#🍥|anime
1 messages · Page 132 of 1
2560 x 1024 
ControlNet for Posing?
I personally use Control Net with Depth Maps but yes, basically anything that can help guide the image without stripping away the creative generation process
I haven't really tried to use the OpenPose Editor and ControlNet with OpenPose images yet
Instead I've been using i2i posing, using poses and details from other images
I also kind of forgot how to get an OpenPose image from an existing image...
(PS. I love working with Depth Maps because it doesn't contain color information used in i2i and I'm very comfortable using them due to my game development background)
some times i using 3D Openpose extention for a1111
Faceless crowd
I've found a decent library of hands in depth map format. People complain the extension doesn't even work but I don't even need the UI Editor to use Depth Maps, I have Krita
you mean library of 900 img depth maps?
yes
Honestly, it isn't the best? But it's not the worst either
It's a collection
There are some good hands in there
I haven't had enough time to actually develop my own depth maps with them yet
Only got it like yesterday or something
But I have successfully used depth maps before and I know how to use them
So I'm 100% safe to say that these can be extremely helpful (if you know how to use them)
Time is always against me...
I just got into work
thank and good luck
Thank you!
Any lady or gentleman going to kiss this one yet?
very atmospheric
Thanks lol, I really like more moody pics
dall3 image
this prompt causes censorship tho xD
dunno how i feel about posting dall3 stuff here, as it's not stable diffusion by stability ai products (i suppose)
I just do it at home, on a local computer or laptop. But you'd need a good GPU
Idk about any free online SD stuff
you mean like this? dream by wombo made
comes up with some good stuff
everyone asleep?
ok his head is not right
more images to compare... model merging takes so much comparing, making sure there's no idiot blocks ruining everything
suits to big but good pic
cute, what model and prompt did you use for such colors?
There's some crazy perspective happening here
wild
Beach Episode
Sorry if this counts as spam
Aaaaaaaaaaaaaaaa
This one is my absolute favorite though
multi-control net?
No ControlNet, I think it might be a side effect from overtuned models or perhaps something about the way the prompt is written out
sneaky eyes looking at viewer
AI just wants people to look at the camera
haha
same with those cat ears
they are a bit backwards
it was online site, dream by wombo
anime v2 , prompt: factory Cat random, epic detailed concept fine art style of mark lee and mark brooks, random epic background, marval inspired fan art
with what model i can get this style?
You might have to try merging some checkpoints, perhaps try merging an anime model like AnythingV3 with something like the Arcane Style model/Lora.
I did see a really cool Chinese/Dragon Checkpoint which could work well if it was anime-fied
Anime style 
Wheres the suffocate me style ? 
that looks like SDXL
most sdxl models, with the right prompting
although, i supsect therre's a lora in there too
Who is this woman?
Peachy 🍑
Shes my local AI template prompt
For when I test styles and stuff

Local peach wants to know your location
Shes trapped in my output folder 
Mangaka
can you send me one pls
sdxl models which are "flexible" in anime output?
yes
what is the model name?
right now, i'm testing a new version, if it screws up anywehere
@nova remnant on a generic sdxl model, anime is totally possible 🙂
but to get the specific style you want, is significantly harder
is that a lora? could you send it to me plz? i really like the flat colors and stuff
I'll upload it in a few mins actually, you reminded me that this LoRA I trained because someone asked me to 
lol
Okeyy thanks
i would train a lora on a style i like but i have 0 knowledge of training
I mean, it isn't too hard, but it can be difficult with some styles
rlly? i always thought it involved a lot of technical stuff
@nova remnant -> anime steve buscemi, art by __manga_artists__
this is all the same seed, so it would take some experimenting to find the artist you like
Papers, I command you to come to my hand
with sdxl or your model?
my model, sdxl based
i'm thinking of uploading this merge, i feel pretty confident about it 🙂
Lmao
Anyway, I need to go to sleep
Goodnight guys
gnight ❤️
goodnighty 
do you not know azumanga daioh?
this scene is actually funny https://www.youtube.com/watch?v=UnktCDi-BVs
the whole series is a riot!
it's a fucking amazing anime
i wish i can say that, i only watched that smol scene on youtube
Smol soosi 
i have like thousands of anime ptw
Luna-y 
how you enjoying the dlc for cyberpunk?
Too many thick peaches


have anime bara schwarzenegger lol
Its really fun 
Though I've been having fun with Cyberpunk in general cx
is it really the redemption arc for cyberpunk? @magic wave
True 
I'd say so 
nice, i should probably complete my base game before deciding if i should jump into dlc 
was a prompt i yoinked 🙂
I'm doing that while also doing some of the DLC stuff cx
but it came out pretty good, i think
yeah, i'm feeling good about this model ❤️ (spoilerr warning, non anime)
Damn that is really dope :o
understandable. I still need to complete so many games, like bg3. But suddenly got back to SD, and lost interest in gaming again
the others are just as wicked, but have traces of nudity
artistic nudity for sure
but alas, them rules are them rules
but this being a general model, that's pretty easy to fix 😄
Fair fair XD
Fair, still dope tho :o
damn, this prompt is TOUGH on any model 😮
well then, time to do a comparison shot, and see if it does what it needs to do
why she dipped her hand into the ramen? 
feeling if temp is right
ramen is finger food now
A beautifully trustworthy woman.
ahhh, i forgotten about the super secret technique to test temperature of ramen, and eat ramen
?
anyway, eminence in shadow s2 ep1 is out
shangri-la frontier started too
the manga is fckn awesome
added to ptw
haven't watched ep1 yet, but if it can deliver the hype of the manga, it's gonna be gud
sadly the score is already at 7.88
basically, dude goes int vr mmorpg, but he goes full dex no hp build 😄
honestly, i'm not even surprised
If he doesn't get one-shot at least 7 times then its a 0/10 manga
i'm literally reading some korean webtoon about similar stuff
of courrse the trope "mc hidden quest line" applies
but he gets cursed so he can't actually wear armor ingame XD
Got a question for yall, what is the most common aspect ratio for ya'll? 
1024x1024 / 1280x768 / 1280x720 -> 2560x1440 (if wanting to make desktop wallpaper)
So 1:1, 4:3 and 16:9 
Gonna watch it later

Hoping for another 2 cour
some slight spoiler that doesnt really matter from manga https://www.reddit.com/r/TheEminenceInShadow/comments/165s4zw/delta_is_lovable_manga_chapter_56/
she's so cute
Very cute doggo
Is there a tutorial available that teaches how to create seamless img2img animations with minimal interruptions?
not sure too much about it, but you might wanna take a look at deforum
Thanks, i can create nice animation but want it to be with less changes and distruptions
i think just using SD will be hard. You might need external editing like aftereffects for that
deflicker effects and stuff
Will check that, thanks 🙏🏻
giga insane prompt that kills any model
followed by cute girl 
you can only come back 6 times
wait no wrong game 
the ones I can do with my gpu
Covert kills I take it?
smol sussins creed
I'd like it to be with anyone at this point, just not a floating house
I'm trying some AND_SALT magic
good luck 
Which are? 
Whats AND_SALT?
This really isn't something that's close to being solved yet. Some people have gotten super clunky workflows to sort of do it, involving multiple controlnet layers, but it takes more work to set up than it would to just use traditional animation methods. You might be interested in fooling around with AnimateDiff which operates in txt2img, using an additional motion module to pull context from previous frames.
3pixel by 3pixel 
https://github.com/ljleb/sd-webui-neutral-prompt for this extension
the dagger looks edited in
imma be honest, i have no idea how that extension worsk
AND = bad
AND_PERP, AND_SALT, AND_TOPK = less bad
all are honestly pretty useless for 95% of gens

O, fair then xD
Whyssat? 
They rely on the concept of basically having separate prompts fight over latent space which generally contributes to incoherency. The different versions essentially change the rules that decide which prompt wins the fight.
If you want images that look very much AI generated then they can be interesting for sure. If your goal is images that look like they might have been drawn by a human then they will almost always be worse than normal prompting.
Their main usefulness comes from using them with regional prompter to help control image composition better.
@supple raptor yep very AI crazyness

hmm this makes me wanna... prompt thick thighs

@lyric zealot 
She is a powerful mage.
tree house 

https://civitai.com/models/156745?modelVersionId=175959 @wet elm Here be the style
if lora from smile = download it

just do image2image at low denoising strength (0.2-0.3)
maybe put highly detailed in prompt
If all you want is upscaling without any diffusion WebUI has ESRGAN, Real-ESRGAN, and SwinIR, and ScuNET upscaling built in. Grab an appropriate upscaler for the image from https://openmodeldb.info/, drop it in the appropriate folder, and head to the Extras tab. If you're upscaling an AI generated image then you're most likely better off using Tiled Diffusion or SD Upscale to do light img2img diffusion along with a good upscaling model.
yoo ty! also whats Gishiki?
it's a model
isn't that aka holokuki v3?
Sort of, it's the same creator
image 2 image upscale from 1k x 1k to 2k x 2k
try broom riding you'll quickly see that one is very coherent compared to most
I got a lot, and more that aren't on this grid that I trained after I made it
Gishiki is gud :o
btw these are all full res, so you can zoom in to take a close look
i should really try it, but not used to prompting with models like Gishiki and even Auroa
Its the same as holokuki 
really?
Though I can related that like... dunno, you grow a bond with the noise of the models, you kinda vibe it out 

ono she fell off the broom 
double ended broom is the way of the future.
hmm... i don't think that's how you're supposed to ride a broom
my local ai bot
Anyone happens to know a LoRA to generate characters with Balaclavas?
What is Rem the Maid doing here?
hmm let's see you are in the anime tab and rem one of better known overrated waifus hmm tough question...

that has to be a fanmade animation...
such great generation, i wonder whose pc you using 
stretches and yawns and looks around
tries not to play any more starfield and make art instead
that game has invaded my dreams. ahhhhh
tries to make things more anime
Booba
dont tell ppl how i live
i need help with hw
im good at math nothing else tho
i need help in anatomy 😵💫
Biology*
boob*
I think I'm gonna be using this emote from now on 

When I get back home from work I'll turn it into a sticker
i just need help with this paragraph 
While sitting on a desk
multi-dimensional , double exposure, Mommaries
What on earth is happening in this image
shrugs
steampunk : multi-dimensional Anime action , double exposure, Mommaries, time, clock, space
hmm transformation scene?
Seems pike something straight out of Fate Grand Order: First Order

heh, I wasn't expecting that one to do as well as it is in the contest
Lmao, well it got my vote
prompt: wet clothes, off the shoulders, Spooky style, multi-dimensional , Anime Miku's delivery service of Eros
ok I should be in bed
thuds
goes to bed
sleeping with adhd brain
tried dalle 3
Seems gud 

With a bit of editing this could be a great wallpaper
my generations are lacking colour like mostly grey how do i fix that?
also what model is that looks amazing
a custom merge i made
not up on civitai
thanks
for colors on your gens, you don't need any more than
Positive: (masterpiece, best quality)
negative: (low quality, worst quality:1.2)
ill try those out but its happened before always looks like the colour was sucked out of them
like it didnt fully print?
only more details but its never done this before
i stopped doing this for a while
getting back into it cause i wanted to make my dnd character
ahh ok, a lot has been updated since last month
xformer new version is out, a1111 got updated and many of the basic and essential feautres in it
i got 1.41
i gotta update it
do you have the link to the new version?
no link, you'd have to type in a command in the console to update it .. you should ask in #🤝|tech-support either Aryetis or CS10 can be great help
ill ask in there thank you!
welcome and gl .. should be easy to do and you will like the new changes
what are the new changes mainly?
to mention the big ones, render speed and optimization along with sdxl native support
oh my god thats probably why it looks greyed out
but i'm still using sd 1.5 since sdxl by default requires lot of resources
grey tone is not cause of the version
if you are getting washed out colors you might be using a deprecated feature face restore
you could also try with adding monochrome, grey scale in the negative box
ill try that on this generation
are you using it with VAE?
hi smol sus ~

nice catch.. vae could be a reason too
yup, definitely a possibility
yeah i am tiled vae
but settings i used to use so it should be fine

i think you can use tiled vae without actually using a VAE model
i might be wrong about that tho
or you can show us a generation of yours with metadata, so we can see your generation parameters
would recommend using this vae for anime in general .. https://huggingface.co/nubby/kl-f8-anime2-blessed/blob/main/WD1-4-kl-f8-anime2-bless1-05.ckpt
whats this?
ill try n grab em for ya
that's a VAE specifically for anime related renderings
ill give it a look!
you would basically need 2 common vae for most tasks with SD 1.5, the one above and this https://huggingface.co/stabilityai/sd-vae-ft-mse-original/blob/main/vae-ft-mse-840000-ema-pruned.safetensors
this is the only img i could grab the others i deleted @lyric zealot
it kept doing generations i didnt want tried making my dnd character and my other party members characters

In the metadata, i don't see VAE. So it's highly likely that you are not using VAE for that
i might of got the vae confused did u mean the bottom on on the generation page?
that's not what i meant. You can keep that on.
Colors actually look alright here tbh, could just be that the model doesn't have a lot of saturation
something like this, but obviously that at the top of my webui because of my setting
that dosent seem to pop up for me?
never seen it before
yup, actually the colors isn't that bad
it mightve been a bad load but it was worse in the other ones
like waayyy worse
found this one it was these type of levels
under Settings > User interface > Quicksttings list add sd_vae
not sure if it's just creative monochrome, or lacking VAE
it stopped happening on the most recent load, but it pops up here n there alot of the time
like 90% of the time
btw do i add sd_vae to this?
just press the empty space and type
actually, you might not be using VAE, but the model you are using probably have VAE baked in
im using dreamshaper
That's just monochrome
yeah thats the thing i never add monochrome to theme
them*
Just happens sometimes if you don't put monochrome in negative
And sometimes still happens if you do 😂
If you use a checkpoint without any vae it'll still have color but just be super muted
im surprised ive never heard of vae when i used to be really into ai art
Yeah a lot of checkpoint creators are terrible at putting descriptions, many checkpoints don't have baked in VAE but the uploader won't indicate it anywhere
that must of been why most of the generations with a model i tried look so bad
@lyric zealot do you recommend and vae?
the one im using should be one of the more popular choice among anime generation #🍥|anime message
im just doing 3d rn for the dnd characters but would u have the link for it i cant seem to find it
thank you very much!
@lyric zealot Lil soosi, you use any SDXL models? 
seldom, i don't have the vram for that 
unless i used comfyui
O, fair 
how do i make skin grey colour it will just be peach colour
i feel that

that is amazing 
that's using the VAE smol sus linked you
damn it i gotta upgrade my version
Am trying a SDXL model and my VRAM go brrrrr

Well
My PC, audibly, goes BRRRRR
XD
but idk how only way i could was a fresh reinstall of the webui
yooo nubby long time no see!!
In your stable-diffusion-webui folder do you have a .git or .github folder?
both?
one is grayed tho
use it in comfyui, might be a lot better 
perfect, that just means that one is a hidden folder
Probs, Imma try
Here's how to update:
- Close your WebUI CMD window
- Click in the address bar from that folder and type
CMDthen hit enter - run the command
git pullin the new CMD window that opened - run the commands
venv\scripts\activatethenpip install -U xformers - wait for that to finish
@worthy delta ^
Impressive. Model?
I always just use the official SD 1.5 vae. Is that weird? Can't tell a difference.
maybe I should do a few test comparisons
it says "operation must be run in work tree?
The base SD 1.5 vae is pretty bad at anime style eyes and tends to be a little worse at small details on flatter artstyles on anime models. It has the best colors though.
What's the difference between WD1.4 "e1" and "e2"?
how do i do stage 4 in the cmd window?
cause it says "system cannot find specified path"
e1 = epoch 1
e2 = epoch 2
naturally
do you have a folder that's called venv inside the stable-diffusion-webui folder?
where do i do pip install -U xformers?
the cmd window?
yep
gotcha gotcha
That's just updating your xfomers to the latest official release.
in terms of colors and lighting also i found this the other day
does it take long to do so?
seems quite cool
is that colour adjusting?
yes
hmm idk maybe the link above me...
ah yes the floor here is made out of floor LOL
i didnt realise u sent a link mb mb
This one #🍥|anime message
the 2nd image and last one look soo good
imma be trying dat one
hey nubby and chat
hello cat with 4gb vram send help!!!
is there any good models i should look out for?
anime style mostly
looking to do my dnd characters
some 3d styles aswell rn im using dreamshaper
@supple raptor u udate kakigori to sdxl?
update*
nah, no plans to. I don't think XL is worth the extra resources to run currently for anime since there's no good base model. also, training on XL seems to be a huge pain in the ass from what I've seen.
you can really tell the difference? 😐
ok
yeah i got a good eye for small diffrences
also nubby the vae hasnt popped up even tho i enabled it?
yeah v2 improved a lot about the model without many compromises.
no safetensors? What is this, 2022?
aye thats good to hear
how do i use vae???
it dont be popping up
if you're using webui go to settings and vae and set it manually
can also set to override the model vae
it could be that your model doesn't have a vae so it's just..not using one
you may need to restart the webui after adding sd_vae to the quicksettings list.
yes
i did restart it alr
updated it shoulda helped the restart
ohhh yep found it
Well, first VAE: None is terrible because it's blurry compared to the rest.
Second, the color on VAE: trinart is closest to no VAE, so I'd rule that out.
For me, I think the ft-mse-84000-ema-pruned is the better of the other 2, but they're nearly identical.
was in the settings
the WD bless vae nubby linked is nice bc it doesn't give weird textures or too vibrant colors
yeah, trinart's vae is bad
which is best to use?
the bless
the differences aren't huge anyways between vae so I don't think you'll ever have to switch
lol
well huge in terms of rendering
Yeah I can't tell the difference between vae-ft-mse-840000-ema-pruned and v1-5-pruned-emaonly.vae. They seem identical. Maybe I'll actually check programmatically.
it's just ...like .1 hue
yes
check the override box
or just uncheck and check, generate and see on which settings the vae is working
can vary depending on model blah blah
but like I don't think vae "stacks"
so it should be fine
thank you very much
Ok I checked. The images are exactly the same between those two VAEs. 🙂
nope check the right eye
your right
like right side
v1-5-pruned-emaonly.vae and vae-ft-mse-840000-ema-pruned are the same man
you know what i mean
one eye is more red
Eyeballing it, I figured that might be the case.
either that or im trpping
the vae files themselves, though, seem to be different. So, not really sure what the deal is.
yoo that looks nice
what model or is it custom
my thing messed up n it made grass
HolokukiV2 base model, expmixlineV3 model as refiner, switch at 0.5
oh its hella fucked its just making grass
here's a metadata for you
thank you
how do i stop the grass
running another image through, including the other vae this time

.
it just keeps making grass textures
( best quality, masterpiece, 1:4 ) 1boy, handsome, rugged, beard, brown hair, brown red eyes, face scar, rugged scarf, old armor, rugged armor, rusted armour, dragon scales, scales on neck, red scales, beautiful, blood lusted,
cfg too high/low
you are not using weight properly btw
should be (best quality, masterpiece:1.4)
if you want it to be at 1.4
correct. they overwrite each other in RAM so only one can be active at a time.
based on those names, they should be the same VAE
yesh, kakigori do the cute
very cute
good job nubby
imma stick around doing ai art
this fun
idk why i stopped
they have different hashes, so I dunno
nubby vae 
might be minor rounding differences from converting or extracting 
yeah that one has better eyes, you can tell
blue eyes instead of brown, very different

a lot more shine next to nose
better folds at the end of the sleeve... wonder if that was the vae or just a generation outlier

Still prefer 84000. The colors on the WD1-4-kl are just too subtle for my taste...also, something happened with the hair right between the eyes that I don't like on that one. And again, trinart's just...off.
c4t
doing great 
increase weight?
i don't even know what's gray skin tbh
I can't visualize that
its like a dark elf from skyrim
pale skin?
anyone know how to get gray skin
I would try (colored skin, grey skin, dark elf:1.15)
ooo u ever do pixel art
I have in the past. It's been a while though. Let me see if I can find one
aye pixel art
heres my accidental rug art
idk why it did this
again
fixed it
removed bloodlusted

bloodlusted
dayum lookin good
Those are from back in april
dayum april is like long ago

fr though if you like sad anime then Your Lie in April is great
my heart is fragile a silent voice broke me
A Silent Voice is also great
never recovered
You should watch I Want to Eat your Pancreas https://anilist.co/anime/99750/Kimi-no-Suizou-wo-Tabetai/
still need to watch that 
thats a weird name
it makes sense after you watch it
I post with metadata so you can just open one of my images in browser and drag the image from there to PNG info tab to get my prompt and parameters. That last Flandre Scarlet one is sort of an exception because I upscaled it with a super basic prompt.
I see I see ill try it out sometime
I wanna think of stuff and make it come to the keyboard yknow
But all I get is blank
When I can't think of what to prompt, I just go browse images on Danbooru or wherever and find one I like then try to describe it with tags.
Ooooo that's pretty good idea
I'm making my dnd party's characters
And they turning out great
Here's my dnd 4e characters that I made with Dall-e 3 the other day
dayuuum
And my adnd characters which I made in SD.
we doing 5e
looking fabulous
what model did you use for them?
my current dwarf
has two souls in one body
so he gotta sort it out
Those ones were made with Aurora using a sketch style LoRA. Top left also used an Alphinaud Leveilleur LoRA because a friend joked that I was making Alphinaud when I was rolling the character so I just went with it.
i gotta get aurora again and what was the lora called?

and here's aurora https://civitai.com/models/40199?modelVersionId=45601
guess where i saw that exact generation 
Use smaller base resolution with Hires Fix to upscale it. 544x960 is usually good for a 16:9 aspect ratio
Lower the height resolution
ill try it out thank u
wont let me send the negatives
trying to make a female character without nsfw popping up is hard sometimes
here to cheer yall up
its giving kill me with a knife vibes
and maybe gut your inside, and eat it vibes?
yeah put it in a blender and add bananas vibes
banana overpowers other flavours in a smoothie
screenshot your prompt
it got nsfw stuff in there im tryna block so i wont get banned LOL
oh bet
what you mean
them fully black ?
yup
From what I heard that is hard
way too many negative embeddings. pick no more than 2
ti's?
ti = textual inversion = embedding
im tryna get is a very specific way
eye i mean
turns out like this

might be better off recoloring the eyes in post then doing a low denoise inpaint over them if needed.
yeah prob gonna have to recolour in post
spooky
experiments have gone wrong
you have gave me a dnd idea
mimic dice
i wish guys got alot more focus
like cmooonnn lemme customize my mans
men look kinda bleak in most ai art sadly
listen cat with 4gb of vram
bloodlust isnt a feeling its a look
also dreamshaper
is making the multiple images
what do
what do you mean?
are you using hires fix?
yup


👍

