#🍥|anime
1 messages · Page 219 of 1
@green temple a less panicked one

@brittle rampart funya funya

What is the theme today
balls
belts
big belt balls

where balls?
They are hiding somewhere else

@brittle rampart droopy cat ears

how lovely!
why you punished her and offended her?
What do you mean
Because she's shy?
She looks offended
"Brother buy me ice cream or I will make "this" in your slippers!?!"
😁

Pat pat pat pat pat pat
btw. this is PonyXL?
Yes
Crap! I still collect in my pig-cap on a new video card!
no floppy cat ears!
take more cat ears 
kill him!!

more cat and fluffy cat ears and ice cream!
evil boy!

What model?
Default pony
Nice
charming, what a puppy lol
What model?
Indigo Furry mix + LoRa`s
ok
lora ?
i think this is his own model.
ok , what prompt ? do u know ?
try look it on PNGInfo tab in WebUI
👍
yep. this is his own SDXL model .
yep
our answer to you
good combination, a wolf girl and a mouse girl
haaha, I know she look like a cat🤣 , her character setting is wolf
FanWiKi say same
a character from blue archive, name is Sunaookami Shiroko
I thought she's a cat first time I saw as well
well... the fact that she is Shiroko from blue archive, Tagger told me😉
my answer to you
now better combination!
The Cat and Mouse Idols group is now debuting
is mama ok?
autismmix is not my model
really?
oki-doki.
then i am so sorry
I've already done it
oh
is she really that bad?
and use it on SDXL model?
no it isn't about good or bad lol it just reminded me of her
this is SDXL
of course. i am use SD 1.5
ah
but I couldn’t make my mouse trapped in a mousetrap
maybe if you give it less human feature?
and work from there
just tryna get normal mouse first
"get normal mouse first"!? but my little mouse girl - normal little cute mouse girl 
well you can slowly add the human feature back after you managed to get on in mouse trap
she's just like a real mouse, damn good at avoiding mousetraps
caught it
this is not quite what I would like. now it’s not a mouse but rather a bird in a cage
I mean mouse cage exist you just gotta make the cage rectangle
yes, but the question was about a mousetrap and not a cage
I don't think SD like treditional mousetrap very much
tbf I doubt they got a lot of sample img for it
not too much. just enjoying my day after a short day at work
just genning some touhous right now. thinking about playing some more Wrath of the Righteous soon.
what else would you be generating
if not tuho
what's a touhose?
hose
two hose
antenna cables?
occasionally I make 2b, holo, miku, or 1girl, au ra

touhou is like 1girl, but better 
These are all tuho
All anime
Same thing
runs away
i'm experimenting with 1.5 models out there. while depth is still a problem, mutations are less and less
really incredible
play NieR:Automata instead
naraka?
Ugly
Yes, free game
ner remids me that time i completed the training at max difficulty
at the start o the next level i already developed ptsd lol
interacting with everything scared the s*it out of me XD
Naraka is a melee combat focused battle royale with various cultures in it
Amazing game
No servers near me

I'm from nowhere
better being nowhere than interacting with this crazy world
things are going crazy out there, totally nuts
I can't find the SD 1.5 model, that is, the base model to draw, I can only find Loras in Civitai, how can I search for it?
filter on 'checkpoints'
that's remind me the first time i touched models
we only had 1.5 and the very good models could be counted on the palm of one hand lol
and the mutations lol
cursed anything.
so many cursed models come from you XD
Lol, back when I started pretty much the only models that existed were SD1.4, WD1.2, and Trinart.
they were terrible
NAI wasn't even a thing yet, yet alone the leaked version of it.
We didn't even have negative prompts in any of the UIs at that time.
i supposed txt2img wasn't that good
oh god, without negatives lol
at least, vae was a thing? lol
considering what we had to work with, we could sometimes get some pretty decent images. the big problem was that it was completely normal to generate these in batches of like 50+ just hoping to get one that was decent.
the VAE is just part of the Stable Diffusion architecture, the whole thing can't work without it. at the time there were no custom vaes though. it was all the default vae that stability trained.
and there was this other model, dark something
darksushimix probably
yeah that one!
i spent so much time genning nostalgic images with tha one
i loved the colors of that model
yeah, it's colors were really nice
DarkSushiMix was very nearly included in the original Kakigori recipe, but ExpMix_Line - V3 turned out to be a bit more compatibile with the highly cursed model that that I was starting from.
no the starting model was Toonyou - JP - Alpha 1
never used it
super cute style but it had tons of problems, fixing those and keeping most of the style was my original goal with Kakigori https://civitai.com/models/92834?modelVersionId=98960
what? i digited civitai, opened it and the site directly opened toonyou
how's that possible?
hands were the biggest problem with it. they were extremely bad. notice basically all of the previews have clenched hands
then on top of that, it just wasn't very compatible with merging so often broke things when merging to try to fix it. took a lot of trial and error to work out something that was mostly serviceable.
wow
what a pain, already from the smudged details, i can tell how much unstable that model is
nice
thanks friend
AutismMix confeti
also using a lora I made using art from gbf fire emble heros and asanagi
works well as I had hoped at making cutes with weapons and armors
i would love to train too
that stuff is too complex for my feble mind lol
yeah, i can see it
somehow i lerned how to merge well enough models
@supple raptor whats a DPO model? combined with lora or smthing?
I need to make a lora for my scifi charaters at some point
it's taking me longer then I would have liked to get moved over to XL
been busy with other shit
yeah, dpo is a lora you merge with models and gives a little more precision to prompts
hmm..so overall how is it?
or you can simply load as a lora, but i noticed is a little less effective
when we compare with the original model, are there any weakness?
fixes some smudges, i like it
i don't see why not using it
even if little, is still an improvement
DPO = Direct Preference Optimization
It's similar in concept to RLHF (Reinforced Learning from Human Feedback). Basically instead of just setting hyperparameters and letting the training run it's course, feedback is given based options that are created during training to help guide it to a more desirable outcome. Using a generic DPO lora sounds like a silly idea to me, basically that's just extracting a good model as a lora and then using it on a worse model. https://arxiv.org/abs/2311.12908
Large language models (LLMs) are fine-tuned using human comparison data with Reinforcement Learning from Human Feedback (RLHF) methods to make them better aligned with users' preferences. In contrast to LLMs, human preference learning has not been widely explored in text-to-image diffusion models; the best existing approach is to fine-tune a pre...
There's no special sauce inside a DPO model, it's training parameters have just been finetuned through human feedback during training.
it was developed for xl? weirdly enough, works on 1.5 too
They made versions for both. I assume 1.5 was probably done first as a proof of concept since it's a lot faster to train it so if the results were garbage they wouldn't have wasted as much time and money.
eheheh
Are there any pretrained models based on the DPO models?
where do i find them?
Their example models are the only ones I know of off the top of my head, though I imagine some others are probably out there. Hard to say for sure which ones because there are plenty of models that are labeled as "DPO" models that simply have had the lora extraction of that one merged into them, so they're effectively just diluting whatever training was actually done. Here's the SDXL one which has a link in the model card to the SD1.5 version as well. https://huggingface.co/mhdang/dpo-sdxl-text2image-v1
where can I find an anime dataset? Do i need to assemble my own?
Depends what your intentions are. If you want to finetune an anime model from the DPO model, then you'd want something like https://huggingface.co/datasets/nyanko7/danbooru2023
You're going to want something more robust than just sd_scripts or dreambooth to handle an 8tb dataset though.
so, the ideal would be to simply link lora to their demo models
The ideal situation would be to actually train using DPO methods when training the model that you want to call a DPO model. Merging in the DPO model (whether that be the full model, a lora extraction, or applying the lora at runtime) is not even really comparable.
oh, i wonder what setting they used for training
Imagine extracting Aurora into lora format, then using the original NAI model for inference but applying your Aurora LoRA to it. Seems silly right?
yep, that's why the results are so small
but at least don't make things worse, like lcm XD
8tb????
Yep, you're basically just diluting an allegedly better base model with the model that you assume is inferior and the result is just somewhere in between the two. If you're happy with the results you get with it then there's nothing inherently wrong with doing it I suppose, it just would have made more sense to at least train from the DPO model if that's what you wanted to do, though that still wouldn't really result in what I or (I'm assuming) the people who published DPO would consider to actually be a DPO model.
My whole pc is only 4tb
Yeah, full finetunes need a lot of images. You can could train a model with significantly less, but if your goal is to teach a model that doesn't really know anime, how to do anime then you want basically as much as you can get otherwise you're going to end up with a model that doesn't generalize very well. That's been the problem with most of the SDXL finetunes so far. Most of even the big ones have been trained on relatively small datasets which leads to things like Animagine being unable to generate most of the top 100 most popular characters on Danbooru.
That's why actual full finetunes are not very common. Well, that and the amount of compute hours it takes.

animagine looks pretty good
Can you name any full finetune models? I’m making a merge.
xl or 1.5?
V1.5.
relatively small. 3.0 was 1.2m images and 3.1 is finetuned on top of that with 870k images.
ah, 1.5 is tougher. most of them have been merged to absolute hell already. I'd say just pick models that you like the look of, that seem to obey prompts reasonably well and try merging them without worrying too much about whether they're a merge or not. If the recipe is included in the model cards, then it's usually best to try to avoid too merges with the same ingredients.
for comparison, Kohaku-xl-delta was trained on 5.2m images and used only 4x 3090s for training instead of the 2x A100s that Animagine used for both of those versions.
manually tagging millions of images must be such a pain
I have installed a 1.5 model, but when I put the lora in the folder it does not appear in the diffusion stable, could it be because of the diffusion stable version?
nah, if the model loaded is 1.5, you should be able to see all the 15 models, unless the gui failed to hook the model to a version
usually happens when the model donesn't contain metadata, or when you failed to download the data from civitai
Ok
Thanks
And what can I do?
in setting there is check to force the gui to show all the models
on the bottom you can open all the setting a with ctrl+f look for lora
Is pony basically mainly good for composition and prompt adherence? Might just use that to make characters and photobash...
Depends on what prompt you want it to adhere to, specifically it’s pretty good at character prompts like expressions, poses, hair and clothing styles, etc
It also has a pretty large set of trained artist styles so you can get a lot of different looks from the same model
If you want interesting backgrounds or secondary aspects, it’s not great at that
I see. I was thinking of trying political cartoons so the characters (like Uncle Sam for example) would need specific poses and expressions and such. Then I'd cut the character less the background into another image and with other stuff going on in the image. So pony sounds like it would be just the thing needed for this type of endeavor
yeah, should work well for that
depending on the specific poses you might need something like controlnet, but you'd likely need to use that with any model
lol
1boy, \(uncle sam\), patriotic, american flag suit and tophat, white hair, old, pointing at viewer,
that's with autismmix_confetti, a pony mix
only a matter of time now that there's API access
probably be a while before any decent tuned models are available though
Why does it do this when I rescale an image?
And am I in the correct section? img2img?
yes
thanks
usually 0.30 0.35 is the max, with a res of 1024x1024
Standard pony vs autism, what're the major pros and cons?
One pro for using standard is that you don't have to tell people you have autism.
Mix of pony with some stuff. It's an attempt at making pony more predictable and less dependent on schizo negatives without removing its comprehension and artist knowledge.
from the autismmix page
Does anyone know of any model or lora that can generate images similar to this one?
depends on what about that image you mean
something like autismmix if all you want is characters
pony based models have a lot of styles trained in them, the hard part is tracking down which one to use
evenin'
Hey omnia
Make any new loras/ xyz plots?


I'm playing game
soon

@nova remnant I was doing some beltiful ones earlier
mmm no plots in a little while... at least nothing i've kept or found remarkable 🙂
weaker than a normal horse vs you can only eat checken nuggets
the hell is that comparison kek
Tragic 😔
Tell me if you find cool anime stuff 😭
I'm really not the resident expert... I just like messing around. 🙂
Messing around gets the coolest results tho
I just make cat girls because I'm super original and creative 😄
this @vital raptor
this is @lyric zealot stoopid
why is she in a supermarket LOOOOL
Gotta dress to go to the supermarket
everyone needs to shop
To buy more makeup!
slay 😌
smol witch
Smolll

I hear controlnet doesn't work with pony/autism like it does with regular SDXL variants, is that true?
smuggy
the only thing i hope SD3 actually does is make better text
if it's going to generate text anyway, if it could at least be a real word
yeah, but they're all prompting for text, it's not random stuff
I see
rawr 
mspaint text vibes
grab
grab what 
everything
Stupid fake horns
I'll take them out

When I dominate the world that'll be my logo
Strongest weapon in modded Skyrim
😄
anything else i can get you, good sir?
i can think of a few things 😏
yes, i have those too 😛
😄
"i ask you remain decent in our establishment, good sir"
yes, everyone is interested in actual elf ears 😄
@dull jackal you awake? we got funny letters scam this time
lol, funny letters
cyrillic confuses me even more than katakana, hiragana, and kanji combined.
@bold ether not sure if you can delete/ban but sunny doesn't seem to be around and you show online.
ouh deym, dat form also?
its her best form 
she voted for trump 👍
i forgor i was going to send this one earlier
wow, that's good
yeh, nai makes local models look pitiful. I have a hard time genning on local since my NAI sub ended.
i feel you, bro
heres one from a similar prompt
very suggestive
she's kneeling over a corpse
this one is probably one of my favorite nai images
that's pure txt2img. no inpainting or editing.
reminds me so much of eien no aselia
spudorately touhou XD
i dig it so much
sakuya = best meido
this goes hard
NAI moment
this one got uploaded before I could fix the missing fingernail, but it's good too (i'm just pulling these off my aibooru)
me when nai:
or there's fujiwara
put more images so i can turn them into loras effectively stealing your images
i mean meow
just search nubby on aibooru they're all on there
put more lora so that i can merge them in my model and steal effectively your lora
charming lol
flan best touhou after marisa
jk bro
i was thinking about the warhammer mess
i feel like someone is intentionally corrupting and than sinking every francise i love <.<
whats going on with warhammer? only thing I heard recently was that female Custodes were confirmed canon.
I don't really follow that one, just have some friends who are invested.
Why do these go hard
fan's backlash.
they refused that change, because radically change the mindset of the lore.
so, the reaction of gamesforge(the publisher of warhammer) was like ceXsorXhip gaslighting and attacks to the fanbase
because NAI is flexing on open source XL models
so they walked back the female custodes?
no, and the shares of gamesforge are sinking for this reason
eh... i'd put money on it being far worse if they did walk that back.
looks like that change was pushed by people they where contracted with doing a film about warhammer
and they wanted female custodian in power armor
honestly tt games are already unpopular amongst females. taking steps to try to make them a bit more open is a good thing for the overall health of the game.
warhammer already have well inserted and deeply exposed full female factions and female characters
this impositions just waters down the dumb and obtuse virility of the custodians
take D&D for example. in 1st edition AD&D published books, all female characters took stat penalties simply for being female with no benefits. it's commonly accepted that Gary Gygax (the original author) never used that rule at his tables and didn't intend it to be included in the publication and it was removed in later editions.
like sending soldiers in pink camo
i just turned on colab to test my new lora (failed miserably) and now i have the session turning not doing anything with it cause i have no idea
any cool idea i should try out
yeah, that was very dumb XD
an arbitrary restriction on one of the canonically most elite groups being exclusively male adds nothing to the game. all it does is ostracize females who might otherwise be interested.
i feel warhammer is to captivating exactly because the characters are not good people and follow to death obtuse ideology
you're not wrong, but when it comes between that and making the game more welcoming to a wider audience the choice should be obvious.
to an extent at least.
they are not the most elite, but they are the most popular group, exactly because the prevalent fanbase is made by boys that like that
allowing freedom of race and gender is almost always a good thing in any sort of roleplaying environment
what's the point of offending that fanbase, when you can simply write more about the female factions or even make new more inclusive factions?
because there are absolutely females who would be interested in playing custodes, who want to play a female character.
what does disallowing females from the role add to the game?
nothing, but than they have to fix all the lore in way that can make everyone happy, when that becomes canon
nothing is worse than betraying the identity of a story, because this blocks the immersion of who loves that story
What does it break in lore besides "Space Marines and co are all male"?
Honest question, I really don't know.
the fact that the empire is made by a bunch of obtuse machinist, racist, radical and fondamentally all evil compassed bunch of people
and people like to read about that, not because they are right, but because they like to read about how a world of evil people evolve
Asserting dominance
they are evil, they are wrong and that's their identity
why can't a bunch of obtuse machinst, racist, radical, and fundamentally all evil compassed bunch of people clone both males and females to fill their super soldier power armor? adding females doesn't really change any of that as far as I can tell
if people want to push for such a change, such a change must be slovly gained inside the story, not changed like this
given the whole debacle of the gamergate 2 thing, forced dei, puts it into another perspective, imho
forced inclusion just because of inclusion is just watering source material down
i'm not saying female's shouldn't be there, let that be clear
forced exclusion just for the sake of exclusion adds nothing.
agreed
changed the image people have solidified of that faction.
exclusion is evil and this make the empire more evil, and that's good XD
i'm no 40k lore speciliast tho, but as far as i know, the empire soldiers were men, pretty much, except the sisters of battle (?)
not only them.
given that they're all bio-engineered clones, it doesn't really make a difference either way on the good-evil scale. literally the only thing it does is ostracize a portion of the playerbase in favor of a different part of the playerbase.
honestly i love to read sometime some pieces of warhammer, but me too i'm not an expert
anyway, as anyone who has an IP with a pretty hardcore fanbase, 'retconning' your own lore is never a popular thing
i was more focused of all the religious idea behind warhammer
i'm not even that interested in custodian
Are you talking about warhammer?
yep
Is it bad that i knew that was warhammer based on that one paragraph?
and that's all the point
no, that's the rot of all warhammer lore XD
a bunch of bad people
the only thing i really know about custodian is that they fight literal gods, just equipped with faith powered obsolete weapons and fanatism XD
at the end of the day, warhammer is a game first and foremost. they make the vast majority of their money off people who want to play, not just people who want to read the lore and from a gameplay perspective, telling a player that they have to choose between the gender they want to play or the role they want to play is absolutely always a bad thing.
art is there to fuel the imagination of people.
everyone is free to make their own story, but what is canon, should be clearn from start to end
when the initial canon choices are bad, they should be fixed. idk shit about how they went about the change, maybe that could have been handled better, but the change is unquestionably a good thing for the game.
even if they added space, they are essentially a virile, dirty, rude and violent confraternity of marins in power armor
that's why the counterpart is called "sisters of silence"
confraternity are made by males and sorority are made by female
in the end of story, as you said, that's not too important, what i really think is that the fanbase fears an infiltration of political propaganda in their franchise and they are essentially saying, back off or we will sink you
this, probably
honestly, the fanbase is shit if they're going to leave over that unless the introduction of females was just absolutely trash tier, in which case the quality of the introduction should be the complaint, and not the change itself.
as DEI tends to become an issue in some games
even if thrash, that's the fanbase that elevated them. without that fanbase, there wouldn't be any warhammer
they can change fanbase.
but that new fanbase will love and follow them like the old one?
warhammer is almost a 50 years old project.
to follow them for so long, that show something about them
I'd put my money on the people who are actually bothered by it being an extreme minority, unless like I mentioned, the execution was just terrible in which case everyone should be offended.
huh? first game was 1987. 37 years is still a long history, but it's not 50 years old.
like the joke about the girl in the last of us which i don't remember the name being lesbian.
the world is ending and even the last survivors can't have children because "duh, I'm from the other party".
seriously, at this point, the humanity should just get extinct Xd
like i said, the shares are sinking, so, i'ts not so simple
i said almost, don't count the single digits.
it's not like i have intentionally controlled
i do like how helldivers kinda made the statement -> "no extra flag for anyone"
better than making someone uncomfortable by giving space to political agendas
people join the play to distract themselves from such stuff
i love that they banned both side of the argument on the discord server tho.
"this is not the place for your b.s."
i'd love it if more companies woudl do that
yep, me too
and inclusion has a place for real. But when i read when some of those DEI consulting companies were hating on kingdom come deliverance because not enough DEI -> then i'm really wondering what those companies are trying to do?
1500's slavic had no people of color in that region... why want to even push that?
politics have inerenthly 2 sides and giving space to one side inevitably starts a fight
that's why people are tired of political infiltrations in art
if you are one side and read propaganda of one side, you inevitably get triggered
even when you are not radical
it's a dip for sure, but it's a pretty volatile stock to begin with.
let's see how things will go.
it's just the start and even after the big voices, the fanbase still loves them
it's not even a week after all
if we're being real, minis are by far their biggest money maker and people with a hundreds to thousands of dollars in their armies for wargames are unlikely to walk away for something like this.
time will tell, but there is still the risk of sinking. that already happened many times
and that's not a good thing
reminds me to the fantastic four movie remake when they made the fire guy black just because and the movie flopped
look at marvel and dcu
yep XD
every one of the fantastic four movies has flopped because they've all been shit regardless of the race of the characters 
i love the first one. laughed from start to end
a really fantastic comedy
the one with jessica alba wasnt a flop instead it had her wearing a tight latex suit
20 Century Fox just needs to hand over rights to the franchise to Disney. At least Disney can make a watchably shit superhero movie.
they flopped because people can's appreciate funny stuff
well the 2nd turned galactus into a giant cloud so ofc it was gonna flop
i couldn't stand it, so i didn't watch it
that's a flop
most films don't. only the shit ones.
because they focus in regain the losses in side stuff, like toys , clothes, publicity ecc.
that's why even after a flop, they don't stop
films are a very pricy way to commercialize a francize
and looks like it works
even when films are a flop
silver surfer was an even bigger flop
looks like thats what happens when u take a dump on the source material
yeah because silver surfer was somehow an even worse movie
i think the real question here would be,the females they are adding are they hot or thats forbidden too?
they should be as pretty as the custodians -> pretty much roided up hyper soldiers that basically resemble nothing even something close to a woman
that's my take on it
they should make them like Baki characters
you can hear the outcry: "BUT THESE ARENT WOMEN"
they are, and they're custodians. they're just not your kind of woman. that's a whole different issue
idk i have seen plenty of roided women in the sport they are huge so theres def women who look like that
is anyone in warhammer's universe hot? I haven't seen anything to indicate that any of them are unless you count the unhinged fanart.
pony hammer is a thing lol xD
these are the only forced identity politics i subscribe to
yes if u are into roided meatheads🥩
and the imperial soldiers in warhammer should resemble the emperor, so virile, bulky and a ugly snout on his muzzle lol
lol, this convo and samus reminds me of the whole "samus is a girl?!" thing
when even the first metroid already showcased it (justin bailey -> people here who used the code too?)
yeah, i loved that reveal so much
i remember when they added samus to smash and ppl were bitchin cuz she was wearing the bodysuit
was so funny fighting against samus zero suit
well, samus with bodysuit is significantly weaker than samus chozo battle suit
lorewise...
i agree the bodysuit is useless,just remove it
samus is just very resourceful, like a true bounty hunter
yeah, comparing samus with warhammer is a funny idea
samus intentionally laked so much an identity, that was totally meaningless who was under the armor
nintendo is very good in creating a character as a blank slate, but still make it lovable
but even from the first moment/game -> it was clear she was female
very low pixel female, but female nonetheless
i didn't know and didn't care, till i saw it on sness
on nes, i didn't have the brain to care
the good old days 😄
at the time i was, dah, platform, jump, good, i'm good, missile
dah!
yeah, i loved the good old days
back when we wouldnt notice bodysuits because of like 8x8 pixels 😔
i did it and still do it because feel's right
back then this was the sexiest char we had in games u would see tons of ppl on arcade cabs playing her 
that warhammer stuff reminded me a thing
the nazi had that fetish of tall, bulky blonde with blue eyes fetish, but their leader was short, flaccid and with black hair
wtf? XD
flaccid 
for sure they never depicted him as a muscular and statuary man XD
talkin about flaccid austrian painter reminds me of the "sex with hitler" game on steam
mai ❤️
i don't want to know XD
i should prompt this in my new workflow...
the weet dream of many kids
you know, i'm pretty satisfied with my new workflow ^^
yup, that's on a pony model
what's so special about this workflow?
nothing too special
just applying lightning lora, PAG, freeu, dynamic cfg, and custom scheduler (wrote my own)
oh, that's good
mornin'
I like this one 🙂
oh right,i should try that prompt again, now that i've dialed in the new scheduler
basically, i'm using that workflow to try to get as much leverage from the fact that lightning lora can help a lot with image coherence
so instead of using it for speed, i'm trying to use it to get generally better images
oh no, you made it obvious i'm not denoising enough in the upscale phase 😮
oh, that's good
whadya mean?
making it make BIG steps early on, and then leaving a lot of small steps for later, making sure every small detail is actually right (looking at you, fingers)
ther's some visible dithering on the lines -> the denoising isn't strong enough, so i'll have to increase it, but that's actually fucking with this formula 1.5*max(sqrt(x**(sqrt(s*2))),sigmin)
it's just me, to prefer turbo? it's still fast and less sacrifice in quality
but i admit lightning is way faster
using 2 different ksampler?
i think i used pony turbo as the base model for my merge
same sampler, actually
but vastly different schedules
they made a ksampler that can change settings half way?
pretty nice, now it's less blurry
basically, i'm just doing it the regular way, but i feel this way, pony does a LOT better backgrounds/details then using a normal scheduling/cfg mehtod
yeah, i totally agree
i mean, if you told me a month back that this comes from a pony model, my first question would be -> how did you actually get it to do backgrounds reliably
do you know kritra is compatible with comfy?
tried some similar in the dallster...
they are good
dall3 being very good on the compositions
i'd love a finetune on pixart sigma, but alas
interesting.. now that i think about it, backgrounds have been pretty tough to get in autismconfetti
ikr?
so, it wasn't just me
yeah dall3 is my cheat code... it's my favorite of all the various offerings out there right now by far...
pony is terri-bad at swords tho, lol
i agree
1.5 is worse but sometimes u get a lucky gen 
do you seriously expect for pony to handle swords... with hands?
they should hold the swords with feet
"YOU UNDERESTIMATE MY POWER"
BURN THE WITCH! 🔥
yeah, that script is gold
lets try it on actually autismmix
see if it's the model merge i did, or the cfg
@steady grail yup -> my model merge was a bit succesful it seems, autismconfetti isn't as good at holding things
might be unlucky seed tho
autism vs my merge
i feel my merge knows just that little bit more on how it should be held
awesome!
see? he was a rogue, told ya
dual wielding
cool
Woooow
Neat, What model?
anything XL, with some style prompts
ok
man nai so peak. wonder if therell be anything that reaches its qualitt
another good one with the same prompts
幽深的巷子
jeans
dalle3 is nuts... I need to figure out how to replicate that in SD...
and at widescreen too
dalle doesn't like the beach, but i at least got that one
i like sd more
I still like 1.5 looks more

trying to get widescreen is so hard sometimes 😦
mid images are back
I feel u man
yknow i got lucky af
i made two accs when novel ai was still doing novel shit
I was more of a Shermie enjoyer
👍
and so got 4000 atlas for free
altho i did make someone wonder whether my works were nai or not when they werent

Maybe leaning forward is what you want
oh they are leaning forward alright
turns out bent over and jeans make for some really good images
😭😭😭😭😭 wtf happening to my sddddd
my dream in life is to be a wall
Are the all dalle?
i believe he said those where dalle
😭 maybe tommorow problem fixes itself
It's omnia, they are probably dalle
adds up
I wish I could run dalle locally
but the model is probably huge
"top down bottom up" is the best natural pose

Lmao
My gens all get bottomless 
computer simply knows what u want, thats all
What is your prompt for that
(masterpiece, best quality, high quality),1girl,angry,pink eyes, blush, white shirt, skinny, jeans, wariza, glasses, black hair, top down bottom up,
no idea, i tried it because u wrote it
ill remove one then
ok lets see what we get
oh god
now i know what that prompt means
i think im brave enough to post a couple here

rules are ment ot be bent a little
the top ones are dalle, the bottom was sd
i'm running another sd set so we'll see how it goes... it's widescreen so they might be abysmal
Lmao
Smol dragon
Does this count ?
I got high hopes for PonyV7 considering the dataset is confirmed to be 4x larger than PonyV6 (with even more Anime images), plus I think Astralite has even more A100's to train it thanks to their "secret sponsors", other than that tho I don't think anything will rival NAI (in terms of base style and mixing artists, cause that's the only thing it's good at imo)
Smolll
indeed 🥹
pony can do normal art quite well 😮
Yup, it really does make you wonder just how insane the model would be if he had more people helping him out

I hope it keeps how smart sd3 is with prompt
And not go fully into tags like pony kinda is
I did see someone mentioned that NovelAI are his "secret sponsor" but that's probably just a false rumour
The thing is natural language still works with Pony for some specific poses which is one thing that it does better than NAI imo
i made a speedpaint except its a robot that drew it
https://cdn.discordapp.com/attachments/1079501847720640682/1231279626366619739/MUF8zlO.mp4?ex=66253e02&is=6623ec82&hm=62724c3ad6057ddcc6a7d488dfeaba814ee322aa2e39af8cad66be4919ab623c&
no human intervention involved
uses sdxl for the original photo and then autoit for the actual drawing process
😭😭😭
so charming lol
i like sweet things. they warm tired heart >.<
indeed
more smol 
Animu?!

I got some pics...
Actually... wow. Too much work to wade through all the NSFW stuff...
Have some HuniePop stuff, 'cause I don't do anything else 
kawaii 😍
Smol version 
Well I'm trying anyway
the setup I'm using is a bit messy sometimes, the hands maybe aren't as crisp as they should be for instance and the carrots are all over the place if you look too closely
i`m used to the mess so np 🤣
I mostly make this sort of stuff where the mess is much less of a problem
Unfortunately it's not exactly anime, there isn't really a fully appropriate channel to post it to
Here's a chibified version to remain on topic 
cool
Hand over your soul uwu
look like smt only SD 1.5 can do
SDXL doesn't like maximalist style sadly

This is Pony
I'll give the prompt first and attempt to explain later
Positive :
score_9, score_8_up, score_7_up, otic, ssal, (mamimi:0.6),
1girl, expressionless, skulls, blue eyes, ginger hair, red horns, claws, red lips, nose, (chibi, cute), hand on hips, wings, dress,, (fangs), catmouth, smug, narrowed eyes, solo focus
Negative :
score_6_up, score_5_up, score_4_up, mgfx, zvu, simple background, sketch
Most of the heavy lifting is done by the weird tags
and mgfx / zvu in negative
Basically yes. ssal by itself does shadowy silhouettes, so in combination with other stuff it makes everything ripply and helps achieve maximalism
I really missed those style from 1.5 days
Me too, it's why I went down the hidden tag rabbit hole in the first place
Pony has pretty surprising stuff hidden inside it. The rabbit hole is quite deep
I wonder if anyone made a list yet
Of course
I don't think you put something like score_6_up in **negative ** for Pony, that would mean "6 and better", excluding higher quality imagery.
Column P has nice comparisons
thanks I'll look through them and now I shall be more powerful than ever
You're right.. but everytime I tested another score setup I ended up liking the result less, so I stick with this nonsense. Probably placebo effect, idk
Seems to add noise and ends up as grain and detail
Oh uh I just realized the spreadsheet has a decent amount of nsfw in it for example images and comparisons, imma delete
hmm I test out your prompt and we def didn't get the same result

I also have the line art lora at 0.4, I use 4 cfg scale and 20 steps
Oh yeah that's way smoother than what I'm getting
But also the fact that the emphasis on chibi isn't doing anything in your case seems odd to me
canceled image
Kind of works for me (without LoRAs)
Yay !
Anyway, the point is that those tags empasize these muddy organic fantasy-like patterns that get on anything they can reasonably fit into, so prompting stuff like "black armor" is more likely to get the look I was initially asked about
What do you mean canceled, you stopped it mid generation ?
yes
i didnt like it at the beginning
New jutsu unlocked
lol
Turned out cool
indeed
Agreed
I kinda like that
clothes prompt ?
There's none 😔
score_9, score_8_up, score_6_up, otic, dwc, (art nouveau:0.3),
white hair, head tilt, tilted head, orange theme, maple leaves, warm colours, autumn, brown trees, narrow waist, horns, profile,
Again, the weird tags are doing the heavy lifting
Yeah. But it's hard to get so many little details explicitely
I've learned to embrace roundabout hacky ways of getting stuff that doesn't have a tag.
It's fun, it's like alchemy
Is there a tag or a name for those ?
Another one of those things I've seen a zillion times and have no idea how to describe
i didn't prompt for them, but they often show up with happy, excited, or surprised
i've noticed that autismmix tends to comflate victorian dress with maid, which isn't specifically wrong, but putting maid in the negatives gives better dress variations
not sure if the same holds for other pony variations
noticed line in use
oh
nice
I feel like in her case, being noticed might mean you're in danger
I didn't want it to seem like criticism lol
it`s ok
🤣🤣 see
I thought long and hard about how to answer this but in the end I couldn't find anything, however I don't want it to seem like I haven't notice lined you back
long and hard
thank you for thinking so highly of me
👍
If I thomas face something its probably just a thomas face moment 
need thickness and more gothness
Ima cook it for you
Thomas face for me has always meant :
@normal nova
I tried to do it while keeping the white hair and clothes
But that's not very goth
@nova remnant You also get this but somehow I feel like it's not your cup of tea

The good
and the bad side
x3
rawr XD
evenin'
@native halo
not anime char 🗿
better than anime

0.35 denoising strength for upscaling, some weird stuff appear
cat indeed
just let her have some food!
which one you like? ice or fire




