#🍥|anime
1 messages · Page 88 of 1
Can someone help me with a situation I am in pls?
I cannot go into detail here as it includes gore
By situation I mean on the internet
Like I just wanna know where to report it and to what authorities
I don't have any answers for ya, Kazoo. You're in Europe so I don't have a definite process for you. If it's on this server then I'd go to #1010934719455707218
If it's elsewhere, then you'll have to report it via that site's reporting method if any such thing exists and to the police, I guess.
Not on this server
It's on youtube
They'll just delete the vid and that's it
It features animal cruelty
I don't want this to potentially happen again
It's sadly up to YouTube to report it to the authorities, I'd imagine. I'd think there's little change we can effect since we don't know the channel's origin, or personally know the people involved.
This is just one of those things where we report it where we can, but otherwise don't have anything else we can do. :/
Lol. I'm not sure you'd have much luck there.
This has legitimately worked before in the same type of situation, it was for videos showing animal cruelty. I don't know anything about actually using 4chan though so I'm no help there.
4chan justice
Man I am way too tensed up rn
SD blessing me with that nice shoulder design, thanks RNG
Just released the model I've been working on for the last few days. https://civitai.com/models/100505?modelVersionId=107578
I'll make a set piece for it
After a few busy days I finally had time to sit down and get something with SD done again :D
@scarlet kelp I released it
scuffed hands but she's cute anyway
Did you manage to improve the hands?
I'll give it a try tomorrow
Yeah, they aren't perfect, but they're quite a bit better than when I started.
Hands!? This model is committing war crimes against books!
Who needs books anyways? Just ask ChatGPT to give you a summary 
I got 2/4 good books. We won't talk about whatever happened to her legs in the last one.
Is it just my model or is it really hard to get decent looking fangs? Mine does it so badly that I prefer to just not prompt it.
Yeah it is, most of the time I just add them myself in post.
cute
I ask for potions and get a bottle of jack daniel's instead.
witch loli ready for the pre-game
thanks for releasing it 🙂 I look forward to trying it soon
I hope you enjoy it!
I'm digging it I'll try and post some SFW ones lol
is vae baked?
Yep, my modded version of kl-f8-anime2 is baked
so I should use no vae with this model?
your prompts render a lot better than mine 😄
Rip feet
das a lot of fingers
Oh is that akkaimix
I really like the eyes from akkai
yeah akkai is so good
I don't think I appreciate it enough
I dig the semi-cartoony look
not sure if I can describe it in words properly
bubbly?
It does indeed look bubbly
I wanna try merging a non chibi model but the hands always suck
YOOOOO
That is indeed bubbly
you can add it to a regular gen with no keyword to kinda make the eyes pop a bit more
also makes my girl more crablike
Make everything underwater
ah here we go, a non-bubbly render with more pop from the bubble lora
Testing claw pose is a must
oh shit that was without a vae
I guess it's already baked in
Cow
Cheese
"hugging teddybear" 🤔
like a furbie without the fur
Look at its cute little face
Try SweetParfait
link?
same in lottamix2, a lot less crisp hmmm
oh man
I had friggin latent couple on by accident
the prompt was at 0.2 str lol
ty 👀
@supple raptor took a while to make, but I think I could have done better.
@zinc flax it's cute
:3
cute! I like it
Ooh nice
I'm not really that creative with prompts so it's nice seeing something unique out of my models lol
Mixception
that I've been using over the past few months
cute n creepy
Straight outta Made In Abyss
Never was there a Manga I wish that I had never started reading like that one.
I'm still recovering from Higurashi 12 years later.
higurashi is so good
Thankfully I knew not to make that particular mistake, lol.
Thic
higurashi does work as a style prompt
Oh? We doing old gens now? Here's one of the first I ever made with 1.4
nice
Cream of the crop, most of them were terrible, heh.
Hmmm I wonder what the focus was there... 
Top 10 mysteries scientist can't solve
too obvious
It's the purple theme of course.
Here's an even earlier one I think
old gen
they still kinda hold up
@zinc flax my mix on your sweetparfait is really promising
That one was upscaled after the fact, I believe. But yeah, it's not too bad
Ooo glad it's turning out nice
It took forever to get an actual tornado. You still can't easily get them.
I'm pretty happy with the hands on sweetparfait
Look how bad this old gen was
yeah they seem better
Yes yes
That's actually not bad. Nice armor.
interesting 'rain'
Yeah, a lot of strangeness to these older gens, heh.
I've been testing stuff lately and "perfect hands" tend to show up on chibi/loli models
Pretty interesting
DarkAlfa 1.0 is fantastic at them, usually.
Like the hair to this
lmao this is another old gen. wtf
This. Is. Fantastic.
I'm trying not to laugh too loud as it's very late here, but man.
lol yeh i laughed very loud just now
oh the good ol days
when this channel flew so fast you couldnt even keep up with the current state of things. people begging for models, workflows, etc.
I expect that level of abstractness with midjourny and not stable diffusion.
Here's an old NAID waifu in a box. Back when I had a subscription.
Someone dropped off their waifu and didn't come back for her.
i remember how many times I had to type (redacted) just to talk about certain models
Ha. Yeah
I always try peace sign waving and claw pose lol
I want a speculative biology model, so I can create my own evangelion angels
Aight. I'ma stop monopolizing the images now.
Still need to finish that series. Stopped at like a couple eps before the end, and then there's the movies.
It's been a decade since I watched what I did though.
80s-90s anime had the best style
For some reason I still have a folder of hundreds of failed lightsaber design gens.
composition is so clean and natural that only the eyes give it away to being AI.
That and the banding pattern on the crosshatching texture in the background. It really does look like a screencap from an old anime opening though.
Staring respectfully
Moth girls are better, 0/10
Okay, this is fire. 
Look at those hands, properly holding that pen, perfect shading, great book... aaaaaaaaaa

Lighting is abnormally ultra harsh, but nothing photoshop can't fix.
That's often a style to some images. I don't see anything wrong with it.
Edit and upscale of earlier post.
That's all from me tonight. G'night folks! 
O, gnite 
@lyric zealot Hmm... this character seems familiar.
my waifu number 100 
So many waifus 
because im stealing yours 
I haven't seen none of mine yet tho 
you know they at the other folder 
They better be 
Fair 
Don't worry I will take good care of her 





||metadata||
i HavEn't SEen NoNE of mINe Yet ThO
opens forbidden folder
:O
closes forbidden folder after seeing the content
:OOO
Those are really cute :o
Time to normalize midriffs and chokers for men!
great work i will check it out ^-^
femboys
-h[
Swiffelfesteungustischunstachten
electrifying power up
maybe
something with electricion tbh
yoooo
what model are u using if i may ask?
How can i make this character
I am using reference_only of controlnet
But is not working
Probably needs a lora since it's pretty complex
how should i use highres fix properly? its always destroying the original generation :c
this is what im trying to use so far

hi, you dont usually need so many steps for hires fix, your denoising strength should be much lower, somewhere from 0.4 to 0.53 might be good
thanks :D, how many steps would be optimal?
depending on your sampler, but usually about half of what your generation steps should be good
thx
And can i do it with only one image?
Ok thanks
im confused on what im doing wrong it feels like the high res fix is just messing it up more then the original
the first is before?
i just took a screencap of it mid generation before it got sent to the highres fix
ahhh, generate one without hires fix, then generate another with hires fix, same seed
so we can compare
maybe the upscaler is the issue?
Use anime6b
i would say it's for sure better, look at the eyes, and fingers. The upscaler can change the stuff as well
what is your denoising strength at?
0.46
damn 
Pink haired waifus hit different
Gudao from Fate as chibi and a version with moustache and beard
hmm even thoe i have 1male, it turned him into a girl
haha
offspring of Mario and princess peach.
more anime-ish
well, that turned out diff
using adetailer and dynamic prompt to get txt2img crowd generation
Amazing
1male isn't a real tag. It's 1boy. If you don't want to use the tag then you're better off just using the word man or male with no 1 in front of them because the 1girl style tags only work because the models were specifically trained on those tags.
I'm good. Not a zoomer. Don't need to be fast.
Well I guess you have a point
Very nice, did you prompt for every single character or what?
Well it is only to be expected out of LottaLewds
This is so fucking good wow
Man the images people are sending today are all so good
img2img on rimworld characters is fun

This was one that I got doing the same thing when I first started. I need to try to replicate it for each character next time I do a run. Waiting on vehicles to finally come out.
Nice! It does seem hard for me to get legs and arms though with img2img
Try adding full body, mid shot to your prompt.
I think I used relatively high denoising, and I also got lucky. This was, if I remember a few months ago, the best out of a batch of 8 or so that I did.
I wonder if you could cherrypick and then train a LoRA to get that specific style.
Weird. Usually when I go above .7, I get stuff completely unrelated to the original image.
.8 causes too much deviation from the face and hair style and outfit
Lower denoise with loopback might work better. Let it change from the original image over multiple passes instead of trying to do it all at once.
🤝
To be fair, my original was this. So what I got wasn't that close.
I type like I'm 95 yo whenever I use my phone
I think I could probably type faster on an old flip phone with physical buttons and T9. XD
It was more the character in spirit. I had previously been just prompting for characters, and I decided to attempt img2img.
The main problem for me is that the keyboard keys are way too small.
Yeah, should probably practice swiping instead of tapping each key. I do the same though.
never did loopback before but this is interesting
they're all scantily clad and looking miserable
lol
ohh you used wildcards to make the various expressions
that's a good idea
yeah
Just another day in #🍥|anime
I don't really mind, people make awesome art here

using img2img and addetailer + wildcard allow for randomized character with various expression with zero inpainting
10/10, would crush her exoskeleton
Should I play it?
It's a pretty good sandbox game with lots of complexity
Yeah I saw that online
But like I usually don't really like sandbox games, is it a good game even for someone who is not really integrated into the genre already?
If you are a player like me that needs more defined objectives to stay motivated to play, you might not like rimworld, still a good game though.
Reddit took down one of my OPM chapter colors for copywrite 
OPM?
One Punch Man
ok now I gotta switch again
Wait, are people training new models in SDXL 0.9?
I'm sure researchers are training on it to see how it responds, but I hope for their sake that they aren't dumb enough to release anything they've trained on it because I'm sure at the very best that would exclude them from any future previews that SD releases of other models and at worst could maybe result in legal action taken against them.
When inpainting does anyone else save the preview image at like 30% because it looks like 20x cleaner than letting it complete?
I don't. That sounds like either a case of too high of denoising or too high of resolution when inpainting.
hi, how can i generate stuff like that, cuz that looks ai, like which model or checkpoint i need pls
Interesting, it tried to apply the lora to the dog.
common problem with some loras they literally hate any living creature to be in the photo besides the character of the lora. I think 'all' character loras have a hell of a time being combined with each other it just meshes them into a chimera
didn't think Totori could look so terrifying
When Satania brags about her diabolical plan but actually follows through
infinite crazy anime girl art, I am living in heaven right now
There's no metadata on the image so it's impossible to be sure, but that looks a lot like AAM to me. https://civitai.com/models/84586?modelVersionId=89927
You might enjoy these if you aren't already using one of them: https://civitai.com/models/48579?modelVersionId=88330
yep I got and am using both of those
This is the way
You ever see those loras where the lora maker just uploads cursed example images?
like damn man
just generate something with arms behind back or something if you lazy
on the other hand, I kind of appreciate the lora maker being honest about the hands being bad if they really are
I guess that's true
I'm hoping the hands problem gets solved before i die of old age
Hopefully the extra parameters with SDXL will help with that. I still think we realistically need more specific hand tagging than danbooru uses to truly solve hands though.
I honestly didn't realize how much hands add to art before I started using AI art
lol
thanks, wanted to try but i never heard about "base model"
yep, hands and the direction that eyes are looking are super important to a piece looking right.
Has there been any cool new stable diffusion extensions?
reading about sdxl right now, I'm curious about compability with all current loras
bubby the thing u sent me is a lora ? (just to be sure)
base model is supposed to mean that the model was trained rather than being the result of merging other models. I'm not sure that's actually true for AAM, but it will function like any other model that you find. It's really only important for people who are merging their own models so that they can be aware of whether a model was merged or trained so that they can try to avoid merging with the same models if it is the result of a merge.
The one I sent is a model, not a lora. It should have very good LoRA compatibility though, if you want to use them.
It will have absolutely 0 compatibility. Even controlnets are not compatible. SF is releasing their own controlnet models alongside it though so hopefully those will be good, and LoRA makers have been holding onto their datasets to retrain since SD2.x was already out when LoRAs for SD were invented and everyone thought we would be transitioning to it, so hopefully LoRA training will happen quickly.
Type " (huge breast:1.9)" and hit generate, realize the model you are using is trash and try a different one. Rinse and repeat until the models you use aren't trash.
@pastel bluff made his own model that is quite good though.
I'd use it if I did realistic stuff other than TTRPG characters.
Yeah I'm up to date with model training methods
Less up to date with new toys like controlnwt
check pins here. I made a tutorial for outpainting with controlnet.
Great texture!
If you don't mind, can you tell us what this model is?
CRAZY!
super CRAZY
lmao
switches gears
hmm still came out more forest than cyberpunk..
behold the rtx 60900
where do you change defaults in webui? I wanna do batch count of 50 or smth
sick
of course, always glad to share. it's not even an anime model, the "studio ghibli flat 2d cel shaded" does the trick 🙂 (((flat 2d cel shaded anime studio ghibli))) a ((futuristic)) all terrain vehicle ((racer)) ((sleek spoiler tuned)), technology future cyber, dramatic lighting, motion blur speed tire dust streaks, intricate elaborate ornate Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1933746512, Size: 640x512, Model hash: 980cb713af, Model: reliberate_v10, Denoising strength: 0.64, Hires upscale: 2, Hires upscaler: Latent (bicubic antialiased), Version: v1.3.2
adds "studio ghibli flat 2d cel shaded"
result:
ok thats better
and i'll share the prompt (studio ghibli flat 2d cel shaded:1), Tottari (robotic:1.5) , anime-cartoon animalistic young small and short divine goddess spirit of the cyber Tree's and leaves ,, , line art style, intricate, cute & adorable, art by makoto shinkai, pov facing, perfectly formed pretty faces, anatomically correct, golden ratio, , Pixar , Cel shaded, Full body portrait, divine, illustration, soft lighting, lush detail, elegant, smooth, sharp , 3.5D , in style of Mark Brooks
Negative prompt: an6, kkw-extreme-neg
Steps: 59, Sampler: Heun, CFG scale: 9, Seed: 2133425891, Size: 960x512, Model hash: c0986c2136, Model: ElegantRPGAnimeMagick, Clip skip: 2, Pad conds: True, Color Enhance: 0.2, Version: v1.4.0
Used embeddings: an6 [40f4], kkw-Extreme-Neg [5dd3]
prob could be better
very nice
gimme more prompts to play with 😄
I keep meaning to upload that merge to civitai
but i get.. oooh shiney
the style I saved is this bit---> line art style, intricate, cute & adorable, art by makoto shinkai, pov facing, perfectly formed pretty faces, anatomically correct, golden ratio, , Pixar , Cel shaded, Full body portrait, divine, illustration, soft lighting, lush detail, elegant, smooth, sharp , 3.5D , in style of Mark Brooks
I call it soft anime. tho I guess it's more soft elegant anime
yeah I like 3.5D rather than 3d. tho that bit can be taken out
I usually dont add rendering engine or type, leave it to a model, useless I want the efect badly 😄
well make it your own
tho I've just been random and tried all kinds of words or such, interesting results
might need to redo the prompt to unconfuse it
I get to random at times and come out wierd
honestly i love this 80° toony style so much
looks good
for the style, its nice
O wao thats pretty :O
cute
corrected
Wow that's gorgeous
i'm trying to get a wonderful faputa (in made in abyss)
There's a LoRA for her on Civitai that you could try.
I had assumed that you were using an anime-type model.
I didn't realize that you can control the texture so much with prompts! It's fantastic!
@supple raptor Hey nubby you're smart, do you know why my brother is getting weird potato faces?
the problem seems to be the webui itself
Why are my img2img outputs getting messed up? I played with cfgs and lora and it still gets messed up
This is the input btw
He's got (((MASTERPIECE:1.3))) in the prompt. Each of the extra sets of () multiplies that 1.3 by 1.1 so that's equivalent to (masterpiece:1.573) going that high on emphasis can definitely hurt image quality. I haven't used dreamshaper much myself, so I'm not sure how well it handles 1024x1024 resolution, but the very high denoising could be at least partly to blame. I usually recommend not going any higher than 0.61 denoising in Hires Fix with latent upscalers and no higher than 0.52 with ESRGAN upscalers.
I doubt this is actually hurting image quality, but he's using 80 steps on DPM2 a Karras, which is probably at least 2x as many steps as he needs for the initial gen and 4x as many as he needs for Hires. I think any of the DPM++ samplers are also better than the DPM2 samplers personally. I run them at 20 steps for both the initial gen and the hires pass.
could be the model, or the prompt, or just not enough resolution for them.
I think We got it. Tried to enable xformers, but were forced to manually update python and pytorch for it to activate, that seem to fix 70% of it. The other 30% was terrible prompting and sampling steps
hmm demon girl
found someone on diviant art trying to sell ai art like this so i decided to bully them by making more
ahh
been out of town a while - any news?
SDXL 0.9 is in the hands of researchers currently so that they can figure out how to train it and have tools ready for us when it's publicly released. Target release date is mid July. Nvidia + Windows users will require an RTX20xx series or higher GPU, 16GB system ram, 8GB vram. AMD users will need 16GB vram and Linux.
I don't know all the specifics, but Loras can just barely be trained on 8GB vram with some tricks and RTX 3090/4090 can just barely finetune it (I suspect that might require Linux as Windows has significant VRAM overhead).
thanks for the news 🙂
5
Generative fill to the rescue
What is this model? 👀
Wao, free discord from a random shady site, amazin
@dull jackal
Ayyy thanks Sunny 
O, another one 
@dull jackal You got extra work tis morning 
Ayyy thanks 
Rip
Sorry; I'm going to my monthly infusion today
Ohhh thats fair then, no worries, hows stuffs goin? :o
Well, I think the infusions are working but
They're not appearing to do jack for the vasculitis lol
So my meds may end up getting switch some
Oh damn that sucks then :s
How are you?
I'm alrighty atm, had some bad sleep though, woke up quite achey but I'm not feeling achey now :o
Ugh. I so didn't sleep either lmao
I get that achy feeling; moving does help
At least I'm surrounded by cats!
hey
Yeee it does :o
Oooooo, cats are always nice :D
dogs are better
im planning to turn FF XVI into anime
Yesyes, HEAR HEAR
Dogs are good too :o
I love them both
Shouldn't be too hard with controlnets
Yeee same 
im new x.x
so what is controlnets?
https://github.com/Mikubill/sd-webui-controlnet This is the controlnet extension
https://civitai.com/models/38784/controlnet-11-models And these the models
Basically, turns your image into something else with some fancy stuff, and it can use that new image to make things based of the image you give it, like the depth model detects the depth and uses the same depth, there is the openpose models that use the poses, canny uses lines, scribble uses bigger messier lines (like a scribble), reference_only uses your image as a reference (its sort of weird but its pretty good)
we need @magic wave to be mod!
True 
Imma submit my application for mod as well 
Good luck, I been waitin for 84 years 
XD
png supremacy
you have to download all pth files?
Just the ones you want :o
yes
linux > macos
yes
its so bad
idk never really used it
sounds like a problem of the image viewer, just use another one
im new to stable diffusion
I might be talking out of my ass here, but it's probably better for you to use colab.
its funny but painful..
Yeah :o
nice shirt










