🍥｜anime

[native halo]

damn she rockin the new Oscar Pistorius boots

[supple raptor]

make sure both your Adetailer and Controlnet extensions are up to date, then you select the control_v11f1p_sd15_depth model. The hand thing will be a preprocessor or "module" under that model.

~~https://github.com/picobyte/stable-diffusion-webui-wd14-tagger.git~~ wrong one

[worthy delta]

thanks#

also how do you use adetailer

[supple raptor]

wait, that's the wrong one. one sec

[worthy delta]

i alr got it dont worry

[supple raptor]

https://github.com/DominikDoom/a1111-sd-webui-tagcomplete.git

that's the one you want.

[supple raptor]

Just install and enable it. Default values are good in most cases. The face and person models are the only ones worth using. Even with the hand depth preprocessor it's janky for hands, and will hurt more often than it helps. The hand pre-processor is best used for manual inpainting.

[nova remnant]

I heard the books at the top of the shelves are the best, want help to reach them

[broken osprey]

It doesn’t show up and both are up to date :(

Im not sure if its cause of the symlink between the folders but I do have a ton of control net models and all of them show up including the new one :(. I’m not sure if I should put the new model in some special folder for adetailer?

[native halo]

let me climb on your shoulders to see if i can reach them 🌚

[sonic narwhal]

Hello, Can anyone help me fix this?, when i use roop, or Reactor, this happen to the face, i tried to change Restore Face Visibility, CodeFormer Weight (Fidelity), same thing happen, and when i use ip adapter to change the face, it not recoganize the face,

[keen hollow]

I just ran a single image that took 25 minutes to generate... and it's terrible. DANGIT lol

[nova remnant]

Long and grabby

[nova remnant]

Try not using face restore

It's not gud

[sonic narwhal]

i don't use it,

[nova remnant]

[supple raptor]

There isn't a new model. It's just a new pre-processor that shows up specifically under the control_v11f1p_sd15_depth model. If it isn't showing up, then you're likely not using the Controlnet v1.1 models.

[broken osprey]

.-.

It shows up under the model tab on controlnet

Under the depth preprocessors, any of them for that matter

I don’t know where adetailer picks up the models for it’s controlnet since it should be showing all of the other models like openpose f16 if it were reading my files

[supple raptor]

I assure you, it's not a model. It's a preprocessor, which the controlnet extension refers to as "module"

[native halo]

yea its not a model just a preprocessor like dw-openpose

it uses the depth controlnet model you always use

[broken osprey]

But it is selected under the models tab and that hand thing doesn’t show up

Even though I updated control net like today

[supple raptor]

no. it isn't as my screenshots show.

[broken osprey]

Im@not saying you are wrong I can be confused but I’m seeing it under models, with a new pre processor yes

[supple raptor]

That's the same depth model we've been using for months. The preprocessor is the only new thing.

[broken osprey]

No no I mean there

I got. Anew model

This monstrosity

Which showed up in a huggin face repo I was redirected to

This is what confuses me :(

https://huggingface.co/hr16/ControlNet-HandRefiner-pruned

[broken osprey]

Got it from@here

:=

Didn’t ping though so I’m safe

[native halo]

that model requires its own installation as shown here https://github.com/wenquanlu/HandRefiner/blob/main/docs/installation.md

[broken osprey]

So weird I’ll see if it shows up tomorrow somehow, I installed from url control net but it shows up as up to date

I’m not sure if I should update it somewhere else

[native halo]

it literally says here it hasnt been added to controlnet yet https://github.com/Mikubill/sd-webui-controlnet/issues/2429

[supple raptor]

You already have the useful part. The preprocessor is the unique part and you literally have it selected in the screenshot you sent.

[broken osprey]

In which one?

[supple raptor]

This one. depth_hand_refiner is the new thing.

[broken osprey]

That was a remix of your own screenshot

I’m on my phone rn lol

Can’t check if I have the tab

I’ll check on it later today sorry to bother you ^^

[supple raptor]

ok, then do you have the control_v11f1p_sd15_depth model?

no worries

[broken osprey]

I do have those all of the models including the t2i ones

The issue might be that I installed torm@url or smth idk

I had a ton of problems installing control net cus their new requirements installation messes up windows permissions

[supple raptor]

in case anyone happens to be missing them, here are the pruned fp16 controlnet v1.1 models and loras for sd1.5 https://huggingface.co/comfyanonymous/ControlNet-v1-1_fp16_safetensors/tree/main

[broken osprey]

Wait Lora’s?

What are those for.

[native halo]

for comfy users

[supple raptor]

yeah, basically comfy just allows using a lora extraction of the model which saves vram and loads faster. it isn't currently possible to load them with A1111 afaik.

[broken osprey]

Damn gor scared for a second, in comfy afaik DW pre processor and the rest installed their own shit and it works out of the box

[stable crane]

really cute, very sharp

[steady grail]

don't press it. it's a scam

[stable crane]

thx mod

[steady grail]

wow, so fast

[steady grail]

charming

[nova remnant]

She wants it

🍨

[vital raptor]

nappy time

[nova remnant]

[steady grail]

just found out something about comfy

bypassing a node, contrary to what it says, totally truncate the node and everything is connected to it from the workflow

instead, setting his mode on never, just deactivate it, but everything is still connected to the flow

[steady grail]

or maybe not... the flow is interrupted...

[steady grail]

i just noticed that ai is surprising good in generating close up portraits

much more than a complete landscape

i suppose details are much more focused and the models trained with many of these images

[steady grail]

wow, cute

that would do a great desktop animation

[topaz juniper]

to smull

[steady pumice]

cute

[nova remnant]

elfie

[topaz juniper]

[pure oyster]

I would say that staff guy is the most coherent ai image I've seen so far.

[steady pumice]

ty he's a baddie from a story I'm working on

oh and if your wondering about the lora, all my males use the bleach style lora as for the check point I use it's based67

[pure oyster]

My stuff comes out good, but only because I do a lot of manual work on it.

[steady pumice]

thats great

I do some inpainting mainly on the hands most of the time

[wooden coral]

Kinda wonky dragon

[steady pumice]

Yeah I could not get it to look perfect

that was awhile ago

[pure oyster]

damn, your GPU just raw doggin 2500x2500 images?

[steady pumice]

Lol no

I upscale them

when I'm done

[steady pumice]

it wouldn't gen right at that high of res anyway

you would need to use XL for that

[wooden coral]

Kohya high res fix

[pure oyster]

what tool? for me I blow it up to 4k then inpaint it like 10-20 times

If I could cut my workflow in half I'll take it

[steady pumice]

a1111 upscaler

I don't use comfyui

[pure oyster]

ultimate sd upscaler?

[steady pumice]

I think?
I had friend helped me set it all up and he showed me how to install loras and models that's about as far as my knowledge goes with it

[broken osprey]

Is that depth hand refiner the new depth thingy? Also how should I use it to fix hands since it always gives me an error that it has no input image :c

[wooden coral]

I use ADetailer for it

[broken osprey]

It doesn't show up in my adetailer, would you know which folder adetailer picks the controlnet models from?

[wooden coral]

Gotta play with the denoise strength tho

Uhm

try updating I had the issue as well

[broken osprey]

latest :c

[pure oyster]

inpainting hands never worked for me, all it does is make the eldritch abomination high res.

[wooden coral]

Did you restart it as well?

[broken osprey]

Should I put the model eitherin adetailer or the other?

[wooden coral]

I didnt have to download anything, when selected it downloaded itself

[broken osprey]

I got the model to show up, but im not sure what those things at the bottom are

[wooden coral]

If you look at my screenshot I think you got it the wrong way

You have to select the depth model and then hand refiner

[broken osprey]

Exactly those are not showing up, I think I gotta connect the models somewhere inside the models folder on my install, everything is on symlink so I think I am missing one of them

[wooden coral]

weird

[broken osprey]

I got the preprocessor for controlnet but it threw an error so big idek what its doing anymore

[vital raptor]

there should be a controlnet folder in your models folder

wait, I haven't read the whole conversation, ignore me

[wooden coral]

Have you tried disabling controlnet and deselct the model there? Maybe it interferes

[broken osprey]

This si my install and that link is a symlink to my hdd where all of the models are, idk why but controlnet finds all the models easy, but adetailer only finds the one that is directly in there

[wooden coral]

The other possibility (seem more correct) is to download the three models and place them manually in webui folders
https://huggingface.co/hr16/ControlNet-HandRefiner-pruned/tree/main and place them in the 3 following folders:

1/ stable-diffusion-webui\extensions\sd-webui-controlnet\annotator\downloads\hand_refiner\hr16\ControlNet-HandRefiner-pruned\graphormer_hand_state_dict.bin
2/ stable-diffusion-webui\extensions\sd-webui-controlnet\annotator\downloads\hand_refiner\hr16\ControlNet-HandRefiner-pruned\hrnetv2_w64_imagenet_pretrained.pth
3/ stable-diffusion-webui\extensions\sd-webui-controlnet\models\control_sd15_inpaint_depth_hand_fp16.safetensors

Restart the console and the webui

Now I can use the controlnet preview and see the depthmap:
In controlnet model select control_sd15_inpaint_depth_hand_fp16 and preprocessor depth_hand_refiner. Check upload independent image and allow preview. Add your image in controlnet image. Click run preprocessor. The depthmap will show.

Taken from here https://github.com/Mikubill/sd-webui-controlnet/discussions/2400

[vital raptor]

you might also have problems with symlinks, defining custom paths in the webui-user.bat might be better

[wooden coral]

It should show it without downloading it, so there shouldnt be an issue with symlinks imo

[vital raptor]

for example:

[broken osprey]

hnmmmmm I might move the models it needs liket2i depth and the normal 1.1

Also I thinkmy controlnet downloaded all of those excet that other thing, the last inpaint depth hand that one I already have

[vital raptor]

I just installed it to see what's up, had to update adetailer for it to see the models, but then it downloaded some when i tried to run the img2img

also i'm dumb, i forgot the controlnet part from the webui-user.bat line i posted above 😄 --controlnet-dir "D:\AI\models\controlnet"

[steady pumice]

💀

[vital raptor]

people really need to stop clicking those

[steady grail]

as they are there, they could at least try to gen something for us

lol

maybe this would convince me clicking their links

[vital raptor]

well, no luck for me getting the hand thing working, it's not working with sdxl yet it seems. Also seems to be faster to just keep generating until I get proper hands anyway 🤷‍♂️

[steady grail]

well, at least, in hooking and deforming good hands, it's very good lol

the staff should add a room for free gifts. maybe such people will stop spamming at random XD

[vital raptor]

it's not people

[steady grail]

yeah, maybe it's automatic

[vital raptor]

they click the link, type in their password because it asks for it, then a bot logs into their account and spams every server they're on

[zinc flax]

@dull jackal

[broken osprey]

Hipoly lora looks great, also I thought this looked like animation style thought I would put it here ^^

[broken osprey]

A fascination for green and blue hair this thing has lmao, at least the eyes change

[broken osprey]

YOOOOOO FREE NITRO?! clicks

@dull jackal

[zinc flax]

They found out the hard way lol

[broken osprey]

nah imagine clicking one of those links lmao

[zinc flax]

They are certainly clueless

[nova remnant]

@zinc flax

Where are the sleepiest cuties at

[zinc flax]

Haven't genned anything today lol

Maybe later

[nova remnant]

It's so hot here today I don't want to move

[broken osprey]

Can anyone explain the difference between t2i-keypose and t2i-openpose? I see tht in some cases the model adheres more to the image using keypose model than openpose, so like which one is better or what are their use cases?

[silk narwhal]

believe keypose is more for image/video to video

[broken osprey]

Ah that explains it then haha thank you ^^

[gloomy scroll]

[nova remnant]

2 reasons to love elves in one picture

[steady pumice]

A little bit of Pyromania

[nova remnant]

Do you put extra work on those weapons

[steady pumice]

indeed

[nova remnant]

Why can't SD do them from the get go

[steady pumice]

Man I wish

[native halo]

it can with controlnet

[nova remnant]

[steady pumice]

yeah but I don't really like to use control net

I do sometimes tho if there is a pose I want to do but most of the time I like to leave it up to rng

[brazen rapids]

what you mean

[native halo]

what model is that or its an edited image?

[brazen rapids]

||https://civitai.com/models/31472/based64||

[steady grail]

why the spoiler on the model?

[native halo]

the

[steady pumice]

based mixes are great

I use 67

[native halo]

based and sd pilled

[steady pumice]

holy shit c control?

was not expecting to see her

[native halo]

no just a lora,someone just released her char on civitai

[steady pumice]

interesting I wonder what she will look like with my style lora

might try later

[stable crane]

same

[brittle rampart]

and this hand correction thing really works?

[vital raptor]

I couldn't get it working

Like most things, it's probably highly dependent on what checkpoint you're using, but also like most things, it's not really needed because if the model can do it then normal inpainting or just generating more seeds will get good results too

If it works, it might save time for people with low end hardware, so it has that benefit

[brittle rampart]

that's why it became interesting to me. in my case THIS THING will be quite promising if it really works

[vital raptor]

it doesnt work with sdxl yet so I didn't mess with it too much, just tried with one of my 1.5 models and it didn't do anything better than manual inpainting and running a few more gens, but it took all my vram and was really slow

but it says it can run on low vram cards if you use the lowvram or medvram options

basically for me, taking an image to the inpaint tab, drawing a quick mask and generating 5-10 images to try to get better hands was faster than 1-2 gens using adetailer + the hand thing

but i also just kind of generate random stuff, i don't usually spend much time to get anything super specific

[brittle rampart]

in that case, for now I will continue the same process of work

[bitter latch]

anime music cover 😄

[bitter latch]

very nice

[steady pumice]

ty friend

[bitter latch]

ok this nudreamao3 merge I just made.. done this well.. wow

[steady pumice]

I didn't even know you could managed to get anything to gen under 4gs of vram

[bitter latch]

nuelement >envydream>AO3_oranges mixes low 0.05 weight merge

<----3050 4g

thats how

[steady pumice]

ah I see

makes more sense

[bitter latch]

wish there was a way to upgrade ram for video in laptop.. dammit

this card actually does pretty good. I can even play starfield

[native halo]

is that even a good thing? playing starfield?

[bitter latch]

hey leave my starfield alone 😛

[native halo]

🤠

[bitter latch]

tho lately been playing alot of a game called GunFire reborn

tho need me a controler for ---> .hack//G.U. Last Recode

now thats anime game 😄

[native halo]

theres also an anime for the hack games,havent seen it yet so idk if its good or not 😔

[bitter latch]

hmmms not wanting those laser lights dangit

yeah i'm in random mode. can't sleep so trying to tire my brain out

ok thats trippy

tho reminds me of a oil painting

ah ha

old style anime

hmm anime transformation?

oops

anime manga

[bitter latch]

falls over

[bitter latch]

uWu
Steps: 25, Sampler: Euler, CFG scale: 6, Seed: 1905744891, Size: 512x512, Model: nudreamAOM3, VAE hash: f921fb3f29, VAE: mangledMergeVAE_v10.pt, Clip skip: 2, Version: v1.7.0

Miku uWu

[brittle rampart]

is this called a cube smile?

[bitter latch]

shrugs

[brittle rampart]

my one fav theme steampunk...👍🏽

[steady pumice]

yeah I like it a good amount planning to do a story they are the mcs but I'm mid way through another one atm and I have one with vampires in mind next

[brittle rampart]

in that case, good luck with your story.👍

[steady pumice]

ty friend

[nova remnant]

[wooden coral]

Double crit with one button prompt

[native halo]

something wrong with mine i think

[steady pumice]

Oh no

[brittle rampart]

that guy definitely reminds me of someone....

[steady pumice]

I think I know

[brittle rampart]

of course, if you cut off his ears!

[steady pumice]

https://tenor.com/view/shunsui-anime-gif-9421128

[nova remnant]

[brittle rampart]

conceptually!👍🏽

[nova remnant]

@brittle rampart

Steampunk spaceship

[nova remnant]

which program are you using to compose your story? scribus? Indesign? something else?

@native halo 5000 dollars star citizen ship

[native halo]

cheapest microtransaction in star citizen

[steady pumice]

I haven't really messed with anything like that I am just wrting out the text for it all first, I was honestly planning to upload them on webnovel since they have a built in comment section where you can post images by replying to sections of the story

But now that I know about scribus i might give it a look sounds cool

[brittle rampart]

looks like my space train

[nova remnant]

Makes me wanna play space engineers

[native halo]

heard u like horns?

[steady pumice]

trippy

[nova remnant]

Horns : ✅ pointy ears : ✅ flat: : ✅

[steady pumice]

shes from c control

[native halo]

https://tenor.com/view/msyu-anime-cmoney-cute-female-gif-21178494

[nova remnant]

I have no idea what that is kek

[steady pumice]

decent anime

[nova remnant]

I'm the one who talks here and doesn't watch anime

[steady pumice]

[native halo]

https://tenor.com/view/kenshiro-hokuto-no-ken-fist-of-the-north-star-omae-wa-mou-shindeiru-nani-gif-14473693

[nova remnant]

[wooden coral]

TIL that you can call a whole wildcard folder with a *

[nova remnant]

Boto

Make it night instead

There's a new avatar game with bioluminescent stuff all over the forest

And at night it all becomes pretty

Reminds me of that

[wooden coral]

Yeah ive been playing that a bit

Looks sick on an OLED

[bitter latch]

i'm gonna go pass out now, but first here...

[nova remnant]

Need more looking up elves

@wooden coral tried my best

[bitter latch]

ok 2nd.. (THUD)

[nova remnant]

[bitter latch]

joins in

closes the programs so he can sleep

[nova remnant]

My SD folders

[brittle rampart]

first - anime interpretation of the singer from the 5th element move?

[steady pumice]

you bet

that and it's a scene I have in mind but the mc views it on tv he's not there

They are crazy (in love)

[steady pumice]

cute cirno

[wooden coral]

Kinda late but made me chuckle

[steady grail]

what a funny face lol

[gloomy scroll]

More Cirno

[steady grail]

charming

kawaii

[wooden coral]

Found a Yareli Lora, my time has come

[gloomy scroll]

Yareli main spotted

[wooden coral]

100% me

Got around 100 mil XP on her and is my most played frame

[gloomy scroll]

I just play melee only Mesa

[wooden coral]

Come to the dark side, we have K-Drives in every mission

[wooden coral]

Mesa mains

Im sure there is a mesa lora too, you should try

[gloomy scroll]

Also I think my basic Mesa's exp bugged and didn't increase after it got to that 105m at least I remember checking it at one point

Peacemaker ate everything

[steady pumice]

nice warframe

[gloomy scroll]

[bitter latch]

😄

[bitter latch]

ok not sure where the tights and the snow bunny came from

hehe ---> Bunny Small eared elf Miku in white winter dress with dark blue over coat that is open, black tights, standing in snow, large snow flakes, smiling, long braided hair Steps: 30, Sampler: Euler, CFG scale: 5.5, Seed: 1122031264, Size: 512x768, Model: nudreamAOM3, VAE hash: f921fb3f29, VAE: mangledMergeVAE_v10.pt, Clip skip: 2,

uhh oh

gets an idea

oh no someone made Elsa Bunny angy

😄

ok this is the best merge yet. hs

not using any of my saves styles or even a negative

"run.. I said RUN!"

hmm put in elsa and it overrides . interesting

haha love this one

hmm almost

stop driving that prompt into the ground

oops

ok maybe not yet.. 😄

[quiet bramble]

Have any great anime illustration models dropped in the past five months?

[supple raptor]

I use Soushiki - https://civitai.com/models/211628?modelVersionId=238399
and Kakigori_v3 (my own merge) - https://civitai.com/models/100505?modelVersionId=191903

[quiet bramble]

Thank you. They both look great.

[bitter latch]

almost

anyone know who this is without cheating?

ok she doesn't have rainbow hair. oops

scooby doo anime ?

[nova remnant]

My finest example of soushiki, amazing model

@supple raptor

[supple raptor]

https://cdn.discordapp.com/emojis/990032236919279656.webp?size=64&name=ChenGyate&quality=lossless

[native halo]

this one cuter

[nova remnant]

[nova remnant]

But where are my small books at

I'm here to learn

[bitter latch]

haha I put in Soushiki Zerozaki and got this . wtf

[native halo]

they banned me from the library for loud reading 😔

[bitter latch]

o...k

[nova remnant]

[bitter latch]

The anime flash

[native halo]

speaking of the flash

https://tenor.com/view/flash-baby-baby-microwave-flash-baby-microwave-flash-cooks-baby-in-a-microwave-gif-12517949662562511074

[vital raptor]

anime girl reading loudly in the library 😄

[silent kraken]

Let him cook

[nova remnant]

@native halo

[native halo]

[native halo]

(pov,from behind:2.5)

[nova remnant]

I have no idea why but... From behind is blocked on bing

[native halo]

from the rear 🤠

[nova remnant]

@native halo even more books

👍

[vital raptor]

(pov,from behind:2.5)
(pov, from below:2.5)
😏

just a heads up for anyone using TensorRT extension in a1111, don't upgrade it

It seems to be completely broken

[bitter latch]

meh never used it

tried it. waste of space

[vital raptor]

I only use it for models I use a lot, the speed boost is nice

I don't use many loras though, so i don't have to bother with that part of it

[bitter latch]

ah ha this is what I wanted

she has found the forbidden book

Anima manga comic page with panels, two tone, two sides, 4D render, Shadow and Bone, Magic and death, thru the void all are doomed, flame and lightning, plasma and anti-matter, in a girl it does live, extra details, elegant, sharp, gothic neon holographic light and color, fine art, soft colors, black outline style, very detailed Cel-Shaded With shaded detail's, soft shadows, illustration, detailed

sigh.. bored

now this looks more interesting

well dang that took a turn

goes back to his corner

[dusty siren]

I like the style, do you mind sharing the model and lora name?

[bitter latch]

not a lora

[bitter latch]

not a lora and custom merge, hmm lets see, nuelement, envydream and anime orange 3

now if I can just find the upload button on civitAI

pp[ o

i'm blind. doh!

[brittle rampart]

you have reached your upload limit 😂

[bitter latch]

no wasn't looking in right spot

I've renamed it NuOrangeMixerDream (NOMD)

uploading now gonna take about 20 mins or so

while that's doing that need to make me some sample images

[bitter latch]

here ya go https://civitai.com/models/260019?modelVersionId=293270

[brittle rampart]

more 5Gb?

[bitter latch]

.

[bitter latch]

5.28GB

I am Groot?

[brittle rampart]

i am Librarian?

[bitter latch]

nice

[brittle rampart]

maybe.... maybe...

[brittle rampart]

reminded me of an illustration I recently came across

[brittle rampart]

👍🏽 but where kawaii style?

[brittle rampart]

this is

[bitter latch]

and nope did not look at meta data

[brittle rampart]

did it "by eye"?!

[bitter latch]

yep

[brittle rampart]

well she is a very serious librarian

[bitter latch]

added ( kawaii style)

hmm didn't reallychange it that much

[brittle rampart]

yep...

[bitter latch]

tries a diff approach

changed to ( CGI 3D kawaii style)

[brittle rampart]

hmmm.. nope

[bitter latch]

it does give it more the look tho

I mean I could sorta cheat and use a lora 😄

[brittle rampart]

she looks pretty plasticky

[bitter latch]

so does yours

ponders

[bitter latch]

and just because

tried it with depth of field and cinamatic lighting 😄

[bitter latch]

hmm didn't mean to make starfield

a girl at the end of time looking out over a starless valley at night, epic detail, concept art, cinematic lighting, dynamic shadows, (unreal render:.5), 2.5D resolution, bright vivid colors, minimal lens flare and glare

[brittle rampart]

Watching move? interstellar....

[bitter latch]

nah

<-- random brain

plus was play starfield earlier.. so that might be why

[brittle rampart]

heh I have an interesting result... a rather similar composition concept comes out...

[bitter latch]

yeah different model and settings prob

[brittle rampart]

but I was talking about the composition itself. lighting

[bitter latch]

ahh

[bitter latch]

sigh, hate it when I got a image of a game that I want to play in my head but the name. nop

nope

would goto bed but I know I couldn't sleep cause of this wind

[brittle rampart]

look...

[bitter latch]

meh gonna try to sleep. nite. thud

[brittle rampart]

couldn't resist 🤗

[supple raptor]

My prompts are always fair game

[brittle rampart]

I don't steal...

I couldn’t resist answering with an illustration 😁

[brittle rampart]

I really didn't mean to offend...sorry. peace, friendship, chewing gum?

[nova remnant]

hopefully my sd works

[supple raptor]

No offense taken at all

[brittle rampart]

btw i have a question about SD upscale Script. you can answer?

[supple raptor]

Probably. I don't use it much anymore, but I used to use it a lot.

[brittle rampart]

here is this parameter. This is how I understand the overlap of small parts, but how does its value affect the output image?

[supple raptor]

Larger overlap value helps to reduce the visibility of seams between the tiles.

[brittle rampart]

i.e. the optimal value is more than 90? and in the process of overlapping there is some averaging of images at the seams?

[supple raptor]

I usually ran about 64px for tile overlap. More might be better sometimes, but can slow down the generation if it requires more tiles. The seams are blended together in the overlapping area.

[brittle rampart]

@supple raptor now how you upscaling your image? if it not a big secret...

[supple raptor]

I use tiled diffusion/multi-diffusion upscaling. These are my default settings and I run it at around 0.35-0.45 denoising. I have 12GB VRAM so you will probably need to adjust the latent tile batch size and/or tile size down to run it on lower specs.

[brittle rampart]

at what step is this? after the procedure hires.fix?

[gloomy scroll]

I'm in

[supple raptor]

Yep, I do txt2img > hires fix > adetailer (face only) > fix any problems > upscale with tiled diffusion

[brittle rampart]

in fact I am faithful to the process

[gloomy scroll]

I just turn on highres fix and adetailer and every image is perfect

[brittle rampart]

come back to SD1.5 😝

[gloomy scroll]

I'm using Soushiki

Look at these perfect images

[brittle rampart]

Well, now I have added to my list of favorite models. on par with the @Nubby models, it some Hello* models, Sakura, bluePencil

[brittle rampart]

i am get this is model and try to using...

[gloomy scroll]

I'm using quite a few loras atm

^pnginfo

[brittle rampart]

of course.. i know

[gloomy scroll]

I just mentioned since I mostly copy straight from ui and this server still doesn't have prompt inspector bot

[gloomy scroll]

Also loras should be here
https://huggingface.co/Surficialruby/LoraDump/blob/main/LoraMerge_dim128.safetensors
https://huggingface.co/segole2003/Lora/tree/main

[brittle rampart]

what this is?

[gloomy scroll]

Links for loras I used to get style in my images just posted if you were interested in the style

[brittle rampart]

ah oki-doki

[brittle rampart]

very Large list! 🙂

[gloomy scroll]

Yeah that person used to do quite a lot loras but now days they just use Dalle and Nai

[brittle rampart]

alas, I cannot spend tens of dollars on a subscription for the sake of regeneration NAI.

[brittle rampart]

this is really slowly

[gloomy scroll]

Depends on your settings since Tiled diffusion splits your image into tiles and the upscales each of those and then merges back to upscaled image so sometimes it could mean that it has to upscale tens of tiles

[brittle rampart]

these are the ones for that

time is

[gloomy scroll]

Does it say how many tiles it created in console? haven't used in long time

[brittle rampart]

of course

[gloomy scroll]

Ok no idea why 6 tiles takes that long

[gloomy scroll]

Actually how much vram you have?

[brittle rampart]

only 6Gb

[gloomy scroll]

Yeah I believe it did 3 images at the same time which isn't rly what you want with low vram

[brittle rampart]

Is it possible for my memory capacity to have very high settings?

[gloomy scroll]

very high how?

[brittle rampart]

dont know...

tiles size

[gloomy scroll]

If you mean like making 4k images then yes it should be possible

[gloomy scroll]

If you make 1 tile per batch then you should be able to do same as txt2img size you have

[brittle rampart]

hmm... i am try to make this is in img-to-img. this is my setting

jut try resize: from 1152x768 to 1612x1075 using T/D

[gloomy scroll]

Try changing tile batch size to 1 and see if it's faster

[brittle rampart]

Keep input image size?

[gloomy scroll]

If you are using Tiled diffusion it should use Scale factor in its settings for your output size

[brittle rampart]

I don't fully understand this moment

[gloomy scroll]

Input image size is just size of your image there shouldn't be setting for it

[brittle rampart]

then it can be checked?

[serene spoke]

Have i set it rightly?

[gloomy scroll]

Tile width should be like <512

[brittle rampart]

ammm this is another script for upscaling?

[gloomy scroll]

Yes

[bold ether]

https://github.com/Coyote-A/ultimate-upscale-for-automatic1111

[serene spoke]

Yes, they say me that improve quality output just using that

[brittle rampart]

ummm oki-doki then try it too

[serene spoke]

Is the only thing to correct? And 512 or less is better?

[gloomy scroll]

I'm not sure if tile height is needed or if it uses tile width if it's 0

Don't have that script installed atm

[serene spoke]

ok i'll check

[gloomy scroll]

Default settings should work just select upscaler model you want and if out of memory decrease tile width/height

[bold ether]

This is the only obvious thing to correct.
But understanding what those settings do would help you figure it out. If there was "one preset to rule them all" the devs wouldn't have bothered to expose those settings.
(by default tile height is at 0, and type the pattern / Type is "linear")

[serene spoke]

Hi, i'm a noob could someone help me with hydrus?

[gloomy scroll]

No idea what Hydrus is

[serene spoke]

it should be a tool to download images from gelbooru and danbooru

[serene spoke]

I read it here https://huggingface.co/hollowstrawberry/stable-diffusion-guide/blob/main/README.md#index

If you have enough VRAM, you may increase the Tile width as well as the Padding. For example, doubling both of them. Tile height can remain at 0 and it'll match the width.

[bold ether]

yes ?

[gloomy scroll]

How much vram you have?

[serene spoke]

12gb

[gloomy scroll]

Well then you can increase your tile size from default but not sure how much

[serene spoke]

yes but it say to increase only tile width so i'd did that

[gloomy scroll]

Yeah it probably uses tile width as tile height if tile height is 0

That's quite normal

[native halo]

because its a square tile theres no such thing as a triangular tile or a rectangle tile

[serene spoke]

So at least did I do correctly? or should i change something?

[gloomy scroll]

Yeah there wasn't any error but I just think that 2000 tile width was just little bit too much

But it's just trial and error what settings give you best results

[serene spoke]

Yes i set to maximux to check the vram, i'll make more try

[bold ether]

Very short explanation cause I've got 5 minutes for my stuff to finish cooking in the oven.

How does Ultimate Upscale Script works for mimics and alike.
Basically it's gonna split your (too) big image into multiple smaller ones called tiles. Upscale each one individually and then stitch them back together to make the higher resolution output.
What do those settings do ? cf screenshot
why blur the edges ? Because it helps them blend better. SD will figure out something correct from the blurry mess, don't worry.

So basically there's no real point pushing for higher tiles unless you're trying to output at absurdly high resolutions.

[gloomy scroll]

I normally don't even use those upscalers since I can just highres fix up to 2k with tiled vae and I'm just fine with that quality

[serene spoke]

Wait, but tiled vae isn't only for who have limited vram?

[bold ether]

Tiled vae works in a similar fashion-ish (but only for the VAE stage of the pipeline) and helps overcome vram limitation yes.

[gloomy scroll]

I have 8GB

[serene spoke]

Cristal clear

[serene spoke]

Since you have share it, cutoff and llul are extension for limited vram, right?

[native halo]

no

[brittle rampart]

heh i see you using Region Prompting! this is what i want know to use.

[native halo]

llul is for inpainting and cutoff is to limit the tokens reach within prompt

[brittle rampart]

btw ... now i have this is

[serene spoke]

and lora block weight?

[brittle rampart]

this is looks worse

[native halo]

to get different styles within the same lora by playing with the lora weights

[serene spoke]

Thanks i'm gonna study that

[gloomy scroll]

Did you use same seed?

[brittle rampart]

yes.

[gloomy scroll]

Well just try which settings you like and gen speed you are fine with

[gloomy scroll]

Seems fine to me

[brittle rampart]

but quality... left from Ultimate Script, right from T/D

[gloomy scroll]

I'll just take txt2img output and run with it

[brittle rampart]

same.. hmmmmmm

[gloomy scroll]

Witches gather

[steady pumice]

CUTE!

[brittle rampart]

what the LoRa dpep4?

[gloomy scroll]

I guess this
https://huggingface.co/closertodeath/ctdlora/blob/main/dpep4-768-000004.safetensors

[brittle rampart]

and some strange weight

[gloomy scroll]

Oh yeah that's lora block weight thing when I was testing the extension not rly sure if it's good or if it just made lora worse

[brittle rampart]

oki

[brittle rampart]

with Tiled Vae faster work controlNet?

[gloomy scroll]

If your vram goes to 100% then tiled vae can increase speed with controlnet otherwise it's probably just slower

[brittle rampart]

@gloomy scroll you use only Mask in RP?

[dapper ore]

Share the 5 most stable and beautiful anime models

[steady grail]

you forgot the magic word lol

hello

[brittle rampart]

he dont know it, hi hi !!!

[native halo]

the one with the 1girl

[gloomy scroll]

Pretty much yes

[gloomy scroll]

Aurora
Soushiki
Kakigori v3

[brittle rampart]

Mask you draw where?

[gloomy scroll]

You can do it in some external image editor also but you need to check correct layer colors

[nova remnant]

@gloomy scroll (now send the elf looking up)

[gloomy scroll]

https://tenor.com/view/milk-and-mocha-pat-cuddle-love-heart-gif-15740749

[nova remnant]

haha

[brittle rampart]

how you change color for mask layers?

[gloomy scroll]

Draw something and click this to apply to selected layer it will become colored

[gloomy scroll]

https://i.imgur.com/6KYMX2S.gif

[brittle rampart]

ah!!! so this is how it works...!!!!

[gloomy scroll]

Same color = same layer

and then in prompt every BREAK is prompt for each region

[brittle rampart]

closed area.. what if this is a mask for a layer that will be the general background?

[gloomy scroll]

I just use one layer for background althought there is probably better ways to do it

[broken osprey]

Hey guys one question, how do I use an already preprocessed posed on a1111?

I have the image with the nodes and all and comfy recognizes it well and produces a good output but in a1111 I’m not sure how to just send the image in and follow the pose without a base image

[steady grail]

do you mean an open posestuff?

or the pose of another image?

[broken osprey]

Yeah open pose stuff, Ive got some poses that are already processed so in like the skeleton form, but Im not sure if the preprocessor should remain empty or what

[gloomy scroll]

Empty

[steady grail]

have you installed the extension openpose? you can load it in the tab

[broken osprey]

oh okay so preproccesor goes empty and model with control 1.1 or t2i right?

[gloomy scroll]

If you have open pose stickman then you only need to set model in controlnet

[broken osprey]

ah thank you thank you ^^

[steady grail]

the preprocessor is used to generate a stickman, right?

[gloomy scroll]

Since preprocessor would try to process those stickmans and fail

[steady grail]

i don't touch auto from so long

[gloomy scroll]

Same with all types if you have preprocessed image you only need cnet model

[broken osprey]

Has anyone managed to get ipadapter to work on a111, Im trying to run it throuogh controlnet but i get these distorted ass faces and idk, the image sizxe is already as large as it can go and im not sure if I can use adetailer with the ip adater face model to fix it based on that :c

[nova remnant]

[broken osprey]

halp I need to reproduce the same character :c

[nova remnant]

See if someone in #📝｜prompting-help knows

[obtuse trellis]

I used it with a1111, usually with openpose or lineart for controlnet

I'm not sure if image size matters or anything I usually have mine output at 640x360 or so with hire fix and scale higher, img2txt seems to work better than img2img not home right now to check my settings it's all for an SD 1.5 model though no idea sdxl, the img input i give it is around 1600x900

I can give some basic examples in a thread once I have my computer might be a few hrs

[nova remnant]

@zinc flax

[zinc flax]

Cuute

[nova remnant]

I just noticed the tea cup in the corner

[zinc flax]

Oh no she's gonna spill it

[finite herald]

looks like the girl from pocket mirror kinda

[zinc flax]

That's one old horror rpgmaker

[nova remnant]

Guess I'm glad I don't know what it is

[zinc flax]

I can't believe a game from 2016 is old now

Scary

[nova remnant]

Noo

[nova remnant]

I still remember writing "2016" in school as the date

[vital raptor]

8 years is old?

[nova remnant]

8 ears old

fennec

[zinc flax]

I'd say it's old lol

Time really flies

[vital raptor]

it gets worse the older you are

[zinc flax]

https://media.discordapp.net/attachments/932280015809306784/991719785165504542/image0-1.gif

[finite herald]

for it to be 2016 and rpgmaker the art is 10/10

v gud

[nova remnant]

[vital raptor]

I dunno why she has wings 😄

[nova remnant]

[nova remnant]

BluePencilXL v3.1 dropped today https://huggingface.co/bluepen5805/blue_pencil-XL/tree/main

[serene spoke]

Hi, i need some help for training a lora. There is something that can help me?

[obtuse trellis]

https://civitai.com/articles/138/making-a-lora-is-like-baking-a-cake I know some ppl who make lora often refer to this guide, I don't myself so maybe someone else has more specifics

[serene spoke]

I'm gonna read it

[nova remnant]

Very beautiful large guide

👍

[obtuse trellis]

I can't take credit since i just link it but thanks

[broken osprey]

Hey guys, I was wondering if anyone could direct me to do one guide or idk if you would mind and know about it explain, how can I create for example characters hugging, or the like, I know maybe controlnet is key but how would I do this and also separate the characters into different hairstyles and the like

[obtuse trellis]

not sure the specifics ur asking but yea controlnet is best to start https://github.com/lllyasviel/ControlNet and then from there if ur using 2 chars u can put regional prompter on top https://stable-diffusion-art.com/regional-prompter/

this will control the character appearance they can be used with controlnet, results may vary depending on model and such but thats the general idea

alternative is as u mention just make them separate characters but that requires some photoshop or external editting tool to put them together after

[raw steppe]

love it

[raw steppe]

@grave snow what prompt/model/lora was it for this one? ^

[grave snow]

that last one? ah, since you asked 😄:

Positive Prompt:
poster for your album which stars wet dingrus and many digital art intricate (generated via nightmare-promptgen LLM)

Negative Prompt:
<neg-anime>, (lowres, low quality, worst quality, 3d, render, photo, photography, watermark, artist name, thighs, revealing clothing, hair intakes, japanese, chinese, text, copyright name)++ (Compel syntax)

model was two SD 1.5 models of mine edited for cranking out thousands to train on in my workflow, one to generate the starting image and one for latent --> tile controlnet upscale

my LoRA used on each denoise stage: https://civitai.com/models/95391?modelVersionId=101815

style LoRA @ 0.54 strength during upscaling: https://civitai.com/models/211819/retro-anime

[raw steppe]

i regret asking (jk)

[brittle rampart]

soda water 💖

where i can get upscalers?

[native halo]

here https://openmodeldb.info/

[zinc flax]

Huge

[bitter latch]

sighs

me b eing bored and can't sleep

[gloomy scroll]

[bitter latch]

now that's Ghibli

[bitter latch]

hmm the new Pokemon Arena's?

[bitter latch]

hmm

[bitter latch]

art fusion by Suzuki Kiitsu and Ghibli (Miku Hatsune:0.5) super girl of Japan, vivid pastel colors, line art , (pixelated background:0.5)

[bitter latch]

hmm it sure loves pink

[bitter latch]

tries to make that a saved style

[bitter latch]

ok thats diff..

[bitter latch]

ok that time of day. last one

[primal wharf]

🥢🍚

[nova remnant]

Yellow wins

[nova remnant]

:oh:

[broken osprey]

Okay this is a step forward to get my characters to actually hug lmao

I mean you can see the hand, the body, and the faces are appropiate Just need to refine the control over the bodies and the like

[nova remnant]

hugs

Hello vol

[broken osprey]

hi HIiiiiii, ye hugs, its something Ive wanted to do for stories you know, seemed like without controlnet it struggles with this task

[nova remnant]

I'm taking that one

Fluff

[nova remnant]

Did you try open pose

[broken osprey]

I haven't, since it needs to skeletons for the pose I would use it and try with the AND keyword or BREAK to separate the two characters but idk

[zinc flax]

Hinaaa

https://tenor.com/view/blue-archive-hina-sleep-mimir-gif-1645225618444061710

[zinc flax]

Very cute

[worthy delta]

hello fellas

ive improved

@supple raptor

[supple raptor]

looks nice

[gloomy scroll]

Rawr

[nova remnant]

Model

[supple raptor]

Animagine XL 3.0 Base

[nova remnant]

what I thought

It looks pretty nice

Is it deserving of mr nubby's approve

[obtuse trellis]

not that nsfw i hope

[nova remnant]

@supple raptor feet detected

[obtuse trellis]

🙈

[nova remnant]

Nubby is the expert in that area

[supple raptor]

https://cdn.discordapp.com/emojis/1182089870005502063.webp?size=64&name=aifeetpics2&quality=lossless