quality tags, to get art out of my normal gens

NO MASTERPIECE
Yes masterpiece
Drawing is more base, illustration is more detail and is umbrella term for many styles
that's why my model = best model. no masterpiece!
(masterpiece:9999)
that's just picasso drawing xD
technically, masterpiece

Ffs
looks good
This is art
REAL art, not that pussy waifu-simping
why use masterpiece, you make the goddess venus sad

guys, invert your prompt and negative prompt, it will be fun
Yw
you want deformed nsfw?
I hate you
i haven't actually tried
is it really that bad
it's... glorious?!
uhmmm....
why are you negative prompting intestines?!
i'm gonna try too ig
oh wait. my style was on.
this is bad

WTF
this is nsfw LOL
if i told you id get in trouble
lmao is it
biblically accurate angel!
DO NOT BNE AFRAID
turns out when you do this and you have nsfw, skimpy in your normal negative. You get nsfw outputs... 
i also had nsfw and nude
"i like to live on the edge"
some people
dm
x6: i seriously have no idea what the F my model is even trying ot make except... a crappy pokemon game?
Funniest part is that I actually interrupted my work to run it.
dm. plz
now, i will join both negative and positive, and post them in the negative AND positive!
is that a real 3rd leg? XD
that went as well as i expected. this is all auto-destructive art
I came for beautiful images
…well, its in the eye of the beholder as they say.
Idk why but I feel like someone would buy this
THIS IS ACTUALLY SO COOL
It is?
yeah thats kinda nice actually
was the best result tho, the rest was just incomprehensible ... perhaps that's the value of it tho
thats cool lol
make like 150 of them then make a lora
profit
call it "fucked up negative diffusion"
name it "incomprehensible cosmic horrors"
im just imagining mashing the chaos with the coherency lol
That kind of lora/checkpoint exists tho
girl in front of planet venus, goddess, ethereal, celestial, cosmic, nebula, abstract, -> yeah, it should be cosmic horrors as that's what i was kinda prompting xD
dragotaur on img 2










