#🍥|anime
1 messages · Page 189 of 1
I keep intending to play with control net to figure it out but every time I do it ends up with some weird conflict with another extension I’m using
Maybe I’ll spend the time to setup a cnet specific install this weekend
What other extension 
At one point it was tensorrt, another time it was adetailer.
Most recently it was one of the extensions I was using for model merging. Don’t remember which
Adetailer shouldn't have problems if both are up to date
just wanna ask, for your adetailer, do you usually use it with controlnet, and if so, how does it help?
like i saw you using it with depth map if im not wrong
yeah i'm pretty sure last time i installed it, it was fine with adetailer and trt, but it was conflicting with something else
i'll have to try again
i still have some confusions with img2img use case from what i'm able to do in txt2img tab along with CN
on the other hand, i havent really had a need to use cnet for anything
You can use it with cnet yes it just detects part of the image for example hand and gives that for cnets pretty much same as doing it manually
Anyone here know of a reasonably popular anime frog girl/frog adjacent girl besides Tsuyu and Suwako? Trying to find another one I can gen on PonyXL without LoRA because I don't want to mess with the style. It's for a composite image I want to make.

hmm, my model seems to discard 'anime' when i tell it to make a kigurumi
i tried asking bard, but the only decent example would be eruka frog from soul eater, even then it's not really frog based
frog suit seems to work though
Tsuyu is the only one I can think of
I tried her and sadly couldn't get even remotely close. 
what are you trying to get as a result?
I'm going to composite these two images together and would like to add 1 more in between them.
😄
are these generated with comfyui? 
since the metadata shown is in a slightly different format
yesn't. SwarmUI with Comfy as the backend.
Stable Swarm

basically it's an actual UI for comfy so you don't have to play with spaghetti
oh, is that one of those ui thingy, where you can use a1111 or comfyui for their backend?
ahhh, just for comfy
It can use A1111 as a backend too, I just wouldn't really recommend that unless you need to be able to use multiple GPUs.
@keen hollow

Miku tifa
no spaghetti? what's even the point then! 😆
You still can use the spaghetti, you just don't have to
Here's the normal generate tab. The image history is really nice. So is the automatic segmentation in prompts which you can use like adetailer but for whatever you want.
is there the autocomplete thingy like in auto1111?
the only thing i don't really like about comfyui, other than the confusing node thingy, is not having the autocomplete tagging
There's one for Comfy, but I haven't found a way to make it work with Swarm yet.
I still use both. I'd rather use A1111 for most things, but for XL, it's worth using Comfy with Swarm.
auto complete is for the weak
I know tags well enough that I don't really need autocomplete anymore, but for adding wildcards and loras I like it. Swarm does have a good UI for adding those though so it's tolerable.
I tried comfy before (like before xl was a thing) and it wasn’t very mature then. Haven’t really had a reason to try it again
but is integrated in the lora loader and works like a concat
So far a1111 has done everything I wanted and it works fine
@supple raptor is that artist tag working ?
I also got a little frustrated with the developer when he told me he would never implement a feature even when I did the code for it
using an artist wildcard is cool
With all of the extensions available for A111, it really isn't. There very little that's actually useful that Comfy can do and A1111 cannot with the right extensions.
of comfy?
Seems to be, I have no idea how faithful it is to the actual artist. I just ran some xy's on the artist list that's been passed around and picked one that I liked the result from.

The pony creator wants to remove artists tags
for his admission, he is still learning and maybe refused at first, for safety reason, fearing he wasn't good enough to check the code
Do you know how to do style similar to Soushiki's style in Pony model?
it wasn't even that big a deal, i just wanted a way to turn off or disable parts of a workflow without having to make a completely different workflow for it
I was told specifically, "this will never happen"
oh, they added a custom node for that, right?
my solution was a custom node, and only required like 2 lines of the base code to be changed for it to work, apparently now you can just select a node and ctrl-b or something to disable it
i usually just bypass a node i don't want, or simply cut the connection
at the time, any cut lines caused an error and it wouldn't generate
it had that function from the time i started using it
this was quite a while ago, i know it's changed a lot
and even if the guy is a jerk i'd use it if i had a good reason to
You can just use CTRL+B to toggle on and off nodes. Have been able to for a long time. Not sure when exactly that was added or when all this happened.
but like @supple raptor said, it's really not any 'better' than a1111 and extensions come out almost as fast as anything else, so it never really gets ahead
tasty jerky 🤤
I think people get the impression that it's "more advanced" because it looks the way it does
Btw where do I get this pony model?
now that i think bout it, some complex stuff can be done in auto, but it's more time consuming
there are some wierd things you can do, like running 2 checkpoints at the same time and merging the latents, but... meh
you need to surf the dark web 
Is this complex enough?
im kidding it should be on civitai > ponydiffusion
beware there should be a lot of nsfw previews
Not surprised Bronies would be hiding there
Comfy also has several things that make it worse or harder to use. for example emphasis on words. A1111 normalizes emphasis across the prompt, while Comfy doesn't. The ideal situation would be a toggle to choose, but in the absence of a toggle, A1111's method is objectively a better user experience because it does not require manually lowering emphasis on everything else to accommodate the increased emphasis of a single tag. Inpainting is another pain point in Comfy, but I'm not going into that whole rant.
i use lcm, dpmpp3, exponential, karrass and depth controlnet on the same gen
probably. you want this one https://huggingface.co/madebyollin/sdxl-vae-fp16-fix
did they rename it, or did they removed it?
speaking of inpainting, i've heard the in/out paint canvas on invoke is super nice, keep meaning to try it out
I think it's just renamed
it's the sdxl_vae.safetensors at the bottom that you want. I renamed my copy to differentiate it from the original.
with some workarounds, you can do that on auto too, but require paradoxically more work
and the prompt normalization was added on comfy too with a custom node
What was XL models minimum resolution?
1024x1024

minimum? technically 512x512
LCM holds no interest for me so I can't really comment on that. I don't like the way LCM models/models with the LCM lora look.
you get deformities under that one
but yeah, 1024x1024 is ideal
depends on the model. kohaku xl is 768x768, pony and the majority of others are 1024x1024
they are trained on 1024x1024, should be able to do 768 and above
me too, that's why i only use it, to speed up the first steps of the gen and than remove it
Where can I find this list of forbidden names?
lcm is very good in prompting poses

the first few steps of the gen are 2 seconds for me anyways so it would make no meaningful difference in speed. If I had a slower GPU, I might be more interested in trying to make it work.
you will still get better poses 🙂
Medium
unless you use sdxl that aready generates fantastic poses
i don't want boring samey poses that LCM tends to make though 

what really put me down is that the pose editor of comfy is very rough
not even completilly supported
and outdated
yeah, anything to do with controlnet is a headache in comfy
yeah -.-'
Should I do something special before starting using XL model in a1111?
Feet 
a godlike hand model lol
that's one of the major downsides to comfy, everything is made by someone else, and nothing is ever consistent because of the random custom workflows
@gloomy scroll , download an XL model
pray to the god of vram

well, auto suffers of the same thing, if i'm not wrong
Going pretty badly
ah yes, second on the list is buy a 4090 😄
I like water and flowers
👍
Set clip skip to 1 and don't forget to change to the fp16 compatible XL VAE (those can both be done after loading the XL model). Also turn off Tiled VAE.
yeah, i too love sitting on water
i mean, sd1.5 should be also
true, can't do with clothes on
I wonder what I'm going to get since I only downloaded that VAE you just linked everything else was wrong
depends, most anime model is better on clipskip 2, tho i haven't really test that
Why 1 :oh:
Technically the clip skip shouldn't matter since XL should just ignore that.
Nubby typing up a whole essay to reply to you now 
if the model was based on the NAI leak, then yeah
It's doing something
Tiled VAE will error when decoding.
but also a lot of them were merged to the point where it didn't matter as much anyway
i would assumed almost all sd1.5 anime model have some sort of NAI leak in it
true as well
Rip pink neko
Pink meow
I used clipskip 1 on sdxl because i read somewhere i should, but I think i changed it back for some reason.. don't remember why
idk why but gives me Rei Miyamoto vibes haha
imma be honest, i was testing to prompt for asuna, but totally failed 
Tofu
why i fall for such women every time?
oh i see i see haha kinda looks like it now that you said it hahaha
the intelligent and cold one is totally my type, but we are totally not compatible
Cat with low vram 

same prompt, same seed, clip skip 1 and 2, i dare you to find the difference without checking the metadata
how many tails does Ahri have
6?
they burned me so many times
I think 9
just 3 tails
1 too many i guess
Because of the Japanese thing
ahhh the internet, ask one question get 3 different answers 
i think so too, 9 tails
she's supposed to be a nine-tailed fox, but i don't theink they all show very often
Poor 1070
She went to the gym I see
i was a main ahri, so i should know...
Should be 9
a custom skin?
counted 9 yeah

or they changed the base skin?
It didn't give error 
maybe it's only with hires fix? I could have sworn I had it work once but every other time it's errored for me.
just looked at the official page, and the default animation shows all 9 too, looks updated from when i last played ages ago
yep
I felt like I didn't want to wait another 7min for highres fix
You probably remember her walking
The tails get close to each other
maybe
Looks like one big tail
I don't blame you
she was different many seasons ago
@gloomy scroll

Oh my god 21.5s 
and my sd1.5 model at 20 steps is longer than this 
7s without hires fix and adetailer
should we be hires fix-ing sdxl generation?
it was a no no when sdxl model was first introduced
well, i do it all the time
i do hires fix x2, 7 steps at 0.25 denoise, it's slow but does well
forgive those weird stuff but...
she looks hungry 
maybe it's a trap with corset XD

wyz reacted to that image, help
tasty
you are on a watchlist now 
I'm innocent
plot twist thats his arm
I guess this is what happens when I disable Tiled VAE
i always forget i can use TRT with this model at this resolution...
me too, i get errors all the time
apparently it doesnt play well with sub directories
after making the default engine, i can't make a new engine
OOM?
even if i did, it will not change to the newly made engine
nope
not oom
then i deleted the default engine, then it won't even work
so 
if you have the checkpoint in a sub directory and/or there's a _ in the filename anywhere, it'll cause problems
i guess maybe cos im using symlink?
because it replaces / in the path to _, and get's confused easily

and they have no plans to fix it currently
well, i will figure it out next time i guess
I only really use it because it speeds up hires fix steps so much, sdxl steps going from 6it/s to 11it/s is great, but 2048x2048 hires steps going from 1it/s to 3it/s is where it's at
this is so incredibly stupid and probably explains the problems that I had.
what's worse is that version 1.0 didnt have this problem
they broke it in 2.0 and refuse to fix it
they literally only fix it when nvidia have some new driver that talks about it
it's the easiest thing to check for in the code too, there's no excuse, just lazy devs
i can't even tell if the new lora integration functions properly because i have to reorganize all my models to get rid of the errors
guys, did you notice? no more bots, someone did something?

pretty sure most people seperate their 1.5 and xl models into different folders (if they hav eboth)
no more spamming bots
as in #1047610792226340935 or those scams?
ahhh
scammers playing/hacking palworld probably
lol
they'll be back
and in number


the perfect generation doesnt exist
that's a perfect blue star, good job!


🤣
@supple raptor You still here? Send elp

they may be a little to much for the channel, expecially the 2°
imma self report that pic, and see if mod is okay with it
nice
eheheh
no
ermmm....
no Shalltear? what is this blasphemy
this model doesn't know shalltear 
I was just watching Shalltear yuri scene
sexy momonga sama
I'd check for too low of resolution, too high of emphasis, and make sure the right VAE is loaded.
Wtf you can't post nsfw here
I guess it was because weights
Time to start searching for style with 9min generation time per image

There is no ice cream there for you
its ok i just want meat
Message #1100170365604483202 hello
dream/Buildings with flowers and cars, next to an elevated overpass relief grayscale map
dream sexy minecraft rapper player
dream Buildings with flowers and cars, next to an elevated overpass relief grayscale map
because it's mostly the only active channel with people posting their generations
looks like it's more because all of the bot channels have this
but this channel doesn't have that
also they're locked so they can't ignore the message and type it anyway
sfw version
i can dm you the first one i generated if you want 😄
dream is the only nsfw i need
first ai generation in awhile
Don't yank it
those tomatoes looking JUICY
The biggest ahoge wins
no tomatoes

any1 know any good male anime models?
i would like to know too
🤮
pony and easyfluff both do pretty good males without them all looking like a typical 16yo anime protagonist from modern anime.
You dare disrespect the pony
haven't forgotten you, still having truma of what i saw there
Cool, gen settings? (Model, loras, etc.)
I keep seeing that term in prompting, what does it mean?
ahoge (アホゲ, "foolish hair"): Refers to any noticeable strand of hair which sticks in a different direction from the rest of an anime/manga character's hair.
Thanks
Thanks
@mossy hawk here's a couple low effort male gens with fairly similar prompts. From PonyXL and EasyFluff respectively.
what prompts did u use?
They both have PNG info so you can get all of the prompts and parameters.
alr ty ty
PonyXL is an SDXL model so it will need more vram to run. EasyFluff is a vpred model so it will need cfg_rescale extension and you will need the hll6.3-fluff-a8 lyco that I used if you want to use it for anime styles. https://github.com/Seshelle/CFG_Rescale_webui
here's a full guide for EasyFluff since it will not work well without some prep https://rentry.org/5exa3#anime-on-furry-models
Alr ty ty
Vpred?
without going into too many details it's a fancy training method and without that extension to accomodate it, you're probably going to get washed out and generally bad images
Does it give good results at least?
For all that setup the payoff has to be great
Yesh
Can I see an example or two?
can u combine two safetensor models?
Most likely, yes. But it depends on what is actually in the file. .safetensors is just a file extension, it's commonly used for models, loras, lycoris, and vaes. You can merge models with other models, loras, and lycoris and vaes can be baked into the model so that they don't need to be loaded manually. Loras and lycoris can also also be merged together with each other, though the useful of that is less since you can just load multiple at the same time.
It's also used for Textual Inversion embeddings, which cannot be combined with the other ones,
In all cases, the file has to be based on the same model architecture to be used or combined together. So no mixing SD1.5 and SDXL stuff.
Gotcha gotcha, know any good models that would go good with checkpoint models?
havent done ai art since a year ago i think been awhile
nvm 2 years
You've possibly heard of "merged" models. These are just models that are the result of merging multiple other models and or lora/lycoris together.
In your opinion, is it better than ponyxl
It's easier to wrangle. Far less character knowledge, but quite good at creating made up characters. The artist list that works is also much more defined since the hll lora/lyco trainer provided their full list, while the PonyXL trainer has intentionally avoided releasing which artists were trained.
How about features like hands, is it consistent with real hands?
aside from the two mentioned above, these are the ones I use most
https://civitai.com/models/211628/soushiki
https://civitai.com/models/100505?modelVersionId=191903 (this one is my own merged model)
i love civitai lol
I wouldn't say hands are a particular strong suit of either of them. Kakigori and Soushiki give me more consistently good hands. If you struggle to get good hands on those models then I suspect you will also strugle on both PonyXL and EasyFluff.
Ok
That could also be because I've made many thousands more images on both Kakigori and Soushiki so I just know those models better. Pony and EasyFluff are still relatively new to me.
/help
its a model for sure
pony->sd15
can it create Mikasa tho
More powerful maids
which one?
the one from AoT
low effort attempt

English please 
how do u use pony exactly? kindve confused how to use it lol
but thats not Satsuki
/ban
Cute tomboy
What part are you confused about
the score_9, etc etc
Just put it on your prompt
is there anything else i should know about or just put that in it?
source_anime, digital drawing, 2d, score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up
Negative: source_cartoon, source_furry, source_pony, painting, (worst quality, low quality:1.2), lowres, fewer, extra, missing, artistic error, jpeg artifacts, scan artifacts, (abstract:0.9), sketch, lineart, chromatic aberration, unfinished, blurry, monochrome, censored
Kinda what nubby sent
You can do many things
Start with that see if you like 
why do _ ?
Because it's what the model understands
(:1.2) is strength of the word
(water:1.4) will be a lot more water than just water
and highet the nymber the more strength?
Ye
Alr ty ty
Don't go below 1 in this model it breaks most of the time
and then more (()) means more importance correct?
Same thing
gotcha
what sampling method u recommend?
oki
oh interesting, I should update my n eg
Not sure how much it really helps
But at least it's more consistent with it
For an anime look
Yeah, I think my pony neg could still use some work. I ripped that one off someone's image in the touhou server a while ago and it's a lot better than whatever I was using before for that model.
@supple raptor have a queston.. when using models for controlnet, I must have files with the yaml extension in that folder?
I think I heard all of the necessary .yaml files are included with the extension now. I've had them alongside my models for a long time so I haven't bothered trying to get rid of them.
How big are the model files? I have, as I understand it, standard ones of 1.45GB, and are there smaller ones?
I use these pruned ones. 723mb and I stand corrected, I don't have the .yaml files. That must have been the old ones I had before these ones. https://huggingface.co/comfyanonymous/ControlNet-v1-1_fp16_safetensors/tree/main
It is
damn, how did you do that. i never saw a generation with speech bubble with anything comprehensive like this
people posted a few here,u just havent seen them
sometimes using something like spoken heart works, for my model it more often works to do something like speech bubble "heart" or similar
a few 1.5 models can do it , but sdxl does it better specially with letters
ah, i see. unforunate, i can't run sdxl
this one is spoken question mark in a (speech bubble:1.2) sometimes you have to force it a bit
and NAI3 do it best of the best yes?
yes because thats the best anime model out there ( not better than dalle3 tho)
is that an sdxl model?
yep. but a lot people playing with SD1.5 models.. 😁
yes cuz they are probably in third place behind naiv3 and dalle, on last place are the other anime models that run like poo and cant make anime chars
i looked at civit, i got a model name nai3-chibi, i don't think you're talking about that, are you?
no its the naiv3 from the novelai website a subscription based service where u can gen pictures
i see
pay and gen pics😁

life is hard!😈
only when you're poor 😛
or when it's not your priority in life😝
@winter fable in general, why do you need hearts, etc.? be original!
i was just interested since i never saw that from a model before
be original like @gloomy scroll @supple raptor @vital raptor @native halo @zinc flax !
I'm so not original
hearts are easy,the hardest ones to gen are emojis or letters
try to gen the poo emoji 
can't you just do that with inpainting
yea its easier that way,just gen an empty speech bubble then fill it in photoshop/csp/gimp
takes like 10secs
💩
🤣
sorry... sorry.....
this is wonderful!
I have a business idea. make a highly specialized LoRa
strictly speaking, that would be the easiest lora to train when it comes to the data set and keywords
even if it was just all emojis in speech bubbles
Well, I’m not at all good at learning and creating. so.. just gave a business idea. the main thing is not to do it in the realism section
based on what AI generates from the lyrics, i need to find some less depressing music...
try with this lyrics I wear a mask with a smile for hours at a time Stare at the ceiling while I hold back what's on my mind And when they ask me how I'm doing I say, "I'm just fine" And when they ask me how I'm doing I say, "I'm just fine" But the fact is I can never get off of my mattress And all that they can ask is "Why are you so sad, kid?" (Why are you so sad, kid?) That's what the mask is That's what the point of the mask is
i just did the lyrics from Crawling by linkin park, it wasnt pretty...
do that barbie song, it should be interesting, i think
most of them come out pretty cool
that looks like demon backshots meme
nah, man it look that one video on YT, guess d*ck or a toy video
tried the barbie girl lyrics, but anything with 'barbie' just generates images of the doll
logo and all
unfortunate
So this is what you have become, Golumn? My darling! 😂
😄
it's generated from the lyrics of Neurotic by Three Days Grace
Angels Fall - Breaking Benjamin
so i got curious if sd can understand different languages, safe to say it can't
i feed it a song about massacre and war, and here's the result
depends on the language, mine understands japanese fairly well
now Nightwish - Amaranth .
really? mine was in japanese too
Pop/Stars - K/DA
it depedns on the model too, i had to add 'nsfw' in the negatives because it was understanding the lyrics too well
which changed it to something completely different, but that's hw it goes
you're using sdxl though, maybe sd can't?
realistically, generating on lyrics is the same as throwing a ton of random conepts at the model and seeing what sticks
yeah, that's possible too, sdxl is much better at 'normal' language
in the end, whatever keywords/phrases are heavier in the model training are what guides the image, in these, the lyrics "rain in your heart, the tears of snow-white sorrow" are the largest impact
also "in a land of the daybreak" probably sets the lighting mood for them
Nightwish - Bye Bye Beautiful
👍
what are the diametrically opposed songs....
white and dark
man, haven't heard this song in forever
Within Temptation - Mother Earth
first 2 lines of the song define the entire image, pretty sure nothing else actually matters 😄
Rivers and mountains she creates```
Ameno Dorime 😄
it's the Era version, but the one on my playlist was done by Kyoresu
Money - Broiler
what the heck is forming behind her 😭
what model do u use?
ic ic
I didn't make it to the list 
Smol ahoge
green hair miku 
@vital raptor lyrics prompts are fun sometimes 
Lyrics do weird stuff

Me most days after anything
Work, food, waking up..
@native halo
removable tails
hmm, it might not be controlnet, something else must have changed
maybe the workflow is loading the model
there isnt a workflow
but i also tested with a clean install, something else is going on, just getting half the performance for some reason
auto111?
yeah
well that was stupid, it was the sampler
somehow got changed
loaded up comfy for a sanity check and it was fine, went to compare the meta and found my a1111 sampler got changed at some point
so annoying
one of the most annoying things about a1111 is that it doesn't save default settings, so if you have to restart the service it resets everything
that's the problem though, i was loading an old image but that old image was generated after a crash lastnight that reset the sampler 😄
I manually edited the config now though, it'll load with the proper stuff next time
it's also dumb that it defaults to a sampler with such bad performance instead of just using euler a like the img2img tab does 🤷♂️
true
Settings > Defaults > Apply
This will set the values you currently have selected on every tab to the default values so that they will apply automatically whenever you load/refresh the page.
Includes your prompt, so if you can include your default negative prompt as well as any quality tags that you always use.
interesting how I missed that option, I used to have an extension that saved everything whenever changes were made so the last state was reloaded, probably why i never found it
i remember having to edit the defaults manually in the past, but looking at the patch notes it maybe have been <1.3 when i last did that
sorry sorry... of course and @nova remnant 😁
Have a cavern 
crap! my computer can't generate it that fast...
Miku, why??! why did you cut your beautiful long ponytails!!?? 😱😁
she somehow reminds me of the heroine of the story "Peppy's Long Stocking"
I think you mean Pippie long stockings
yes. Pippi.. Swedish author Astrid Lindgren
👍
I like both the book and the movie.
Morning
hi hi
palette for mixing paints?
ok I should be in bed, nite.. thud

👍
I tried so many times, but I still couldn’t get the character into the box properly...
@wooden coral
Well, that's how to get a model to draw an archer girl!?!?!
and I hoped that it would work well.
CivitAI Browser+: Error occured during download of "niji-watercolor02.safetensors
Are there no mods online right now? @wooden coral this guy is spamming the links. I'm assuming it's a virus.
spammers... and all our defenders are sleeping...
shhh dn't poke the universe
Just script using stolen accounts
i know but still, could at least program it to post different links instead of the same one
hmms as he gets an idea for a prompt
spam you say?
ok not spam but nice
kinda
Yummy!
do you guys still have your first ever generation?
yep
a have a some
yes i still remember my first time
my first one on the dalle2
😂 so much better
some from me....
Demon be gone!
It’s always interesting to see and understand my own, although not great, but still progress.
my ask for a knight with medium booba with battle dress, doesn't match any of the prompt 😅
Any lora with this style?
there is one on civit, don't know how good it is
I made a baki anime style lora. It is decent, I could make an update in the future. I have uploaded the training data that I used. I encourage othe...
I don't think that lora can create anything similar
you can always create your own? from what i saw, it's not that difficult
btw @winter fable have question. you can download anyting in CivitAIBrowser+ now?
yup
i am cant
file with "other" tag, you would have to set them up yourself
which files you can't download
is that a lora or model?
一个水彩模型 推荐使用anything3作为底模 高清放大选择latent系列 并限制重绘幅度在0.45-0.55 a watercolor model It is recommended to use anything3 as the base model HD zoom select la...
Animezon Delivery?!
on the list of things i wish i could order online... 😆
2 day delivery from A-meow-zon
what does the terminal says when the download fails?
same
but in CivitAi Helper i can download without problem.
and aof course i am create OpenKey
A-meow-zon - delivers the best packages! 😁
they even come in a variety of colors 😄
Such a fragile package and without proper packaging!
@brittle rampart I tried to download that watercolor color lora, mine downloaded without any problem. Don't know why it showing error on your side
thanks... maybe on my side. but in CivitAiHepler a cam get it too
download any problem
/ dream Painting of 2 fairies looking at the camera and smiling,Asian girl's face, full body photo, white wings
well it's not that important...
collect the whole set!
people behave strangely. and why?
?
why do they come here and try to use bots?
because all the bot channels are locked and they can't bother to read
Because they know that this server has bot but don't actually know what channel
there's no "bot is down" message here so they think it has one
and they see our pictures and think that we generate them right in this channel?!
i guess
I never would have thought...
You have made mistake and came into biting range
a couple of seconds before the bite!
you need to feed those you have tamed! 😁
A slightly redundant prompt it seems to me.
Volumetric Lighting, Volumetric Light, Volumetric, and
high details, detailed and intricate, intricate details, high intricate details,
all the details!
i was just recreating my first generations with with what i know now
i didn't really change any prompt or seed
“repetition of the past” so to speak?
this was the original generation
nah, was checking if just changing few setting can gave a good results
sigh the faces are mangled. wtf
adetailer
adetailer detects a faces?
doing that now
but the blindfold didn’t appear here either?
hmm really didn't change it
my model still does this most of the time, even with very little prompts. Don't know why.
that doesn't really help, you know
oki.. i can download your model?
it's on civit
i put black gown
of course
try (black gown:1.2)
Try gown, black clothes
or that
Might be hard trained for tags
Hello people who are awake, I am not 
Debating throwing myself in the shower to wake up or sleeping now
😂 now with a completely different problem, even the hairs are black now
What CFG are you using
5
Try more
same model .
what a positive person!😂 👍
🤣
To balance the mood
oh noooo!!!!
Meow double trouble
Too much 
double the trouble, double the fun
Give them crazy hairstyle, dynamic pose
sleeep
I hope
they dangerouse!
She's standing in front of a picture
Fake background


You don’t understand anything - THIS is an artistic move! 😁
you do it! 👍
Big 
just ignore the right hand 
You didn't wait for me to finish
I was gonna say big foliage
👍
She so smug
and I made big butterflies
btw. good model!
i don't want to use depth library though 😭
too much work
that nose 
It's gone now, butterfly owns it
What about pony model? It's pretty good at hands
😅 i haven't touched it after that scare
well, it's almost midnight, goodnight people
final generation of tonight, entire background gone for some reason. Though at least you can't see the hands
see ya
🥴
Pony model? what ti that?
the sdxl model that is very popular 
also there is a 1.5 version maybe its okay ? 
But I think I'd rather soushiki on 1.5
SDXL agane 
okay imma actually remove this, it looks sus af





