#🍥|anime
1 messages · Page 151 of 1
I don't think there is even a .gif
hi what did you use to get this style
you can download the image and see png info
will try ty
Hello community... I need help locating these old versions removed by the author... if anyone knows of a location still available... thanks in advance... (note I cannot use the backups available on Telegram):
breakdomain_M1808; breakdomain_M2050; breakdomain_M2150
there are some on the tusiart site
ehy guys, someone that already toyed with ssd1b and know if out ther there are loras and embedding compatible with it?
waht is ssd1b
they removed some layers from sdxl to make it lighter and faster
Is it worse than normal sdxl ?
Like is it not as smart with prompts or something 
they say the loss is minimal
what layers
@supple raptor pls explain
this is my go to
i looked a bit into the ssd-1b, since i thought it was just a sdxl tweak. But apperently auto1111 support soon so will be easier to see it then
was not impressed with the demo they had though
well, i can run it on a 4gvram and that's something
the final result was polished enough, without too many mutations
thanks love
Kazoo... don't let reading comprehension become a lost art.
what layers
the conditioning of comfyui stopped working 😦
cool concept but sdxl currently im not happy with, not speed wise - just it feels like its walking through mud on prompts
less creative?
it feels way more likely to just flat out ignore tokens
very apperent when using it with ipadapter.
i just love how good SD can be if used with a good workflow
you tried w/ anime?
I haven't tried at all, just providing reading material for kazoo.
are they also gonna tell me that SD 1.5 is faster than SDXL?

also that graph does not look 60% longer
who made this graph? a middle schooler?
number is the size of the line isn't
wait actually maybe it is
its literally is
🤣
someone have been watching a bit too much ltt this morning
oh no my staff
out of curiosity, how long does it take you to generate a 1500x1500 picture with 4 gb vram
lol

I'm on amd
I can't go past like 700x1050
and hires basically doesn't work
Im on a very old amd card, rx 580
very very old
comfy requires linux for AMD
tho from my testing linux is the way to go with SD regardless
ROCm doesn't support my gpu anymore 
so, do you have even less than 4gvram
basically
its worse than 2gb on nvidia
yeah, i can see that
with 4gb i'm already pushing the boundaries of this thechnology
in performance wise
i doubt is possible to do anything, with lass tha 4rgvram
My main thing to pass time with is games and not sd, still fun to use 
At least there are several services which are free/limited/not too expensive to use which dont require a beefy computer
hahaha, yah that do tend to help.
This is an amazing image.
hmm now thats interesting
yeah.. not sure why
im bored
Not anime????
I don't watch anything
Based
Unbased
I duckduckgo this now
Oh I know this one
It's cyberpunk

slay the princess is gud
just full released like a week ago too
Yeah it's all over the place on steam
1gril

imagine the asmr potential
wait

eating cheese asmr
🙇♂️
idk what happened here but this isnt a girl sleepin on a couch
Catgirl
is there any way i can make the leaves more detailed?
high res fix
HF: https://huggingface.co/OedoSoldier/detail-tweaker-lora This is a LoRA for enhancing/diminishing detail while keeping the overall style/characte...
what a beautiful girl on a couch
Use Tiled-Diffusion to upscale the image in img2img. For the prompt you'll want tags related to the leaves such as outdoors, nature, forest, tree, grass, masterpiece, best quality don't include any tags related to the girl. Set denoising to ~0.34.
try with high res fix set at 4x 
then do img2img ultimate sd upscale
oh its a plugin
ultimate sd upscale is kinda pointless. it's not really any better than the regular SD upscale script with proper setup, is considerably slower, and both are worse than tiled diffusion.
am i supposed to select regions in this?
what region? asia? idk what u talkin about
thats for regional prompts
no, turn off region prompt control. You don't need that part. Just set latent tile width, tile overlap, tile batch size and scale factor. These are my settings with 12GB vram.
what do u suggest for 6gb
This should be safe. Basically you want the tile sizes as high as you can get without running out of vram.
so much to learn
and people say ai art is easy lmao
not dissing actual artists
they are awesome
to be fair, it is simpler than learning to draw, but a lot more go goes into making good AI art than simply typing some random words into the box and clicking the button.
set tile to max size if u get oom reduce size a little bit then try again
after 30 mins i just generated the same image but with lower saturation lmao (prolly cuz vae didnt get applied or something)
Time taken: 31 min. 28.7 sec.
here's my take on upscaling it
where did i go wrong?
did you set the scale factor to some number higher than 1? I used scale factor of 2 so it's 2x upscaled.
yep mines 2
i 1:1 copied your config
this is what i got
seems like no vae applied and very identical to the original image
it did upscale the image and made it more clear, but i wanted the image to have more 'bushy' leaves
You may need a different model if the one you're using struggles to do the style that you're looking for. A more realistic model might work better.
try to increase denoising to like 45 also increase steps from 20 to 60
yeah, this model creates characters just how i want them to be, it just lacks details in background
hiya
i disagre
ha watching a parade in japan, they are playing the Godzilla theme song
Nihonbashi Festival
now I gotta make some Godzilla art
rawr
Unet layers
Where was that written
How did I not see it?
I went to the thing you sent
The blog thingy
Didn't see it
I saw the first part
Then skipped
They explain it so simply compared to the research papers
I could not understand anything almost on the research paper
rawr
I blame my reading comprehension on my ADHD and dyslexia
This is a gouch
A gillow
Wtf did you put in the prompt for that to come out?
1girl lying on red couch
What model
several models i only did it to try diff outputs with all my 98 models
to see which one could handle the couch better
another one
at least this one added the buttons
Ottoman-chan is the only one for me
my beloved
vae issue
is it possible to use the detail enhancer lora in img2img?
yes
in txt2img i used detail enhancer and got blurrier image
it is with detail prompts and highres fix
huh, much better. Guess I'll try to find good vae then, thanks @native halo
yoooo got a real nice one
highresfix > photopea > img2img upscale
my fav girl
parcel 1 of 5
i have to assembly it myself 
looks like you disassembled it yourself
I feel like you might enjoy this dude https://twitter.com/subway_meat
Praying for clothes malfunction

Been using stable diffusion for 15 months and just learned you can right click generate and make it generate forever
Why can I not find a thing that does all the backend things for you to be able to use https://cloud.google.com/vertex-ai/docs/vector-search/overview as a normal user
Like why do I have to make the search queries into embeds myself
Just now
Bruh
Man every couple of months someone says that
Fr
RIP the back part of the car.
What if you infuse the girl and the car together
Gar
{1girl|car}
I think that was the syntax
But you need like an extension
Prompt blending script
Though this is old
Not sure if even works
You sleep?
I don't know enough about how it works to write out an actual explanation so here's the github repo that covers it. Basically they're stripping out Unet blocks to reduce the parameter count and size (and therefore also reducing the hardware demands) of the model.
https://github.com/segmind/distill-sd
More Agawa Ryo style 
I should participate in that CivitAI contest
Though all my LoRAs are anime and I got like 30 of em
Meow 
Hewwo smile

The Lora connoisseur
Hewwo Cat 
That contest is perfect for you
thats why i told u to try with ultimate sd scale it takes less time,with the one u used if u have 6gb vram u need to use tiled vae which will make the whole thing even slower
setting it well, will do it even faster
do it then
Daddy Coronel Sanders
kenta-sama 🙇♂️
yes
beanie with horns 

so my extension > Available > 'load from' not working, its spitting out error
URLError: <urlopen error [WinError 10060] A connection attempt failed because the connected party did not properly respond after a period of time, or established connection failed because connected host has failed to respond>
any idea why?
try #🤝|tech-support

ah, didn't see that. Thanks
This one good
Yeah yeah that's also about what I understood.

how much cfg would be too much without the tresholding extension
try it yourself, but i've seen 10 quite frequently on older models

been doing 9 images on +1 cfg each
instead of a plot
I'm stupid 
well, if it works, it aint stupiod
@nova remnant
. I have done gen by mistake with 20-29 in CFG without problems .. but without benefits either....
sooooooooooooooo cute
cute type of scary
here in my gar͙̟͠a͓̻̬ą̭͉aA͏̳ͅA̦̰͚A̴a̮͚̝a͡aag̤̘e͓̖ͅ just bought this uh..new lamborghini here
What prompt did you use for that shadow over eyes effect?
shaded face
Thanks
but doesnt work all the time, so good luck
Hmm I'll try
anyone here that does comissions? i'm looking for something pretty simple, DM me if you're interested
if you are already here, why not try your hands on making it yourself with SD. Unless it's super complicated, it should be do-able.
I want that
interesting fashion
mike
hello
🐭
mooooooon
no booli my mouse girls :<
Mouse hoodie 
guys, do you know what settings is need to touch to make face detailing faster?
Talking about adetailer?
yep
This is at least one
working to improve the workflow, adetailer is a pain in the buss, when adetailing bigger images
I find it's only worth it for face adetailer
monster hunter fashion 
Say I generated an image, I like the image but it has bad hands and ears, is there any way to improve just that part of the image?
inpainting is one way, but for hands, it never really work, unless you can self draw a little to "guide" the AI
Thanks, I'll try and see what happens
good luck 
hi

evenin'
Smol soosi 

doing great, now kinda looking into the smallest LLM by stable
Oooo? 
I didn't know they realeased a smol LLM 
Smol sus helped develop it
good question
@supple raptor




True 
Nanami from Riddle Joker
Chibi Godzilla Boy
hmm Chib instead of Chibi gives interesting results
or it just auto guesses and uses Chibi
does anyone know how to get that lighting effect?
I swear I have been trying for like 3 hrs now, but mine doesn't even look close to that good
huge
1girl Big breasts, blond hair long, red leather coveralls, tanned shaded skin, standing in a flower field with the snow alps mountains behind her, (style from this point .....> fine art, soft colors, black outline style, elegant, very detailed Cel-Shaded With shaded detail's, soft shadows
added Tanned Shaded Skin to prompt
takes out saves style and leaves the rest.
Model: AOM3_orangemixs, VAE: clearvae_v23.safetensors
VAE: mangledMergeVAE_v10.pt
tho not sure which part gives the effect you want
may have to play with it
1girl medium breasts, blond hair long, red leather coveralls with white button up shirt underneath, tanned shaded skin, standing in a flower field with the snow alps mountains behind her, fine art, soft colors, black outline style, elegant, very detailed Cel-Shaded With shaded detail's, soft shadows, Midjourney style, hypernet:mjv4Hypernetwork_v1:0.8octane render
Negative prompt: nsfw, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, artist name, one tail, no extra legs, kkw-extreme-neg
Steps: 40, Sampler: Restart, CFG scale: 7.5, Seed: 622867340, Size: 512x736, Model: protogenX53Photorealism_protogenX53, VAE: mangledMergeVAE_v10.pt, Denoising strength: 0.5, Clip skip: 2, Hires upscale: 1, Hires steps: 14, Hires upscaler: Latent (antialiased), Pad conds: True, Version: v1.6.0
oops hehe
it's going on 6am, off today so i'm gonna sleep now. thud
But first Miku Tatsumi
not sure where tatsumi came from... shrugs.. nite. morning. whatever
Funky style, I think this style is not meant for this
is there a way to generate images based on a specific character?
yes, by using loras of a certain character
example:
https://civitai.com/models/16250/princess-peach-or-character-lora-1499
Mood
damn that looks lit
and the hair detail
btw, do anyone of you know any negative prompt to stop characters from appearing when i'm generating landscapes?
my current negative prompts for landscapes are
abstract, amputated, animal, face, artificial, asymmetrical, cartoon, cgi, cloned, collapsed, deformed, distorted, foggy, grainy, human, face, incomplete, low quality, low resolution, lowres, malformed, missing parts, monochrome, mutated, over-saturated color, overexposed, photoshopped, scary, silhouette, sketch, surreal, twisted, underexposed, unfocused, mountain, lowres, text, error, cropped, (low quality, worst quality:1.4), normal quality, jpeg artifacts, signature, watermark, username, blurry, monochrome, BadArtist, BadArtistAnime, BadPrompt, BadImages, (monochrome, grayscale, sketch:1.2), (logo, text, words, artist name)
you can just add (no human) and (scenery) to the prompt, that will work better than negative prompts
Just do (low quality, worst quality:1.4) instead of all of that quality stuff
I like using that and grayscale, watermark, text, border
okay, let me change them
And the textual inversion
They are the ones doing that for you
Like the (bad hands)
what's your prompt for landscape?
((masterpiece)), best quality, 4K, HDR, Realism, VFX, Volumetric Lighting, Volumetric Light, Volumetric, Natural Lighting, high details, detailed and intricate, intricate details, high intricate details, absurd amount of details, absurdres, high resolution, [volumetric fog], high resolution, Medium Shot, Single Shot, Shallow Focus, High Angle Shot, Front View, soft shadows, picturesque, scenery, fall season, dark sky, no human, Victorian mansion surrounded by forest and hills, paved road to mansion, small pond, foggy, camera in font of mansion
just got this
I'm mostly satisfied with this, except the watery road
any reason why you were using "[ ]"?
and also, "no human" probably isn't working the way you intended
I started learn 3 days ago, so I'm mostly scouring reddit for good prompts. I found it there
well, then there's really no reason for that square bracket there. You can read here https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features for auto1111 features
but for square brackets, usage will be to go from one prompt to another. Example, [cat: dog: 10] , means to have cat in prompt for 10 steps, then change "cat" to "dog" for the remaining of the steps
or another usage is for the alternating words feature. Example [cat|dog] would generate cat for 1 step, dog for another and continue the loop
ah, got it. Time for reading documentation, i guess
yup, it's always good to read the documentation for auto1111 to get some useful features
hate reading those, college gave me trauma of documentation
like, you can highlight a word, press Ctrl + arrow key up/down to increase weightage on the prompt
well, some useful features are there, or you can just explore the webui yourself
nah, documentation is way better than trial and error
[] is the opposite of ()
but I'm not sure it works at the moment
ye
@winter fable do you have those textual inversions downloaded and in your embeddings folder ? >.>
only have these two, and I rarely use them

In your negatives, hopefully old negatives: BadArtist, BadArtistAnime, BadPrompt, BadImages
Those are all TIs
You need the files to activate them
At least I believe they are, you could've copied them from someone who doesn't have them and just typed stuff x3
almost 90% of negative prompts I have copied from somewhere else
it's okay to copy borrow from others here
but, models nowadays don't really need that much negative embedding/prompts
@winter fable download this one https://civitai.com/models/56519/negativehand-negative-embedding its gud
use that and easynegative, should be enough >.>
other than that, only prompt negatives you don't want to see in the image, like I did the other day with a model that doesn't really know what a dark sky/night is in the positive "blue sky, day time" and it magically worked 
thanks for the help man
easynegative have a certain style, so if you are using it, make sure you like it first
you can do a plot xyz to see the effect with and without

@winter fable 
I put bad drawing at basically nothing weight so it doesnt do much
forgot how to do with and without
lmao
Manmaru mix 
lotta 

loving this theme
Asuka abandoned the Japan thing and switched to belly dancing for middle eastern oligarchs (many such cases)
jk
my teammates be like
Modern witch doing her magic on the internet
bruh i just launched sd and my images are generating like this
internet warrior
did you fix it
@gritty spruce
bad person
Done thanks
nope
have some pumpkin fun (a bit on the not-so-much-clothing side but quite tame compared to other stuff i've seen here, so as long as the upload bot says ok...)
ask him
needs more pumpkin sized features

more pumpkins would be indeed better

its not
Its just cyan hair girl
Come on you need to pay attention to the class
you can by just drawing on top of each frame
comfyui is the one who adopted that first
thats the funi fbi reference when the funi
Everything is a honeypot if you're schizo enough.
prob the skin color chart the feds use 🤠
I'm a middle eastern Neet, we turn green after Neeting long enough.
They need to update skin color chart
I hate that the ai can draw better than me but at least I can just copy it and use it for reference 
also alien girls
You can
You know
Fix it
Well I think the best ai art also requires some art skills
usually over painting, color stuff, some Photoshop, making a sketch and then polishing with ai
at least anime wise
I'm not saying most are bad
it's just
has its downsides
like not being specific enough
weird limbs
ears etc
bad pose
It's a ploy by the government to fill the prisons with 😭 enjoyers.
can't catch me they don't require fairy registration
Sunk cost for profit prisons
They don't care about ethics
They'll get you for profit

You me good sir would find yourself right at home on the Hydrus discord server we too hate the feds
they cant catch me if i cut my legs and arms🦶
Sheep
just don't plan any "big moves"
lol but I don't really know what you mean
more mousey
Portal to another world 
nice ink style
used a Ink scenery Lora on that last one
changed order of style's and wow
You can check the lora mix I use from my meta data
Also different models yield diferent results obv 
Do some black lotus

Like the magic card if you know
ok enough naruto. hmmms
hmm Black Lotus Ninja.. hmm yeah
well that came out wonky
FannGanna?
Demon Miku
dragon ninja of the black lotus village
hmm actually that could be an interesting story.. ponders
takes out the mdjrny4 lora
ok doesn't really work well without it , so I put it back 😄
goes back to bed.. thud

card name lmao
A store, this reminds me of something
Been getting decent results I think. Using a mix between CGI and Anime though for the prompt.
Making some backgrounds.
what model?
Switching between these two. SDXL Base and SDXL UnstableDiffusers.
I think the majority still use a1111, I use ComfyUI as I prefer a node based setup. A lot more customization as well.
newest driver supports disabling shared memory fallback feature https://nvidia.custhelp.com/app/answers/detail/a_id/5490
Any SDXL chibi Loras out, forgot to add that type back to my files for SDXL.
yeah
pretty hard to not get one giant moon
Do you have any tricks >.>
Tried some negatives like large moon but to no avail it seems
not scary anymore 
It's mostly just a problem that some models have. Negative prompting sunset might help on the model you're using since it sort of looks like it's trying to form a sunset with the moon. If you have anything like official art, movie poster or somthing else similar, that might also be influencing it.
my tiny moon 
this pretty cool
I like this image very much
Bro using discord on the counter strike 1.6 console
haha
close enough
I also finally set up ssh between my pc and phone
I just realized that an android phone uses more ram on idle than a normal/basic linux desktop environment
like how tf
bloated
I wish I lived in a country where they actually did halloween
android is intended to cache frequently used apps so that they'll load faster when the user selects them. in an ideal world, your RAM would always be nearly completely filled, and unneeded things would just be dumped when something actively needs that RAM.
unused ram does not help your performance in any way and is actually worse because that means the app that you're very likely to load has to be loaded completely from storage instead of from RAM.
oh thanks for info
That goes for all OSes by the way, not just Android.
in case I need ram (for example for llms) is there a way to clear the cache?
windows
windows is trash at memory management unfortunately. It just starts throwing the new stuff you open into pagefile instead of clearing out the old shit. only real option for windows is to start closing shit that you aren't using
for android
don't run your llm on android 
but I wanna flex to friends and family
Run it on your PC and remotely access it from your phone. They very likely won't even understand the difference anyways. 
But what if no internet
What if not on same network
my isp shit
no port forwarding
set up your own private vpn. being on a different network should be a non-issue
oh yeah I know that solution
hamachi
or openvpn
my new waifu
i feel better now its just that my legs still feel weird
yes just stuff her in your backpack🤠
It's just some forest/red moon/fog prompts and lowra Lora at 1 strength
I know how to get those to fit in my backpack very well
@nova remnant Touhou server made me commit warcrimes.

Okay Google uninstall discord
ok I tried openvpn
needs bigger head 🧠


Have you tried using windows
👍
im back son,btw im pregananante
👍

tried also hamachi but they don't have the vpn feature on their android app and even though they have an arm linux package it doesn't work on the android filesystem (perhaps can use through proot-distro or something but it is not worth it)
I could just use something likke ngrock tho
it won't change anything
this is related to my isp/router
I am terrible at estimating time. uhh, like a couple weeks probably. I like the colors, I made it myself.
how you made it?
I used Clip Studio Paint and was just messing around with colors. Anything good I make is practically just luck.
I like blue and purple so I just did like a gradient between them
nice
I'm also a paint pfp 
deep and cinematic
Pretty, I wish I could paint a cat that well.

i hope u ready
bruh
wanna practice clarinet?
Doesn't exactly follow the prompt, but I guess it's alright for the first image in a bit.
for a firsst image in a while this is pretty good
its ok spongebob just one more lesson
I haven't tried getting those optimizations from nvidia, is it worth it?
I think there was an extension or something
@autumn violet 
also
Why does it looks like jets are flying all over when there are clouds
can't escape the contrails
chemtrails
all those butterflies will have cancer in 2 years
Cottoncandy
I'll take both 
my fav episode of spongebob
lemme help u with that giant buttfly
I should get to work learning to 3d sculpt, I think blender can do it now right?
maybe there's a plugin that can also take an image and create a crude model template
so uh, automatic still the go to or is there a new, better version that's not bloated but gets new experimental stuff?
I've been gone from this awhile
ok before I goto bed
not what I was going for but .. i'll take it
thuds
I mean there is comfy
And it is less bloated
And has some more experimental features
But automatic1111 is still the go to for most people
But you know about comfy already don't you?
I saw some GitHub repos for that
I'm real rusty at this
now I'm gonna go watch something unproductive
well this was a bit better
Agreed
What upscaler?
Tho it could be a problem with some settings or something
I gotta get back in form so I can go back to creating things like this
but I also have other responsibilities, I gotta work on my novel, I have to get ready for chemo, I have to try and clean up my house even if I can't lift more than 10kgs without rupturing my abdominal muscles. Then I gotta play games 😛
Thats the 4th in 10 mins, thanks 
this is cool
Thanks lmao, I really like it

almost all of these are so fucking cool
very nice
I like
Thank you. With the new tools available I hope I might be able to make even better stuff, closer to what my mind imagines. Once I get the rust off my sleeves
JRPG swords with very fatal flaws
@brittle jay I'm going through my old renders, man I really need to get this rust off me and get back into making these. Rubber-hands aside.






