#🍥|anime
1 messages · Page 207 of 1

interesting
My carriage is advanced
not what I had in mind.. but almost
bing ggot closer
to what?
woah...
ws by bing
fill in the miku

when @nova remnant is around 😏

I had to double check to see if kidneys were against the rules
i don't wanna check cos lazy, so are they? 
no

i mean, not specifically
I suppose if you made them lewd enough it might get you in trouble
do not lewd the kidney
I think I'm happy to not have seen it

I'll generate some dragon girls so I can clear my mind
Good plan
👍
it wasn't a picture, just someone posting a link to the rules channel and the word 'kidney'
am i doing dragon girl right?
silly pony trying to lewd the dragon kitty..
smol
The tail mixed with cat looks funny
Ohh, I've got some interesting results with Bing aswell few days ago
it mostly generates a cat tail, but i was trying for more dragon
or a cat tail with scales 😄

Goodbye dragon ball world
bing create subs?
Yep, but sometimes it misspells the text
momdal?
Just had NAI create a randomized prompt for me, added some artists, and hit generate. It's just 1girl.
deym so its NAI...
sadge
1girl
1girl in the style of Masaaki Yuasa
1girl in the style of Yutaka Nakamura
Full prompt is: [artist:ciloranko_2020], [artist:ask_(askzy)], artist:wlop, artist:hiten_(hitenkei), artist:ningen_mame, artist:manuga_(manuga430), artist:modare,{artist:tianliang_duohe_fangdongye}, [artist: fuzichoco], 1girl, scenery, dog ears, dog tail, tareme, short hair, braided ponytail, orange hair, freckles, hairband, pink cocktail dress, {lipstick}
Everything before 1girl is artists that I added to it.
in style of fuzichoco, 1girl, scenery, dog ears, dog tail, tareme, short hair, braided ponytail, orange hair, freckles, hairband, pink cocktail dress, lipstick
kinda generic
except for certain elements
anime, howling wolf, moon.
anime, howling girl, cat ears, moon, from below,
But cat instead 
btw SD3 will be mostly SFW they have announced
But it will be available for people to train and stuff with it right
idk how that will work but the base model will be somewhat censored

W-who would ever gen nsfw stuff.. that's for people like @lyric zealot I'm innocent cat

i gen nsfw all the time, that's life's art to portray feminine beauty
you might be shocked to see my civitai page 😆
surely 
lol steven, how did you get that meme at the right time, i mean how do you guys do that? do you have some kind of organized image catalog to display the right one at the right time, im fascinated
and to put gravity on the matter, there are gazzilions of meme on the internet
i just press the gif button on diskord 
oh you deescalated the fascination meter real fast lol
im bad with memes

Morphing between two images
Where ever I go I can't escape it
I can improve this
one sec
NOO why u blocked me
@finite herald Do you have plushie like this?

Who?
oh its just discord being discord
I don't speak that language what is that?
discord servers that we are both in
I was just about take image of it and then it crashed
same
yeah, no server in common, we are connected by love lol
Who doesnt love chibis
for waifus!
for shrek waifus*
No servers in common but I will always live in your walls
obama chibi
no
Wait, i kinda need a community mod to approve a picture for me
Can i post gens that contain guns ?
only if handled by good looking waifus
There was Yue hiding within your popcorn
nah, it was miku! she's everywhere!
Suspicious pose
They are basically the same
ctrl c ctrl v
🤷♂️ not my fault
regional prompter tends to mirror/duplicate the regions since it's all the same seed with slightly different prompts
I could be more specific and get something different but that requires more work
It’s been two days since I asked this question so… ||SD3?
||
Bruh.
It's not gonna big a big secret. When it's out, everyone will be talking about it.
If you can’t tell, ||its a joke||
If you can't tell, ||it's annoying||
Oh I can tell you are annoying
Yup. That's me. 👉 😎 👉
sd3 when














i'd let her eat me
👍
time for bonks
@lyric zealot help
Hi meow.
Good job getting bonk
Reimu woke up and chose violence
does any 2hu ever wake up and not choose violence?
very nice
joining forces
love the style
ssambatea, abmayo on NAI. touhou is probably also having some effect on the style.
Bnuuy
o https://civitai.com/models/340631/min-lona-loralocon-or-style-or-sdxl-pony imagine my surprise running into this 😄
SDXL LoRA/LoCon for Pony V6/Pony merges, trained on Min Lona's style. Works well at 1 weight. If you want to support me, make comissions or request...
Yeee, I made it a while back but hadn't uploaded it for some reason xD
a what
||open her mouth wider||
anime source?

those horns like ....
hmmm
no idea
wym?
source_anime
https://new.reddit.com/r/StableDiffusion/comments/1ba97v6/emad_sd3_possibly_sd3_turbo_will_be_the_last/ i found the source
you are right about that statement
My bad
emad does plenty of shitposting. I wouldn't take what he says at face value.
true, and it appears that he was just bragging
from that linked comment
so basically saying it's so good, there's no need for another
goodnight 
Yeah, that's what I assumed he was trying to imply with the first message, which I suspect will prove to be very much untrue. Look at LLM's they're more mature than image models and we're still getting plenty of new ones on a regular basis.
well, it's definitely gonna be the new standard tho, or at least i think it will be
gn 

modal?
Yeah, that seems likely from what I've seen so far. I think it's still going to need anime training to be worthwhile for us to use for more than generic 1girl gens though.
flatpiececore
maybe some people are doing their anime training already
xl or sd1.5?
1.5
it will be interesting tho, when it's out to public, to see how long it takes for midjourney and others to train their SD3 closed source model
I'm not certain that Stability has even finalized their own version of the checkpoint yet so any work that might be happening now is probably just finding appropriate hyperparameters to use for training once the model releases.
ahhh, that's true
But they might actually have a finalized version as well, since they said they are gonna push to some for beta testing soon
unless that soon is like 3 months
that doesn't mean tho they are not going to be open source, also about that statement i think he is bit overwhelmed with what they have achieved with SD3 for now, AI has a long way to go still
Yup yup, read further down, where me and nubby talked
yeah i see that now, i wouldn't think sd3 is the last frontier with txt2img by any stretch
however im excited to see what we can do with it
Well, we will just have to wait and see I guess
i also want to add about sd product by comparing sora and sd video, it feels to me sd video is not parallel to sora at all
sd video seems like taking a static image and sliding it across frame, more or less.
sora on the other hand animates the character
one of the reason i never bothered to configure sd video on my end seriously, i tried animatediff but wasn't happy with results
Isn't sora like basically a video
Incomparable, from what I've seen at least

Imagine these other options are making a gif while sora is a whole video cx
yeah sora is txt2 video and does awesome job at it .. nothing exists like it as of yet
Wtf
Real humans?!

ofc that... and i know you moved your fingers and typed that... everything we are creating with technology the basis of it all already exists organically
humans are playing a race with nature 🙂
but honestly there are things we are not able to grasp tangibly and technology can help with that
@native halo using technology and be able to create a beautiful subject like sweet girls, how awesome is that? its the illusion of creation tho
yea it feels great,even more when u spend like 1 week animating something on blender and then no one buys it 
oh yeah, the economy sucks
its becoming more and more evident this economy shouldn't have existed
lets assume the focus on all the pleasures of heavenly concept, the abilities of god concept, human race is ultimately aspiring for that with everything we do
or is it the multitude of god's presence trying to return to the source? 👀
i wonder when pleasures and power merges in singularity for everyone and everything
but i know the guiding desires for that is embedded in girls within this reality
and guess what .. girls don't generically have ego trip, how intrinsically sweet are girls
No anime allowed

@vital raptor big dragon kitty 
no
yes 
I think you're being mean
Maybe you need therapy

i think that would depend a lot on who is actually giving the therapy 🙂
i once visited a female doctor and just from talking with her few mins and how beautiful she looked my physical symptoms of blood pressure disappaered
Lmao
You are something else
lol.... im waiting for steven's next meme to show up
idk i dont like it or use it 🗿
lol ok
i mean if it's following prompts very closely that's cool but i dont understand the fascination about its actual image
for me it's the flexability in the image, it's well trained on tons of artists
gimme gimme dress
prob because of its big dataset so its a good base to train on top of it,like when anything v3 came out
"tight dress"
ahh that's very understandable
Would you imagine
She is disappointed because no gif 
i see, but i think steven pointed out well how people are drawn to it
in reality, tbh, it's the single best hentai model that exists, that's why it's so popular
Oh you wanted to have her dress? Okay I can gen without
👍

Yes idk
Black clothes, pink clothes to give a mix of color
Bare shoulders to see shoulders...
Goth cause why not
The rest is non dress related

And best pony model

sleeveless tight dress, tight skirt, off shoulder jacket? thry it
one button prompt extension 🤠
i usually don't prompt for clothing unless it's just something basic like shirt, shorts it's generally more interesting to see what the model comes up with
And then you make "dragon girl, from below, from behind" just to see what comes out of the model, really scientific test


that's the other thing pony has going for it, poses, hair styles, clothing, etc.. it knows a LOT of them
@junior quest more jacket
biggest downside for me is that pony isnt as "creative" as some other well trained XL models, especially when it comes to backgrounds or landscapes, but it does ok
they'll fix that in v7 ? hope
needs (latex:2.5)
where off shoulder jacket?
Too much of it's raining data must be characters on simple background

dat jacket is goood
i think it's just that it has 0 real images in the dataset, it's ALL anime
hanging jacket? no idea how to do it?
some types of clothing and background details is other models being trained/merged with realistiuc models
It does have
It can do realistic but not that good
@junior quest towel off shoulders lmao
jacket off shoulders
There are no public well trained XL anime models. Pony is sadly one of the closest. Others are mostly overfitted on a few dozen gacha game characters.
I'm aware
lol
I miss 1.5 clothes 
i think i somewhat managed 1.5 clothing look with my xl model
and faces
one of the great things about SD and all the open stuff is that you can just change models if one doesn't work, which one is better really just comes down to what you want to make
I know plenty of people that still use 1.5 for 90%+ of their work because it's exactly what they want
my personal requirements are basically "cute girls and reliable prompting"
SDXL based model with "cute girls and reliable prompting" as a prompt 🙂
😄
if you're still on SD 1.x, the jump to SDXL will make a giant difference in prompt understanding and quality
oh i did make the jump already for prompt coherence and in certain quality improvements
yeah generally the prompting is better, but the quality depends a lot on the models
i was pulled in and pushed away from sdxl for 2 reasons, pulled in for coherence, and pushed away for image quality that were pushed so far into realism that didnt feel real
but over time that changed and more and more models came out to address the actual quality of the images that felt more organic
I was the same when XL first came out, the first models were bad, and it was super slow compared to 1.5
I kinda have the same feeling with SD3, everyone is getting all hyped about it, but no one knows if it's actually good, and the first models probably won't be great
if i had to include performance that'd make it reason #3 .. but now with lightning, turbo that has changed drastically for the better
eventually it'll probably be the new standard, but not right away
im already skeptic about sd3 hardware requirements and im not very keen for its release as i was with sdxl back then
sdxl launch for me was a disappointment with all the initial lackings
i tried to dent emad's excitement talking about sdxl then by replying to his twitter posts how much i disliked sdxl .. back then
i wasn't sarcastic tho, i was pointing to the obvious
hopefully the hardware requirements are like anything else, and it's just lacking optimization
one thing i've understood they will release several models with different parameters, so i guess there will be something for everyone ranging on vram req
yeah, makes me wonder how that's going to affect training stuff like loras and similar
we're going to end up with a flood of low end models that run on cheap hardware because that's what most people have, so we'll have very few higher end models using the full capacity
you know about x-adapter for universal lora, controlnet, embeddings applicaiton across sd and xl models, im curious if they have already figured something out to include sd3
well, hardware requirements, you know...
i have a 4090, so that doesn't bother me 😛
yes but thats on higher end, there are lower spec versions too
she is waiting for SD3 too
i feel the same way when 8gb user tell me something similar about my 12gb card 🤪
going ghibli, again
btw, that's simply adding "(((flat 2d cel shaded studio ghibli style))) "
i've been watching some of lykon's posts on twitter, he was cooking real cool stuff with sd3
but honestly.... after hearing official remark about sd3 being mostly sfw i felt discouraged to use it with the same level of enthusiasm
In early, unoptimized inference tests on consumer hardware our largest SD3 model with 8B parameters fits into the 24GB VRAM of a RTX 4090 and takes 34 seconds to generate an image of resolution 1024x1024 when using 50 sampling steps.
34 seconds is painfully slow at this point
well, SDXL was quite censored as well, and look where we are now (well, don't look because this server SFW only, but still)
34s would be a bit painful, but we have turbo and whatever now, so i expect SD3 to become faster as well
i hope you are right about it lol
i'm pretty sure every model has been 'censored' to a point
that's why we do custom training and merging
well, there's whole communities dedicated to creating NSFW stuff... cough unstable diffusion cough cough
hoping they are not hardcoding anything to limit training
i don't think they can
we'll see. interesting times in any caswe
cascade was supposed to be 'censored' and i got booba in about 2 minutes of prompting
you just have to think around the keywords they remove 😄
or train them back in
i thought you could run any model with the cascade extension on a1111
when i tried the cascade extension i used my regular models instead of a base model
i haven't really played with it, i tested the original release just because i was bored, it did really well for a base model but not good enough to actually use
i haven't messed with it since
i think my interest shifted from it btw now that's sd3 is right around the corner
i currently have just 2 xl models on my drive that i use
and both of them are lightning
does that improve things, besides speed? I'm still with normal SDXL, no turbo, no lighting, because i was under the impression that that drops quality visibly
speaking of... "(((flat 2d cel shaded studio ghibli style))) lightning turbo cyberpunk city "
speed is spectacular due to low steps and cfg but the quality is slightly affected but nothing too gross
to give an example ...
well, 5s for a 1024x1024 image at 30 steps dpm++2mk is totally fine
without turbo i mean
this is lightning model with steps 4 and cfg 1 #🍥|anime message
not bad at all
i impressed a good chef friend of mine with it lol
but when you draw human characters it feels like there is little loss, tho im not sure if its due to the xl model design
this is also with steps 4, cfg 1 #🍥|anime message
it feels to me when drawing fingers things could have been much better, but perhaps its a lacking in xl
another, a little up close and fair amount of detail #🍥|anime message
does it follow prompts at all with cfg1? i always go with 7 by default
oh yeah good question, cause how cfg worked was the higher the value the closer it adheres to prompt, but with lcm then turbo and now lightning those rules are broken, and cfg-1 gives you pretty good results
here is an example of that fantasy art, dreamy cottage, water stream, colorful flowers, pebbles, bush, setting sun.
all with cfg 1
the resolution is at 896x1152 and takes about 2 seconds to generate an image
but w/o hires
that's impressive, nice
i guess i'll wait for SD3 and some iterations of that, but good to see so much progress
yeah there's been good progress, you understand why I don't use tons of different models, sdxl is quite versatile by design, and with performance upgrades its easy to stick to a few selective models, and sd3 is probably going to narrow in on the precision even more
right, fingers crossed - and same here, I very rarely switch models anymore since SDXL does it all
mandatory image, too much text otherwise
also, obligatory muscle girl
totally understandable
new sdxl you mean turbo or?
ponyxl and autismmix and those derivatives
I'm using a custom mix by @wispy canopy now
not enough muscle
ahh ok
0/10
There's the culture im looking for.
as long as it's not excessive i'm ok with it
yeah
I guess if I were more muscular, I wouldn't mind it as much 😂
not a factor for me, i'm not in the greatest shape but that doesn't affect my preferences
Crab 
I post quad mommy combo, just trained a new style that took a lil bit of elbow grease to get to train right
a basic prompt like a hentai girl has her clothes off
even with nsfw in negative
opposed to anime girl outdoors
obviously the model understands hentai word well
Her tail is just like me
Goes up

Can you try this some time, curious to see if 1cfg follows it all cx
1girl, cowboy shot, looking at viewer, blonde hair, seductive smile, ponytail, hair between eyes, blue eyes, red lipstick, nurse outfit, sitting, leaning forward, hands behind back, white shirt, button up shirt, black jacket, earring, eyeliner, table, chair, lamp, painting, flower pot, brown skirt, thighs

i can try but i have doubts sdxl can follow all those detail
could be casue of the model im using... not trained enough
no tail, no nurse outfit ... and some other detail are missing
no cherry picking
not bad
i didnt cherry pick either but they were missing detail
let me try and edit the prompt to reflect on the key elements
nurse outfit and white shirt, button up shirt would contradict btw
what's excessive?
still SFW-ish
probably as well - tell me if you disagree
1girl, blonde, ponytail, hair between eyes, nurse outfit, sitting, hands behind back, earring, eyeliner, table, chair, lamp, painting, flower pot, brown skirt, thighs
well, depends on which details you insist on... theoretically SD3 is much better with that
she agrees
My prompt is like very tag based for pony, maybe not a good example
i have some nice results for that but they are not SFW anymore 🙂
good
i kinda edited the prompt to this:
blonde girl, hair between eyes, wearing earring, nurse outfit, sitting on chair, hands behind back, table, lamp, painting, flower pot
and got this
notice that none of the results really have hair betwen the eyes? 🙂
exactly
The hair between eyes is very annoying to get it seems xd
i had good result for that but using a different 1.5 model
"cute female face, hair hanging down in the middle, hair over nose"
Also as a long hair enjoyer, who wants to have it between the eyes, some psycho stuff

Now do ponytail cx
"cute female face, long hair hanging down in the middle, hair over nose, ponytail"
i can message you the pic i did with hair between eyes, thats nfsw to post here
using realvisxlV30Turbo_v30Bakedvae.safetensors [364cecf630] currently
Okay 
Is it 1.5 or xl
realvisxlV30Turbo_v30Bakedvae
I meant to ask torqx x3
top tier lamp
They still look so good, sometimes I go back to gen some views 
rpg maker ass looking mother fucker
ok, gotta jump, later 🙂
a meow
cat message sent, and im going to try hair between eyes with this xl model too
cat if im not confusing it
tbf, i forgot a setting at 900% of what it normally should, here is the normal version
@nova remnant that was sdxl, im gonna improvise on that prompt you shared
this sitting with arms behind back, leaning forward, isn't going to be easy w/o openpose
a nurse, blonde, sitting with arms behind back, leaning forward, facing viewer
lot of hit and miss
logically this prompt should suffice
blonde girl in nurse outfit, hair between eyes, sitting with arms behind back, leaning forward, facing viewer
but with actual images it still some hit and miss
Haha
It's tricky
i mean i got somc cute pics but not strictly following prompt which is kinda sad
anime. blonde girl in nurse outfit, hair between eyes, sitting with arms behind back, leaning forward, facing viewer, bare thighs in view.
there are distinct difference between anime and realism and how they treat keywords

cat girl, cat ears, cat tail, slit pupils
think i can go wild with all the booru tags
cat girl, cat ears, cat tail, slit pupils, arms behind back, white thighhighs,, <lora:offset_0.2:0.3>, score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, source_anime
@nova remnant
wild
I see what you did there
kitty
yeah lol im testing out booru tags basically
1girl, maid apron, standing on bathroom scale, barefoot, came out bit differently
1girl, maid apron, standing on bathroom scale, barefoot, came out bit differently
How does it work
you need the magic box
https://www.greataiprompts.com/guide/install-stable-diffusion-in-windows/?utm_content=expand_article
Thanks
Hi 🙂 new here 🙂
Spent more time than I wanted to cleaning up the hands. best I got so far.
Was going for basically this and this was the first result.. I tried to refine it and it only got worse, go fig lol
what model?
gift for @finite herald
which model?
A bit of fixing it up photoshop for 40 minutes and it would probably pass as real to 99% of people.
dont wanna put that much effort to fix it
i didnt care to fine tune it cause the prompt didnt work as i was aiming for
that's cool, is that sd video?
but how did you animate
oh .. i was thinking you used sd video to animate that
no no
well you know nothing is impossible these days 🙂
yea it cant even create proper blinking
lol one step at a time
i found a 14 yearl old music video which still sounds cool https://youtu.be/kKTCScoUFWk
See Nerina Pallot play 'Fires' live at The London Palladium on April 13th 2024
'Everybody's Gone To War' is taken from the album 'Fires', available on Spotify.
Follow Nerina Pallot on Spotify - https://open.spotify.com/artist/4aFx7q4WkiTWPLJNW5mkKK
Subscribe to stay up to date - http://www.youtube.com/channel/UCcd3EeqwNISVCUOAxmTGvRQ?sub_confi...
several more years and we should all be able to create full blown music video with Ai
what would sora create from this Design a world where time flows backwards, with people aging in reverse and events unraveling in surreal sequences.
sadge
sad*
hey cat, i couldnt settle with pony model, i was missing the sweet faces and 3d images
model?

A versatile lightning model. (VAE is not baked in) Recommended VAE: sdxl_vae.safetensors https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/tree/...
wanna ride?
srry id only ride a FRANXX or Aquarion 
mechs that need ❤️ to function
im really anime illiteratew
@native halo
Do you like her eyes

i mean i like anime art but i never read a single anime book
big eyes 
Anime is boring
Not losing anything

anime girls are hot tho which is why i got interested in anime

W
Don't you do it
wut?
if anime is boring then you should try better shows
literally something for everyone, it's not a single genre
Play games instead
👍
i do both 🤷♂️
i have difficulty watching 2hour long movie
I have difficulty watching 1 minute movie clips
But I watch 6 hours of some gameplay of a game I like
well yeah 1 min is too short
👍
but really imagine missing all the differnet stuff you could watch on youtube instead of 2 hour movie
Yeah
Also I rather just listen to music lmao
too many good songs no time to watch
lol thats not too bad, i have music playing in the background most of the time
White shirt and rain prompt 
Let's see
yeah, i listen to a lot of music, especially when i'm torturing my gpu
xl gen and spotify are good combo!
i use youtube music since i have premium
cool, i usually play yt music in 4k while playing video game lol
but theres an anime about a giant mech that needs girls to function,when the mc merges with the girls mech they experience extreme pleasure 🌚
ouh
that one

What even
Ugly
hmm nope
that one is so old
This is a Gundam

the original has bad animation,but the 2012 has great animation,also its coming back in 2024
mecha oni horns? need to try
where do you watch anime online?
I need a sec season for my one
Aniwave, formerly known as 9anime, is the best website to watch anime online for free, watch anime with DUB, SUB in HD. WATCH NOW! No Ads GUARANTEED!
i use crunchyroll
rich boy
i buy the japanese blurays for the ones i like 🗿
yeah, $6/month is rough
but you can watch free, there are so many sites
but i've been subscribed for like 12+ years
ouh deym
"free" comes with slow bitrates, bad subs, broken links, ads... meh
no ads
it's cheap enough to just not have to worry about it
no broken links

dunno bout subs, I can kinda merge EN and JP so fine
oh, crunchyroll is $8/month now

but online sites have bad quality,if u wanna pirate better to download BD's
Me when people talk about anime in my anime channel
almost half a meals worth 
I dun care bout quality, someties I watch 360p with upscaling (nvidia control panel 4x)
gross
it only works for animes tho, for real images dun wrk
i like my 4k

yea 4k with the new rtx video thing from nvidia 🤠
i will actually do that with my old E7 bluray but ill use topaz instead 
TOPAZ do real time?
no it takes like 2hrs for me for a 25min episode
im afraid i wont be able to watch most of the long anime
good thing most anime arent long
average series is 12-13 episodes, ~4.5 hours total

1episode 24mintues
damn 13ep in 4.5 hours is crazy reasonable
new ones are 12ep per season, ones from 2012 and below are 25eps and ones from early 2000s are sometimes 50 eps per season
People are lazy now
got it ... i think i can handle that
Soon 6 bad episodes per season
so much to experience
no,new animes will release in tiktok only and will be 1min 30 secs per episode,6 episodes only
the number of episodes hasnt actually changd as much over time, its just that back then less was being licensed overseas so they usually went with longer series to sell more dvd/blurays
lol
Didn't mean you but the ones making it less episodes per season cx
a tv 'season' in japan is 12-13 weeks, so that's what a series usually has, some have 2 'cours' making it 24-26
Good, the less the better 
just tell me how it ends so i dont have to watch it 
One image anime when
you may still need to watch for girls
Then it will be worth the while
i only watch for cute girls, i just get really good story animation and music sometimes
i wanna watch the girls ride the Aquarion and merge with the boys every episode 

the merge is the new singularity for all the fun
if im not mistaken
the abyss king has all the black rings
i have this playing casually Tales of Wedding Rings
the abyss king with black rings
i wonder if abyss king wants to marry all the girls
and he'd be a fulfilled king
but those black rings are evil
that aniwave.to is going to my entertainment bookmark, thanks for the share @junior quest
this one is w/o adetailer, i noticed adetailer as of recent takes a bit of time

@nova remnant thick thighs 
looks cool
why censor that lol
For the surprise factor

if only faces were more realistic
better question is why have censoring enabled 🤨
i dunno, i left most of the options at default

Eepy
close up of a female face. green eyes, ruby red lips, bangs.
well if we're posting realistic images..
im experimenting with face id
also some of the raw faces with the model comes out pretty nice
this one is with face id
wings attached correctly to the back on first try. nai really is impressive.
but how many times does it miss?
I've only tried once so far, but that's still infinitely better than other models that i've never seen do it right.
2nd one is upscaling, but it got it right too.
if its following prompt every time that's impressive but i have doubts it would
3rd got it right too. 4th as well, will send once the upscale finishes.
that beats all of sd models
pop
but wings look right, so that's a score
im thinking you get some of the prompts right and there are something that are off
5th is a little off, but way closer than i've ever seen any other model get in dozens of attempts.
are you paying to generate those btw?
🤷♂️
nope. i got a gift code in another server from one of the people that works there.
nice
pony seems to do fine, it's harder to get it to be sfw if you don't specify clothing though 😄
it's autismmix
oh nice. I thought I tried it on autism/pony but I must not have or I would have remembered that.
i've noticed sometimes certain artist/styles have an affect on whether or not it can even do a certain thing, but i haven't bothered to find a specific relation
Yeah, many styles get overfit on things that the actual artist draws. Like, if you use an artist that draws a ton of catgirls, you might get cat ears without prompting for them.
but that's a horse in shallow water 😛
well yeah, its ok to call them pony too
added anime and this one looks more babyish to be called pony
yeah, that's more pony like
the girl kinda looks underaged for too much fun
its something related to all XL models
and i often have to change girl to woman
that's intentional
but woman can be 18+
XL models have 2 clips, one that uses tags and one that uses proper english
hmm i see
that's over slimplifying it, but the basic gist
this is not how they work at all.

lmao
yeah, good thing i can use words like woman for the desired image impression
base XL is trained with natural language captions. not booru tags.
you notice that sometimes with the combination of anime and woman .. the female character can still look underaged
most anime characters tend towards younger looking
i know but there is a gaping difference on that with sd1.5 and xl
I mean
You prompt for what you want
Not really a difference because of SD version
well in context of language XL is giving what it should i suppose, but even with same prompting sd1.5 girls dont look underaged
yeah kind of but to give a better example if you compared sd1.5 realism and xl realism with the word girl
There are basically no SD1.5 models anymore that don't have a significant amount of NAI in them which means that the word girl is largely associated with 1girl which is used for females of any age.
so 1girl is more generally applied for females
xl treats that bit differently
xl puts age conotation on that word
xl text encoders haven't seen booru tags as much so they may lean more on the base training for things like that.
nubby i havent seen a girl lookalike this one from smol sus in XL
prompt: girl, first 3 are 1.5 models, last 3 are XL
first and last are personal custom mixes
wow.. those girls on the left are rare that i've seen with sd models
maybe i didnt dabble too much on girl with realism
the more i think on this chart i confuse myself with all the image data in my civitai page
Tentacle girl will get you

What do you mean?
i used 1girl with most of the images somehow i didnt get underaged looking girl
xl models that i've used dont come close to it
as far as anime goes
This one?
this one?
Cursed 1.5 models Cx
no way
nowae
Someone with stealthpng can get the metadata from that pic
autism mix has this level of aesthetics ?
like i said, it's model specific


