#🍥|anime
1 messages · Page 62 of 1
why is 3 out? 😮
when used in upscaler in txt2img no difference
ah okay!
Can notice difference if using upscaler separately
which folder do i put which files in?
Models -> ESRGAN
racin
you both racin
Well im heading back to the lab
don't make me prompt absolute thiccness again!
nah, just 10 more minutes and then it's tomorrow -> so i won't be tired anymore
flawless logic!
too old, out!
Y not ? 😏
Coz he will be blasted xD muhahah
well...i threw it in my llm prompting machine... absolute thicness seems to want to be a man in the first generation
Excuse me ?
my LLM prompting machine
just say "gimme something" and it'll generate a whole description for me

yaaay!
Somehow I wanna see
kitty
New here. Where can i create some images?
😂😭
hahahaha
guys are in here artists I can talk with?
Just trying to show a friend how AI girls look because he follows a page on IG and thinks they are real.
excuse me
this looks like my prompt...
the storytelling of the king looking over his kingdom
Oh wow I did those on monday that’s wild
less muscle XD
Oh my
LOL SHUSH

Spicy
its amazing how the same prompt can have such different results just by changing seed
testing character consistency using controlNet reference only.
this is from the same prompt as the two girls
It can get real easily 😄
so is this
seed determines the initial noise pattern from which the image is formed
NEAR clothing failure!
damn
Gyat
do you use restore faces?
I never use it
Im getting distracted by Aiska boys tho

restore faces is only used for photo images
it will ruin anime faces
I like him a lot
one wing for each 😄
😏
I'm using poems written for loved ones as my prompt
thre's no prompt in here!
but prompt shows up for me xD
absurdres, (((masterpiece))), 1boy, mature, (dark skin), exotic look, great composition, highly detailed, upper body, fluttering hair, a tall and muscular man stands shirtless, his broad shoulders and defined biceps glistening, strong jawline and piercing eyes, his hair is swept back, adding to his confident and charismatic demeanor, lora:styleBoysGlassyAndGlaze_v10:0.9, fata, solo, outdoor
Negative prompt: ((000000)), ((ffffff)), (EasyNegative), bad-picture-chill-75v, badhandv, ((out of frame)), lowers, (((text))), error, cropped, worst quality, low quality, normal quality, jpeg artifacts, side view, ((signature)), ((watermark)), ((username)), blurry, artist name, cgi, 3d, brown hair, femine, braided hair, ponnytail, white skin, long neck
Steps: 12, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 1882033726, Size: 500x700, Model hash: bfe7fddceb, Model: Anime_blueboys2D_v20, Denoising strength: 0.65, Clip skip: 2, Hires upscale: 1.8, Hires steps: 20, Hires upscaler: RealESRGAN_x2plus
not all!
Some cannot be uploaded here 🤣
it's always the hands
you'll get used to it
it works well
darkalfa is pretty good on the hands true
but creator is stubborn prick
yup, never satisfied, keeps whining all the time
wait isnt darkalfa your model?
😄
made with flat2DAnimerge model
cool ^^
be sure to use nerfguns negativehand embedding 🙂
this is almost perfect, except her right middle finger splits into two fingers LOL
easy edit!
love collors!
like in photoshop?
for AI terms -> this is VERY good actually
yup
i should learn how to photoshop
no hands = best hands!
Words to live by
DarkAlfa overlords would like a word with you
looks good!
Tried the watercolor in my promt and I kinda dig it
watercolor is always cool!
Some models don’t really know what to do with it tho
can you make more of this i love this style
Saved for science
oooh... good mention, lets see if my prototype could handle that
I LOVE BLACK AND WHITE LINEART
I wont lie i have no idea in which point i was during my.merge for this one, the final is dramatically different
oh i see
loooooooooooooooong phone
loooooong
damn remind me of japanese bubble gum commercial xD
LOL
https://www.youtube.com/watch?v=sZsJyCyGBSI commercial xD
fixed hands!
why DA F do you know this?!
Lmao that’s a classic 😭
oh, it is 😄
UGH, the hands of prototype 2 😦
colo
I just have words I use all the time that helps with imgs 😄
a good default negative prompt 🙂
hmmm... have i been going at this the wrong way?!
DUDE, NOT COOL 😄

Nah you’re doing everything right
(more referring to the merging i've been doing all day)
JUST GO SLEEEP!
well.. i think i kinda denounced this merge prototype already too...
Coz I;m going!
You will wake up with fresh ideas!
😛
yeah... this was the issue... it's very well styled... but too far from anime
Bonfire lit
smexy ass for g'night 😄
goodnightt
Night night
alright time to go to work bye byee
see what you did, now his cloths are wet, he has to take them off now
haha
i can try later if you want XD
Kojo Aisina, God of Dreams and Crossroads. From my setting
why is she sad?
does his domain of crossroads include any choice made by mortals? or does it mean literal crossroads like paths?
she only has sour patch watermelon left

in your mind story is this the same girl?
is she going out to buy more sour patch kids?
watermelon is the best kind!
quiet tonight
Both. He's representative of the choices people make in that regard, but also will appear at literal crossroads, either in dreams or IRL.
Selenia, Goddess of The Moon and Secret Knowledge
nice, what model did you use?
My own mix, MothMix. Version I used was just uploaded https://civitai.com/models/50041/mothmix

Selenia's daughters. Arcania, Goddess of Secret Knowledge and Hidden Wisdom, and Aestaria, Goddess of Healing and Mercy
anime dragons : D
Why did Arcania get crossover domains with her mom but Aestaria not overlap at all? Is it related to their other parent?
Also do Selenia and Arcania know every secret in existence? Can they keep secrets from each other?
is this a show or something you're talking about?
Selenia died. Arcania took over half of her domain because she didn't feel capable of fulfilling the full roles required of her. Their father survived, but was changed and sealed away, so Aestaria just kind of did her own thing out of respect to him
Story I'm writing
Mori’s original mythology!
It may be a show one day, if I get very lucky, lol
but until then just chugging along
trying to get everything written up
and concept art'd
Nope, its all AI. I can draw a little, but i'm not the best artist. The model is a mix of a few others, there' a list on the page if you're curious
ah gotcha
Right now I'm building up the setting and stuff. I have a handful of stories in that setting that are in the works, but want to focus on one at a time. A lot of this stuff is supplemental, but helps me fully understand who and what is all there in the setting so I don't lose track/know what I have access to
Is this becoming a game or something?
atm no. It's not anything but what I have written down on my server server and reference docs. It's likely going to be a lite novel first if anything
from there... guess it depends if people like it?
Typhondalos, Draconic Mad God of Chaos
Anyone follow this guide? https://civitai.com/models/34726?modelVersionId=54518
First i recommend reading the Part 1. It explains the extension and settings. Remember to leave some ⭐(~ ̄▽ ̄)~ I have planned to expand more on mult...
Do you see me the way I see myself?
Am I the monster I think myself to be?
Deep inside I know, I'm just an ordinary girl
But then why do I feel so wrong
I want to be strong
I want to be courageous
I want to be everything I see in you
The you I know you to be
Femboy got simps
Good afternoon
Femboi hairstyles
can you send prompts?
Should have PNG info :o
you can get prompts from images uploaded by going to PNG info tab in webui
alr
i basically train a lora just for femboys lol
out of curiosity
what makes it a femboy?
like
just looking at it how can you tell
Ayyy, lets gooo 
but for wd 1.5 beta2
By looking at it, well, I feel like a femenine boy, maybe easy to confuse with a girl :O
Beta 2? Why Beta 2? Is the Beta 3 worse?
i havent retrain it yet
Ohhh, fair then
Fair fair, I was planning to do a furry LoRA, but seems hard
nah just need quite abit of training steps and alot of quality filtering
Yeah, I mean
I think the best would be to make it like a huge dim LoHA
And make a bunch of folders, of like, cat furry, dog furry, wolf furry, and separate by like species
standard lora works fine
I think LoHA would be better for less concept bleed though
also i use kohya addnet so best i could do is dylora
You can't make LoHA?
well i can
hmmm
jeez look like i gonna have to rewrite the prompt quite abit
i gonna reset most of the prompt weights to 1
wd dont like excessive prompt weights so ye
Fair fair :o
this is wd 1.5 beta3
OH it is?
just with 4 style loras
Ohhh, fair then :o
It looks a bit like the radiance finetune, is that the version you're using :o
base model with these
anyhow its time for me to be a goth
hmm not as i exactly wanted
That looks really cool tho :o
do you want to try this yourself?
Still looks great,
Sure :o
I have been trying to make some art and it isn't coming out like I want it,
what is not to your liking?
I want him to be more muscular but it somehow ends up being shirtless,
I see,
Also which Lora is the best to use for male characters?
No a friend of mine wants to make a art for a male character so he was asking,
check dms
so... i'm trying to figure what the problem seems to be, you want him to be more muscular/toned, and keep clothes?
Yeah,
so... where does it go wrong?
He looks thinner compared to what I imagined,
did you try toned?
I'll try again after reconnecting to the site,
are you using a lora for the pixelation?
no just based66
what is that?
a model i found off civitai
What lora?
my own.
You published it or plan to? What is it trained on?
no it's bit a of mixed reaction anyways to post artist loras on civitai anyways
Who is it trained on?
the artist in question was happoubi jin
Thanks
I guess it's christmas now
hi guys, can i run this model on my 8gb vram gpu? https://huggingface.co/TheBloke/wizard-mega-13B-GPTQ
if you offload some layers to ram
what should i set it to?
this right?
i think it's 40 layers iirc -> 30 layers
what is "iirc"
retrained my nixeu lora now it looks hella good
what is "iirc"
Stands for "if i recall correctly"
holy smug
model?
MEH. lets do this. hand-fixing attempt number 4
why is the oobabooga webui crashing on lpoading a model?
is it a gptq 4bit?
yep
the thing where the process just shuts down?
Had that too.. I think latest version fixed it, but never solved it myself
bruh, i am on the newest version, even reinstalled it
idk
you sure you have the right model version? models by thebloke have both act-orders and no-orders
perhaps you got a model loaded which has both activated?
idk what that even means man
using wd right?
SDXL
oh
My laughing/smiling vampire prompt tests. A lot of people got angry with me after I generated these in the SDXL bot channels 😂
tf is sdxl
The new model by SAI.
does it do anime
yeah
Its avail in #1100170312106127410 through 10 wdym lol
The only place i go in this server is here
So can i download and train stuff on it
You can get stuff like this with SDXL but there is a learning curve to the prompts

not yet, they have not released the model to the public
Marlboro ahh background
Also isnt sdxl gonna need smth crazy like 14gb of vram
unfortunately yes, I think this is for the RLHF version, but they've said they will release smaller versions that will run locally with less vram
yeah, the community will probably optimize it
probs a more suited name like sd 2.0 or smth imo
man comfyui is not that comfy, it hurts my heady 
queen of the trashburgers!
No burger is trash
i could go for a burger, my brain is mush
that burger on the floor tho
Except nasty patty
Borger
🍟
I just tried doing this myself (albeit i used different file names) and it didnt register with the webui
you need dynamic prompts
and it enabled
and then have those wildcards saved in the wildcard folder of the extension before invoking it
goddess of pizza
My kind of goddess
we eating rn
@wispy canopy u have any idea how one does this?
i committed a war crime
please tell me that is a diffusion 😄
that's not a warcrime, that's crimes against humanity
ye it is you could check pnginfo
still a crime against humanity xD
also i have more too
ehh...
i think you need to write down the full loras in your wildcard
so your wildcard file has lora:thisone:1
or whatever you want
His screenshot of his wildcards section only showed the name of the lora
we need to invest in big things
It's convenient when everything is set up and you can make stuff in 1 click
yeah im gonna enjoy this once i fully understand it
she has questions though
we busy no questions
i assume you use it?
if so got any tips? am getting stuff like this
and the pre hires looks nicer in a way
too little steps?
oh perhaps, was at 14
It depends on what you're going for. I like to use add detail Lora and upscale with esrgan Anime4x 6b. And also use the custom nodes for automatic face fixing+detail
seems like theres a few QoL things im missing for comfy
(i just got it running this morning)
Steps depend on what you use. For example, huen is equivalent to half the steps as something like Euler a but is twice as slow. But it converges faster and less variance
Where can i find this custom node thing you mentioned
If you have another installation of a UI, you can change one of the files so it uses the models in that one (so you don't have to duplicate models)
Ya i got the yaml set up
To load my sd models
I was able to pull a hires workflow trom youtube
https://github.com/ltdrdata/ComfyUI-Impact-Pack there are other custom nodes but I haven't used them yet because I didn't need them
Yeah it's quite convenient. I don't know if it has stuff like reference_only yet though
I havent messed with CN all thst much yet tbh so thats ok
Auto1111 has the advanced features, comfy is later to come but really stable
Appreciate this
Gn chat
@wispy canopy more steps yea
and have the upscale denoise high enough
its on .5 rn
for dpm2++ types that's a tad too low
they need 0.54-58 at least to be sure to resolve properly
pog ok
ddim should be ok at 0.5
this is really good quality for not using TIs or anything
i like this
must be allot of under the hood stuff going
Best part is i dont need to use any special flags just launch it and go lol
i knew she was a reptile all along!
aiska... did you steal my prompts again?!

funny that if you ran a prompt long enough, you can recognize them on another model 😄
hahah
it has goddess, flowing cloth, shrine, water, candles, night sky
when u are addicted to prompting and SD
this one is good! need to reconstruct throne tho
just higher frame 😄
nah, she just has major strength ass-cheeks. she's just suspending herself!
dual wielding drinks!
in my opinion the best intros of the season(anime) are oshi no ko and first place the legendary hero is dead. what are your guys opinions?
not watching it, sorry 😄
watching what?
I'm not into this seasons anime
bro what, last season was even worse tho!
preferences 😄
name one good anime that came out last season
there's an unused A100 😄
where?
company server 😄
looks like you didnt understand what I've said. Every one of us has different preferences and it's not up to anyone to decide if sth is good or bad. If you didn't like it, sure but there are ppl (people) who liked it. And pls no bro ❤️
why no "bro"?
coz I;m not male
dont wanna be disrespectful, but thats just how i talk
no worries, I;m not mad 😄

and as for last season anime I like "Tondemo Skill de Isekai Hourou Meshi" since I've been reading manga soo before it
oh yeah, that's why i stopped generating on the A100
the images are not saved on my pc
"Koori Zokusei Danshi to Cool na Douryou Joshi" is nice too! Since i;m buying manga 😄 (manga is way better!)
and yeah, I saw myb like 3 anime last season
so not much to compare
Cant get over stopped Nier tho 
you already know how it finishes!
and be glad nier is decent
i still can't get over devil may cry
and... for some reason, i can't get into the rwby anime
saw them before I've finished game!
i like the og a lot more
so it motivated me to finish it!
I'm also very sad about delayed NieR 😭
well if u like series, sure!
Honestly, I'm biased. I just love the NieR universe and want more of it.
Nice!
fangirl simping 😄

trying to see how the clip nodes work and if i can use them in sequence lol
idk if its working
I have all of the Automata achievements and did all 26 endings, followed by another fresh playthrough after getting them. I haven't gotten all of the Replicant achievements, but I did all endings including the secret one. Maybe I should go back and finish those achievements.
recommended if you want to cry more! 😄
I had this for my wallpaper for a while until Windows decided to just stop loading it.
you need 86
You are banned from generating until you start NieR
🤣


no p*rn
yeaaah sure 

we need banana-hammer for Lizzie

am I tasty? 😄
yes


Yeah, comfy is neat. But it's a lot more work.
I should learn how to use comfy and then sell a SD course to people
Well I could do it even without comfy
But since comfy looks so complicated I am sure it would help marketing
lol
Have you guys seen anyone that was interested in learning to use SD?
And is willing to pay money I mean
I'd probably laugh at someone willing to pay money to learn it considering how many free tutorials are available on Youtube.
There are ppl who does it but I dont think audience is big
Well there probably are such people out there
would be better to make vid on YT, higher audience
Maybe I'll also add in LLMs
Harder to make money in short term
And I am not great at presenting myself and stuff
In videos at least
Don't need many people, just few who will pay
well, it;s hard to get customers first
Yes
Do you upload your artwork anywhere?
I should go back to scamming children then
jkjk I am pure hearted, never wronged anyone in my life
But images kinda mid/bad
And I rarely upload
Putting together a decent course would be quite a bit of work if you're only expecting a few customers. Probably better to spend that energy on some other money making venture.
Well
Hard to argue with that
you neet get higher audience on Insta then, some ppl will DM you overtime and if you see that a lot of ppl asks you "how" then you can think of creating one
without proper "free" marketing it's gonna be hard
I just now remembered some woman who messaged me like a couple months ago and wanted me to make an image of her dog with ai and I was like "sure" but then forgot about her and never responded
well that's bad then. You can make an for eg. etsy listing for commissions
Then you can make money, slowly
"you guys are making money?"
WTF SD... that is not 2B's blindfold XD
She doesn't look very happy about it
does anyone has an Albedo LORA?
there's a lot on Civitai
this one is more like JoJo 😄
why i don't find them 
My poor sleep deprived head couldnt take it no more for today. Tomorrow lol
probably searching in wrong category? 😄
wait what... my model knows what rainbow colored hair is!?
why i found 0 of them lol
thank you so much!
@ocean reef if you find more of them please link me!

Those are the best I've saw 😄
the resemblance
no rainbow hair for you?
pats his model
She's very sad that they didn't give her a plate for her burger.
nope, no rainbow hair 😄
my model is just awesome that it understands 😄
this hand tho 
new way of holding things!
lol... if i see what the prompt did in the OG redacted stuff vs this...
no hand at all! 😄
can't have broken hands if you have no hands
no hands, no cookies!
shiny waist armor
looks cool
just generic fantasy prompts i did when i discovered the redacted
hobby just went downhill though
😏
Borger
ooh entered the elysium v2 anime era of my saved images
and i'm finally upon the era where i started prompting lyrics 😄
those prompts did not age well against a 'generic capable' model
darkalfa prototype2 can't do lyricsprompting
AT ALL
we going back to church bois and girls
dont do it!
you'd listen to her sermons too
although i got this feeling it'snot a regular church or smth
BLING
Yes :O
Praise the borgir
Trying out WD 1.5 Beta 3 Illusion, its kinda cool :o
Thicc
Yis 
Any way to inpainting a large image? I happen to be making an image where people come out but smaller and when upscaling their eyes become strange.
Enable this. It will inpaint only a section around your mask at the full resolution that your sliders are set to. The better option (and what I would recommend) would be to just use the Krita extension for your inpainting, but that's more work to set it up.

They asked me to. I said I wasn't sure if I want to.
I already have it, my image is like this, the eyes of the characters are small, so when I upscale it looks weird.
Makes sense, I sent a application, but I think someone snitched and showed them my horni posting tho

Like this
You have to inpaint in their face
dis model comprehension too good 😮
Just smack a blob on their face
You might want to pass the whole image through img2img with low denoising to get rid of the high resolution look on the characters.
And also use a res between like 512 and 768
True too
That's what I've done, I've painted the eyes to fix them, at a resolution of 512x768 everything looks fine, but when upscaling you can see that imperfection that is not visible with a small image.
Yeah, with already being a mod on the other server I just don't think I want to take on actual responsibility here too.
Fair fair :o
Oh, what upscaling method are you using?
Though also talking about mods, I wonder where Guizmoose went, I haven't seen him in a while 
i have tried animesharp and resgan 4x+ anime6b
i heard i was taking a break or smth? read soemthing about in the pow discussion
Imagine, heathen
Ohhh, fair then :o

Are you using them in extras tab, using SD Upscale script, or using tiled diffusion?

I mean, I do like helping people, but Nubby is the gigaest brain out of us here xD
the extras tab
what is tiled diffusion?
I like helping people, but I can do that without a role 
Fair, role is nice tho
BUT
Not as cool as the stable star role
That one has a star
That one is the coolest
its an extension that gives probably the best upscaling results. https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111.git
The main function is splitting generation into tiles within the latent space which can help with coherency and vram utilization, but in the img2img tab it can be used along with an upscaler to allow for AI denoising to clean up the denoising artifacts.
Oh nub. Comfy had tiled kick in when it oom, is there a way i can have a node to always have it on or smth?
Okay thanks i will try it
I'm not sure. I imagine if there isn't one built in, there is probably one made by the community somewhere. I haven't used Comfy in over a month though so I'm not sure what all it has available now.
I still have that github link that was shared to me here earlier
Civit is another place to look?
These are the settings I use for img2img upscaling. With multiple passes, I've taken images up to like 16x8k resolution with these settings. I have 12GB vram, if you have less then you will probably need to reduce the Latent Tile Batch Size, I assume 8GB would probably work at 4. For the prompt, you want to simplify it for the upscaling. I usually remove any mention of my character and sometimes remove the background description too. It can work really well with just quality tags in the prompt.
Okey thanks, in the prompt what should I put? in my case it is a girl playing the ukulele and a guy smiling inside a truck, I should not describe the people or yes?
thicc ❤️
Yes 

anyone know if there's a way to directly paste a prompt(png info) from discrod to a1111?
I still dont understand how u get started with this. Do u just take an image from t2i or some place and place in i2i to upscale?
Also everytime i do try to use it it takes an ungodly amount of time to finish

yeah, tomorrow i'm posting this model. the hands might be worse than darkalfa, but the general comprehension of this model is out of this world 😄
I wouldn't describe the people or the truck at all. Just whatever quality tags you normally use.

In the laptop I am using I only have a 3050 4gb, do you know if it can be used with 4gb of vram?
you open too many tabs 😄
I cant ever have that many tabs open for long
Yep generate normally, then use this to upscale and add detail. Compared to just straight upscaling in the extras tab, it will always take a lot longer since it's doing that plus running img2img. If it is taking an extremely long time, you may be at the verge of hitting OoM, causing it to swap to system ram. You can reduce VRAM usage by lowering tile size, lowering latent batch size, or lowering tiled VAE tile sizes (if it is specifically getting stuck or taking a long time when encoding or decoding the image).
war fairy prompt still goes hard 😮
Should be able to. Might need to lower the Latent Tile Width and Height both to like 96 and the Latent tile batch size to 1. Also lower the Encoder Tile Size to like 768 and the decoder tile size to 96.
What about those wider images people use it for? Like the regional prompting stuff
fun times when you restart the browser 😄
Okey thanks i will try
I've never gotten good results out of the regional prompting that Tiled Diffusion has built in. I find Latent Couple to be the most reliable option for that, though the actual Regional Prompter extension got some updates at some point that make it look like it might be better now. This one is the regional prompter extension: https://github.com/hako-mikan/sd-webui-regional-prompter.git
Yeah i have it already. I had similar experiences as well
Idr if i updated it or not
You want Enable Tiled VAE turned on and Move VAE to GPU turned off. If it still gives Out of memory, try slowly reducing the Latent tile width and height until it works.
TBH most of my wide images are just generated normally with Hires fix. I usually do 512 H and anywhere between 1024-1536 W then 2x upscale. Open pose can help to avoid some of the weirdness and extra characters that wide images are prone to.
With 12GB you can bump the latent tile width and height to 112 and latent batch size to 8 which will speed up the process by a lot.
Lately my openpose isnt even working

I updated and now nothing happens
New models too
God blessed her hands
i'm stillbummed though, that i haven't been able to fix them to darkalfa's hands
they are way better than these
Whose
How often does it get it right
2/4 for the prototype, but the og model is better, but has lesser contextual knowledge
and i've been able to get the hands a bit improved, but at the loss of contextual knowledge in the model, which i kinda don't want to give up
i mean, it understands rainbow colored hair
An unfortunate tradeoff
is giving me a different image instead of increasing the resolution.
ngl, this prompt is really cool
What is your denoising set to? Usually around 0.4 works well with DPM++ 2M Karras, but you might want it a little lower.
how i can make this? awesomeeee
Thank you! I post my images with metadata intact so you can drag it into your PNG info tab if you're using the WebUI to see my entire prompt and all of the parameters.
thanks, im new here, new in this world of IA, im using discord right now, WebUi give more options?
a lot more options
and in general, way higher quality pictures too, if your gpu supports it
Not working
it is fully free
Not* xD
WOW
xD
anyone know how to trigger the faq bot?
@vital scroll
Welcome! Start by heading over to #1072220168534642768 to get yourself situated and help find the channels you are looking for! Please make sure you review our #✍🏼|rules-and-tos and feel free to assign yourself some #👥|roles as well! Answer any questions your may have at our #1072229020520947753. There are many ways of accessing Stable Diffusion, take a look at #1080946152318443610 to start your journey!
Any other way to correct this?
^^ np, nice to have you here
Do you need it to be upscaled, or were you just doing that to try to add more detail?
upscaled but fixing the eyes
in 512x768 i see it good
But when scaled in extras it comes out like this, they are not as detailed and there are typical lines.
I find it really weird theres no zoom feature for inpainting , would help a ton in these situations
You could upscale in the extras tab then inpaint using the Krita extension. Technically you don't need the Krita extension for that, but it will make masking the eyes a lot easier. If you use the Krita extension you'll want to mask the areas you want inpainted then use the rectangle selection tool to select the area around the mask so that when it generates, it will generate that area at a higher resolution.
I use the canvas zoom
but is a high resolution image
after the upscale
so i can't inpaint






