#🍥|anime
1 messages · Page 184 of 1

Anyone try animagine XL 3.0? A dedicated anime SDXL model
search the channel, a few people were talking about it earlier
Yeah I just saw it after scrolling. Maybe because of the reddit post people jumped on it
I've been using it a bit today. Overall it seems pretty decent, but I'm not a big fan of the style and the style is not very flexible. Also, it doesn't know Touhou girls which is a huge negative IMO.
And when I say "the style is not very flexible" I really mean "artist tags are basically ignored and the medium tags mostly just make it look weird"
It's basically an SDXL version of AnythingV3 but with far worse character knowledge.
Hi nubby
Hi King Pendragon

That's really cool what model?
Probably bing
Mr burger loves using bing 

It's ok. Reminds me of the first ever NAI minus the touhou dataset.
I was interested in one fact. why during generation the use of GPU is no more than 15%. Is this how it should be? Generation uses only the GPU Video Memory ?


vil
@lyric zealot
Free elves
👍
I tried Elferyone but couldn't quite gen it perfectly lmao


What model?
is there no metadata? 
On my phone in bed
should be soushikiV1
Can’t check
another lazy enjoyer
I’ll check tomorrow
lazy sleep deprived

ok not sure why she looks sad
1 more than back to bed. cause brrr it's col
cold
then big brother needs Onee-chan😁
I think you might mean imouto, onee-chan means big sister
OOPS
then imouto
uppies?
prompt for pose and angle shot?
hi hi
its in the file
WildCard?
standing, from above, (arms reaching up:1.2),pov,
png info tab if you're using a1111
ah )))) (arms reaching up:1.2),pov, now i am not near my PC
Friday, barbecue, friends, beer... a lot of friends... and for some reason a lot of beer
Metadata says ProtoLarkXL_v5
oh xl
it's a custom XL model
the model was made for more realistic images, but when I force it to do anime it has an interesting blend
yeah, i like the colors and forms too
only real 'problem' with this model is that it starts leaning pretty heavily to the nsfw side of things with certain prompts

like, if i don't specify clothes, it won't generate them 😄
Even if you have nsfw in negative
Are you specifying body parts or size?
Happens to me when I do that in many models
honestly haven't tried to put it in the negative, I try not to use negatives as much as possible since XL doesn't really need/use them as much as 1.5 does
no, even without body prompting
like, my prompt could be a girl and it'll just be a random naked girl
not always, maybe 1 out of every 8-10 images
but it happens a lot when you're generating a lot 😛
👍 model is testing your patience
"You didn't specify clothes so... Here you go."
I'd just leave nsfw in negative if you plan on posting it on servers :b
Shouldn't hurt quality in any way
I just fix the prompt if i need to, I don't usually care if it makes nsfw stuff, just when i'm trying to make something to post here
plus this model will likely never get released, so it doesn't really matter
lol, not what i was after
mmm donuts
funny thing is, there's probably not even donut inside the prompt
yeah that happens
tiny necktie 
@finite herald
looks like me at work lol
Seconded
Hii, how did you prompt for this type of character; chibi or smith? In the keywords or just describe it?
I mixed several loras to get that style
<lora:lk149-more:0.3> <lora:manoneko:0.3> <lora:Rabbit-Artist-LyCoris-V1:0.5> white background 1girl, black hair, blue eyes, blunt bangs, very long hair, dark blue hair, round eyewear, sweater, holding book, standing, denim shorts, black pantyhose, :<, head tilt
But to prompt cute looking type of character you can use either loli or chibi
Though it depends on the model you're using
There are also models specific to making cute characters, especially for 1.5
It really is
I usually take a previous gen
use an interrogator
and then use the prompt and run with it
Here's less thighs
MORE THIGHS!
What model?
That's Bing
yeah what sleepti said... I post a mix 🙂
don't click that
i never know who to ping 😛
@wooden coral
you don't have to ping anyone, right click it and go to apps -> report
or use the ⚠️ react
Ah, I didn't know about that woops

dont think so

It's definitely dying
stuff taking 3 years to load
Can't even see any pics
It's going to be resolved in 5 years
i see pictures, but they won't load the large size
Can't even download actually lol
One day discord will stop being a scuffed chrome tab
lmao
Soon™️
it got fixed?
Not yet

its rias gremory 
Lmao
The model I like is really bad with cars 
ohh
i try to mix & match things
can pls stop sending that emoji
i m Thomas (the cat) ffs
I just love it lmao
Also doesn't notice that
I never knew the cat was called Thomas

Just always Tom to me
Can a mod check deleted chats?
no?
will try fr
Dynamic pose is honestly fun
@nova remnant yo someone told me to fuck off when i was just tryna generate some chinese houses💀
tf u sayin
basically i kept generating chinese house's in #1103708504142925824 till i got a notification from someone saying fuck off chink
And immediately deleted it
I know the name
Rude
I just wanna report it to a mod that's all
thats not legitimate discuss it with the mods
I can i can
I'm just wondering if they can check deleted logs
well
He was being racist so yeah
do you know who it was?
well, then just let the a mod know if you want with the details
then if you want just move over to any other bot channels
there's like multiples
well well what do we have here 
I mean he's generating pianos and I don't see me complaining
Is there like a support ticket
Or any report channel something
Anyway thank you
welcome
Smolsus best bot helper 
well, doesn't really matter tbh. Just dont engage with people like them and move on
@lyric zealot generate cute elf
using Sleepti generation with openpose 

You're right yet i dont tolerate racist jerks, anyways thanks though dawg fr
That's pretty cute
Ima go now
Do you mean this guy?
hugs?
well, let's not witch hunt, if he wants, he can report to mods, and lets move on
I can agree with that but now it got me curious why did lopez even say that to the poor guy
We will never know
Hope he gets banned though
Yup. I lack evidence though
That's what I'm requesting them to read deleted logs if they can i already made a ticket
Have a good day then don't let a low life crawl to you
I need a hug

it's been long since i used ExpMixLineV3 
I tried for awhile yesterday to make my model do hugs, never got any decent gens 😦
Make more ExpMixLinev3

Your own Lora likes kids 
Can you make them more idk "OLD"
Teen years somewhere in the 20's

I could
What you mean they are 3000 years old elves
I heard that in an anime
👍
No, they're drawings/ai generated fictional characters, they don't have an age
It's actually a 10k year old vampire queen
weird, usually 10k yo vampires looks younger lol
I don't believe you
Perfect
squish
not what i was trying to make at all, but it's cute 🤷♂️
now i need to figure out how to prompt the hair to do that intentionally
so uh been away awhile longer, what's the current "meta" ?
type in prompt, click generate
Pony model

Pony diffusion on civitai it's sdxl
Really good 
depends on what you want
some people use sdxl more, some use 1.5 because of the selection of models and loras available
sdxl also takes a bit more vram, so people with lowend hardware tend to not use it
stop! this 10,000 year old PNG must be stopped!
JuggernautXLv8 does okay at anime
if you are intending to ask me how i gen this, i have no idea. I have no model to give you, i have no prompts to give you
no, i was literally testing something called sudowrite, and was typing stuff in their canvas, and then there's this visualize thingy, then i press that, and it gen for me
ah okay
Woman Dancing in Arabesque art with Hobuko style, her arms and legs creating kinetic photography effects in a nocturnal space, photogenic, low shutter speed this is most likely nonsense, I haven't tested it
but someone has to
Not so bad
well, it's a 1.5 based model, so yeah
Im just surprised it did exactly what I was thinking of
my sdxl model
people were doing poked cheeks earlier, so here's one... lol
Here's Grandma.
And Grandpa
There too
the combination of techniques and effects makes it nice, rhythmic and effectfill
like "A victorious warriors sits in a field of corpses,captured in anime art style and abstract brutalist design, gothic-art, dark fantasy influence,"
I sciiiiii
The Emperor of the Eternal Sea, cursed by the Red of the Underverse
Cute 
apparently the whole cheek pinch/poke thing is a concept SDXL models just don't know, or at least it's not a very strong concept and no loras on civitai
i got really close with one gen, but the finger is on her lips, maybe in a few hundred gens it might hit her cheek
probably not
I need to figure out what 'picsart' uses for outpainting.
I seem to recall the old NAI did
NovelAI is a company/website
but you're probably talking about hte model that got leaked for 1.5
yes and they allowed it like a year ago
no
I'm talking about NovelAI
not the leak
oh, well i guess check their website then
cause anything that doesn't allow nudity and violence won't work for my dark fantasy needs
cant find any info on it
someone that has it could just try a obvious hentai prompt
This is a terrible prompt!
ig is
NAI allows violence and nudity.
nice
Can someone help me make an XL model of Civitai run on my PC?
because now the civitai only has an Xl model
Is pony diffusion some kind of meme or is it genuinely good?
I'm testing it myself and I can't decide
it's genuinely good.
it's good at generating pony?
probably, but it's also the best publicly available SDXL Anime model
I made this with ponyDiffusionV6XL
this too
Hd as hell
more pony model
furries and bronies have more money than weebs 
Idk who's worse but Booba gens are made more possible each day
👍
This is... a really weird model. It has some very strange weighting on certain tokens, which makes it barf out nonsense for a lot of my standard anime testing prompts. It may be "sketch" or "simple" is what's affecting it so much, but it's an interesting one for sure. It's cohesive even in its terrible generations which means that it has real promise if one can figure out how to use it effectively. Out of the box though it's very strange, and will likely require learning a different way of prompting that works best for the model. Not an ideal situation when SDXL itself is very natural-language based.
Included in the half-size testing preview is the latest BluePencil for reference.
better be careful of that tattoo. I got booted for 24hrs with that once
Okay, I can clearly see that this will take some time to understand properly. If one includes the "quality modifier" keywords in their example images, it seems the cohesion increases considerably. That and proper sampler and CFG usage.
every prompt shoudl start with score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, score_sensitive, digital illustration, (source_anime:1.15), basically
or something similar
and it really shines on NSFW
SFW is mostly just OK
Thanks for the heads up! This is great info.
artist names
that kinda work
those will help a lot too
Ooooo
this one was ponyxl
Looks like it doesn't work too well with upscaling. It loses too much cohesion, but HimawariMixV2 at 0.4 denoise is fantastic for the upscale.
My thoughts exactly.
Quality modifiers?
thanks
ponyxl is alright, I guess.
🤷♂️
ponyxl is just off doing it's own thing ig
any quality differences between pruned and unpruned models?
in general, pruned models have "extra" data stripped from them, so there is a quality loss overall, but in most cases as long as it's done properly it's probably not very noticable
ok
pony also seems to deviate to less sfw results overall
it's basically just trying to save some space by making the file smaller, if you have the space then unpruned is probably better
also if you plan on doing merges or especially training on the model, you want the unpruned/full version for the extra data
the extra weight is because of the ema weights that are used for training if u arent training then theres no reason to get an unpruned model as it has both ema and non ema weights
Best v1.5 or 2.1 model?
ponyxl, best model period
Sfw results aren't great though
and I'm building a TensorRT engine for it too
it depends on your standards,if you lower them even a cake made of dirt can look tasty. 💩
[E] 2: [virtualMemoryBuffer.cpp::nvinfer1::StdVirtualMemoryBufferImpl::resizePhysical::140] Error Code 2: OutOfMemory (no further information)
I keep getting this error
during TensorRT engine generation?
it takes a lot of vram to make a TRT engine
especially for a XL model

You have a 4090, right?
yeah
Could you do me a solid and make a TRT engine for ponyxl?
sure
thanks
Pony Diffusion V6 is a western cartoon style SDXL 1024px finetune capable of producing stunning SFW and NSFW visuals of various anthro or feral spe...
depending on how much vram you have, it might just be a matter of settings too, an engine set to max batch 1, min/max/optimal size all 1024x1024 and prompt min/max/optimal at 75 will use less vram
having any of the settings different/dynamic will cause more vram use
I only have 8g
😔
ah ok, no way you're doing a XL TRT then 😄
4.8gb for the trt file and another 5gb for the onnx file
Not sure if they can already
yeah, was just thinking of that
Civitai has a good search mechanism, when it works so wouldn't be hard to find your model trt
tbh, i don't even know if sharing them works anyway, but it would be cool if model makers could just have trts there with it
i think that will not work. It's based on your own graphic card
not sure tho
share the gpu then
it's on amazon, you can order it whenever you like
😄
looks like it'll be about an hour for these files to upload
Got it
👍
Squish
she's ready for the squish
the what?
(latex:2.5)
grab these horns instead 
This program whatever it is, is really bad with tails lmao
And tried on fingers
It's just really good with text and logos
@native halo
👍
what a massive tail
looks like an anime 🙂
pov from above?
No the hand on head
or the headpat
i used a lora for it, much more consistent
my model will do it without the lora, but it takes a long time to get a decent gen
this one?
She's a wrestler and about to learn the knee to the face technique ( real )
camera turns on

@lyric zealot she's going to record herself teaching how to do her hair
👍
Well that's one horny girl
cute
i need help guys - so i am creating a work project for my faculty to be greaded - what i need to make is a mobile app using adalo -(https:/www.adalo.com/) - mobile phone chosen for me is Galaxy S20 - so now i am wandering what size background do i need to generate -(already chosen model and everything but i have no idead about size of the image )
1 Inch correspond to 2.54 cm and 1 cm correspond to 37.7952755906 pixels
measure the phone's screen and number of pixels and so, the ideal resolution
the minimium is 512x512, so, you need to resize it after gen
this?
resize to downscale the image
better use a different application
512x512 is too big for a phone screen
it can load it, but it's too pointlessly big
i only have automatic 1111
to downsize, even windows paint is better
the ai is not able to work well on image with a size inferiors to 512
photoshop would work?
FB group i'm in the daily theme Iridescent Orange, Violet, and Dark Blue
never used, but i think so
the galaxy s20 has a 1440x3200 resolution
what's the use of such a big res on such a small screen?
yeah, but they need to still be big enough to be seen lol
it's basically so you'll never seen any individual pixels in a game or app
only to me it looks totally overkill?
it's not functional resolution at that point, it's just presentation
it comes down to details, same thing applies to generated images, the more pixels you have, the better the details can be
it doesn't make the images better all the time, but more pixels helps
Godzuki?
maybe a little for a fine eye
reminds me, i was going to make a little info graphic showing why more pixels matters
lol
day of Godzilla?
i see it, but ... not sure i like that types of art
oh wait, i see girls besides just pony at the bottom
It's the best model lmao
it seems a little random
Put something like this in your prompt to start score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, score_sensitive, digital illustration, (source_anime:1.15)

look like @nubby model
i saw all that in the civitai examples, but it doesnt say anywhere why
hey hey what mean this score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up?
i play with kakogori model every now and then to make bday cards for kids in family circle
Pony Diffusion V6 is a western cartoon style SDXL 1024px finetune capable of producing stunning SFW and NSFW visuals of various anthro or feral spe...
so this is a feature exclusively of the model SDXL?
no, it's a bunch of tags in that model that are apparently required
i'm kinda questioning how the author implemented those trigger words
surely it is, not something common to sd model
same seed/prompt on my model and the pony model
yah..... 😠
the model seems to have a lot of tag knowledge, but i wouldn't say it's a "good" model
maybe merging it with something else would be useful
btw this is not generic to xl model, its that particular model specific
I think you need to play with the prompts more
than just use once and call it x
There are so many ways to use models
ree my keyboard is dying today, not pressing stuff
i mean, i've generated over 300 images in the last few hours
but these keywords apply to regular 1.5 models? if so what exactly do they mean?
but i'm not making what it was made for i guess
no, they don't
it's something trained into this model specifically
sadly
this turned out to be pretty good anime model too, in case you haven't already tried it out https://civitai.com/models/119012?modelVersionId=293405
Merge everything. Stable Versions: v3.1.0 , v2.0.0 , v1.0.0 Recommend Negative Embeddings: unaestheticXL or NegativeXL See HuggingFace for a list o...
yes yes,,, but in my case i am have only version this is model for SD 1.5
crap! I think I should rename myself Vintage Girl
i havent tried bluepencil 1.5 model but looks good from the cover images on civitai
🤷♂️ i don't know what special magic this model needs to work properly, normal prompts don't do very well
bluepencil xl v2.0, the current version is 3.1.0 and can give pretty decent outputs for anime
it delivered the prompt concept pretty well, just some deformity to fix
dont know dont know,,,
maybe they mixed a lora or an embedding with the model, if it's possible
it's pretty easy to merge loras into models
and embeddings?
not as much, those are trained differently
loras are basically trained the same way as full models, but with a small dataset
if you want to learn to use comfyui here is a list of videos that can help https://openart.ai/workflows/academy
personally im not interested in comfyui interface but people can benefit from its memory optimization and some features
thank you so much, but i am try using Comfy.
for example
cute
this is my comfy
and this is too
thats cool, do you get to manage memory with comfyui?
That image is made with NovelAI
ahh yes, i got you confused with yHR
people who use comfyui claim to have much better memory optimization, so if your image generation has improved with comfyui on your gpu then that's great
to tell the truth, I didn't notice. Perhaps my workflow is not so optimized.
on low ram, you can see an improvement, but i suppose, on high performance, much less
without any major tweaking of things, comfy is actually slower for me
my last vertions of workflow that
but its a matter of like 0.5-1.5 second difference on average, so it doesnt really matter
go through that website, the author gives you some good insights to workflows
thank of course i am wathing this is
adventure = grab a backpack and go
Is this pokemon ? xd
that's just me lazy with prompts 🙂
forest, pond, night, moon, girl, adventure
Double the trouble
my fairly old render still holds up, yeah it's 1.5
either there's something wrong with my auto or it hasn't updated for like a month
or more
Hi everyone, I am completely new in stable, but I tried to do an Isabelle from Animal Crossing, here is the result, hopefully I can post it here :p
your creations are awesome, I'll continue learning to be able to create something similar in the future
i would have made her more fluffy, but i love her like this lol
This is cool! I love the flat design and the fact that the skates actually look decent! Roller derby style!
How'd you do it?
hello 
ah looks like an xl model
I'll try to do it more fluffy next time, I wanted it too xD
those scare me because they're slooow for me 😛
I like your name hahaha xD, hello!
I just used what @bitter latch typed above theirs and added 'anime style' for my model, anime style, uWu roller skating 70's style roller derby girls
cute
How about this one
and yeah, it's my custom XL model
Lmao
less furry and more feral XD
Its so cursed for some reason
so charming
I just put the name to gen to see if SD knew the character
looks like a dog i know lol
interesting what it creates only with the name xD
Do one at 3840x1600 🙂
The others were normal looking gens

Nerf!
gotta run an errand, i'll do one when i get back 🙂
Hi :)
it got vibes
oh my gosh... 10,000 maybe?
yessss 🙂 🙂
@thin torrent glad you liked it, but careful PM-ing... gives off spambot vibes and I'm never sure if there's a person on the other side 😛
Is it Tifa overload?
who is alexa bunny?
oh
yeah a bit!
a lot of models know her pretty well, especially if you add the description of her outfit
and obviously Loras will take it the rest of the way if needed
I've tested maybe 20 or so and have my favorite Loras
cute, keep up the good work 🙂
nice vibes
👍
@keen hollow lmao kinda buggy anatomy but a fun prompt for a character like her just relaxing on the grass
(Don't really know what she does but I imagine its fighting stuff in the game
)
yes more
I dunno never played those games 
both! she's awesome!
Do you play any games at all?
FF7 remake is worth playing if you have PS5
@keen hollow I think you can see where I sank my time into here 
oh geez you gotta play the FF7 Remake.
I'm not excited for poe2
i dont even know what path of exile is 😄
thank you, I'll do it, your creations are so good btw, I want to be as you when I grow up xD
The one thing I was most excited for was them being the same game after the campaign you do, but that was cutt off and they separated them, I got really sad 
Path of Exile is what Diablo 3 and 4 should have been
I'm 20k hours deep and thats without standalone launcher 
Yeah I played before steam too
Don't really have that much time off steam, was mostly when my pc couldn't handle running the game on steam for like 1-2 years
@nova remnant well I wouldn't really want them to be tied down to being attached to PoE too much. They will share things like store purchases etc

I just really really wanted them to do both the same, updating poe1 was probably impossible with how bad it is engine wise
yeah that would be impossible
they tried and failed at first
so they built a new engine
I'm sure 2 will be a great game
and used the "scrapped parts" to make the expansion for PoE1
kek GGG is the master at re-using stuff
anyone have a link for a good offset noise lora?
trying to link someone to it, but IDK where I even got mine
search for epi noiseoffset 
ty
Me play POE a lot
rendered this image with XL Model but used 1.5 LoRA and embeddings
not sure how did that work w/o any errors
unless they added cross compatibility
one day you'll get there
this is what soushiki think isabel is
Yeah when I don't get burnt out by that boring story before I get to maps
good evening guys, i have this problem where my colors get all weird when i use controlnets with sdxl checkpoints (the controlnets for sdxl), does anybody know why that happens?
maybe wrong vae or resolution ?
too much/little cfg ?
i am not using a vae because i also thought that could be the problem, res is 1024 x 1024 but i get it with all the resolutions i tested..
i will try that.. but why would that ruin the image only while using controlnet? :,,)
I don't know, I just know that cfg can affect colors :b, not super familiar with sdxl 
if stuff is breaking I go back a few steps to what was working and change one thing at a time to find out what is going wrong
yeah
I need to play with this prompt later
that's always a fun resolution to play with
hmm, that last one was my model merged with 10% of ponyXL.. seems to work fairly well so far
anything higher in the merge and it produces garbage, not sure what's up with that model
seems like the sampler has a huge impact on it, that could be part of the problem
we heard you liked wheels..
hahah
ok, so this model does a lot better if you define a specific character and style or whatever, for generic prompts that i usually use it's just random
never had a model that randomly cycled through styles if one wasnt specified 😕
uWu Furry miku
lol
yeah I like to do widescreen... gives me a use for the images
need a version of that but like.. aged up a bit 😐 and without the weird leg straps... seems like a lot of models add weird leg straps
tried furry miku, all i got was cat ears 😄
try : Furry Anthromorphic Cat Miku hatsune having a UwU day
heh, adding 'uwu' does this
and your suggested prompt does this
not enough anthro in my model even with some ponyXL in it i guess
how to tell a model was trained on a lot of nsfw images 😄
umm well...
lol, this one makes me think @finite herald is at the window
haha
Face reveal
That's what you should expect if the model isn't an "aesthetic model" (read: not intentionally overtrained on a specific style).
It makes sense if it's meant to be as flexible as possible, I guess I've just never seen one outside the base SD models
ok now i'll be posting the rest of these over on a furry server cause I don't wanna get in trouble here 😄




