🍥｜anime

[lyric zealot]

okay, the next one is too nsfw so

[cerulean bane]

...you should share it to me in privat.

[lyric zealot]

[vital raptor]

more fantasy than historical..

[cerulean bane]

[lyric zealot]

[vital raptor]

i hate that this model thinks vikings have elf ears and horns

what garbage source material did they use for training

historically accurate norse tunic dress 👍

[signal panther]

HOW GOOD!

Prompt?

Pls?

[steady grail]

elves and Vikings comes from the same Cheltic culture. maybe they had pointy ears

[lyric zealot]

you are gonna be shocked to see it's super simple prompt, just good rng

1girl, hatsune miku, egyptian clothes,

source_anime,
Negative prompt: (worst quality, low quality:1.2), 3d, censored, abdominal muscle, artistic glow
Steps: 35, Sampler: Euler a, CFG scale: 7, Seed: 1377122372, Size: 768x1024, Model hash: ac006fdd7e, Model: autismmixSDXL_autismmixConfetti, VAE hash: 66f78ad136, VAE: sdxl_vae-fp16fix-blessed.safetensors, Denoising strength: 0.4, Clip skip: 2, ADetailer model: face_yolov8n.pt, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer mask blur: 4, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True, ADetailer inpaint padding: 32, ADetailer version: 24.1.2, Hires upscale: 1.5, Hires steps: 10, Hires upscaler: 4x_fatal_Anime_500000_G, Mask blur: 4, Inpaint area: Only masked, Masked area padding: 32, Version: f0.0.12-latest-155-gd81e353d```

[signal panther]

Cheltic?

Celtic.

[signal panther]

THANX

[vital raptor]

except they don't, vikings are danish/norse, celts are not

[steady grail]

oh

[vital raptor]

i mean, at some point they all comingle, but different cultures

[junior quest]

ohhh gimme gimme info...

[signal panther]

First result:

[steady grail]

https://scandinaviafacts.com/vikings-vs-celts-whats-the-difference/

[steady grail]

they have really too much in common, other than a temporal separation

[vital raptor]

that's like saying the english are a lot like the irish

[signal panther]

Burying people in bogs was a good idea.

[steady grail]

eheheh

[vital raptor]

they're all really close together

[steady grail]

the rest of the world don't see any difference, honestly XD

[signal panther]

FR?

No.

In my country, it's very easy to tell them apart.

Everyone knows that in my country.

[steady grail]

your country is closelly related to them?

[signal panther]

No.

But it's near them )

[steady grail]

so, here you have the explanation XD

[vital raptor]

just in the dane-english wars the vikings participated in covered wales, england, ireland, denmark, iceland, finland, etc

[steady grail]

this a relation close enough

[signal panther]

Wanna know my country?

[steady grail]

not that it's that important

[vital raptor]

due to the proximity and the wars, there's a lot of cross over with religions, bloodlines, etc

[steady grail]

they even have in common some mithology, that's what i was talking about

[signal panther]

Yeah I read Iceland sagas once. Interesting that english rulers had iceland warrios, and they valued them all

[signal panther]

Aha, okay

[signal panther]

https://tenor.com/view/cute-baby-how-cute-patrick-star-dreamy-face-gif-16667571

[steady grail]

damn, Norse mythology was too good to be lost like this 😦

[vital raptor]

lost?

[steady grail]

i deeply studied what's left and it's very little

compared to what should have been

many and many fragments of different legends, all incomplete, but so interesting

[signal panther]

Damn, the languag of Iceland is being lost, 'cause the subtitles on it don't fit the screen and it's a very hard language.

[steady grail]

uahahah

for example we know very little about Loki

[vital raptor]

he has his own tv show

😄

[steady grail]

they totally destroyed the character

[signal panther]

@lyric zealot

[steady grail]

the true Loki was such a bad ass

he knew how to live the big life

[lyric zealot]

i liked the second one

[signal panther]

Prompt: sigma pharaoh

)))))))))))))))))))))))

))))))))))))))))))))))

)))))))))))))))))))))))))

@lyric zealot or sigma tutunkhamen

[signal panther]

Wha-a

Is that the prompt?

[steady grail]

Loki was a damn rebel, fuXking sick of the hypocrisy and abuses of the other gods. Experienced everything was possible, and despised because he refused to cover the bull s*it of the other gods

[lyric zealot]

nope, i was just doing some makima prompt

[signal panther]

Like Prometheus

[signal panther]

Phew--

[steady grail]

yeah, pretty much

[signal panther]

@lyric zealot

[lyric zealot]

[steady grail]

was pretty canon what happened to Loki's childs. One was used by Odin as a mount and the other imprisoned because a prediction said they would have destroyed the gods

and what happened? his sons rebelled because the unjust imprisonment

🤣

[signal panther]

https://tenor.com/view/грустный-смайлик-gif-16570694716073351126

[steady grail]

but we lost most of this legend 😦

[signal panther]

https://tenor.com/view/nooo-meme-xcom-explosion-psionic-ahhh-gif-272505217465622822

[signal panther]

@lyric zealot

[lyric zealot]

[lyric zealot]

@vital raptor all of these on Stable Cascade?

[vital raptor]

sdxl

[lyric zealot]

[vital raptor]

the base cascade model isn't really much better than sdxl, and it's slow, so I haven't been using it

it's pretty good at prompt coherence, but so are my sdxl models so 🤷‍♂️

since it's a holiday this weekend i might spend some time seeing if i can do some custom training on cascade, but we'll see

[lyric zealot]

give it a month or 2 and there's gonna be tons of models out there

[vital raptor]

yeah

I was thinking of training some XL loras this weekend anyway, so if cascade is easy enough to do while i'm at it then i figure i might as well play with it

[vagrant falcon]

weird palm but meh

[junior quest]

what are those eye coloring name?

[lyric zealot]

there should be prompt, but you need the extension

[gloomy scroll]

I still don't have stealth extension in forge

[lyric zealot]

[lyric zealot]

1girl, white hair, hime cut, pointy ears, blue eyes, choker, white hoodie,

source_anime, chibi,
Negative prompt: (worst quality, low quality:1.2), 3d, censored, abdominal muscle, artistic glow
Steps: 35, Sampler: Euler a, CFG scale: 7, Seed: 2200733935, Size: 768x1024, Model hash: ac006fdd7e, Model: autismmixSDXL_autismmixConfetti, VAE hash: 66f78ad136, VAE: sdxl_vae-fp16fix-blessed.safetensors, Denoising strength: 0.4, Clip skip: 2, ADetailer model: face_yolov8n.pt, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer mask blur: 32, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True, ADetailer inpaint padding: 32, ADetailer version: 24.1.2, Hires upscale: 1.5, Hires steps: 10, Hires upscaler: 4x_fatal_Anime_500000_G, Mask blur: 32, Inpaint area: Only masked, Masked area padding: 32, Version: f0.0.12-latest-155-gd81e353d```

[gloomy scroll]

No artists

[lyric zealot]

yup no artist

[gloomy scroll]

Ok I just saw perfect image in my artist test preview

[lyric zealot]

[lyric zealot]

artist name?

[gloomy scroll]

(myama:0.8), (cirilla_lin:0.8),

[gloomy scroll]

Actual each of those images have 2 random artists

[lyric zealot]

you doing some dynamic prompt thingy?

[gloomy scroll]

Just 2 wildcard tags

[lyric zealot]

i should probably setup the umi tagging stuff

is that the name

[gloomy scroll]

I'm just using dynamic prompts extension for wildcards

Although I stole wildcards from Ao's Umi Debloat

[lyric zealot]

yup, i still need to get either that or the umi on forge

since it's like my go to webui now

[junior quest]

whats this mouth called? dat anime embarrassed screaming, or smthing like dat

I couldnt see this mouth from recent animes

[vital raptor]

🤷‍♂️ i didnt specify eye coloring, the prompt was girl,kimono, geisha,

[junior quest]

@vital raptor captain, how much temp 4090 gets?

[nova remnant]

at least 30

[junior quest]

eh? my idle is also 45

when generating its going bout 65

[nova remnant]

thats big

[junior quest]

yeah

Im now afraid

dunno what to do

clean?

[nova remnant]

I'd be worried if it was 90

thats normal you stoopid

Did you check the center temperature thingy

hotspot, was trying to remember the name

im bored

[junior quest]

not for me, ik its safe around 60-80 butttt who knows whats happening right

maybe it shows wrong and real temp is so high?

[junior quest]

gpu die temp?

[junior quest]

wut?

[nova remnant]

GPU hotspot temperature refers to the highest temperature recorded on a specific area or component of a graphics processing unit (GPU) during its operation. This temperature reading helps monitor and manage thermal performance and is crucial for maintaining the GPU's safe operating conditions.

[junior quest]

ohh hot-spot

I thought dat hotspot thing in mobile

[nova remnant]

Lmao

Check that if you can

It should tell you the highest temperature I believe that is what you should be looking at when generating images

[junior quest]

idle bout 48-55, when generating 80-70

[vital raptor]

My temps are 41c idle, 45c full power

During the summer when my AC is running my idle temps are closer to 35c

But I have a watercooled gpu

For what it’s worth, 80-90 is fine. Most gpus can handle at least 110 before they shutdown the system

But the warmer it is the slower it gets from thermal throttling

If it was 60-70+ idle I’d worry about it

[junior quest]

hmm I dun have seasons so mst of time same weather

[junior quest]

dahhhhh

matrix?

[nova remnant]

https://tenor.com/view/the-matrix-reloaded-matrix-reloaded-neo-keanu-reeves-gif-4011236

[bitter latch]

cackles evilly

[junior quest]

to mappa?

[native halo]

mappa with their cgi?

[bitter latch]

huh.. confused

[native halo]

https://tenor.com/view/kermit-diary-gif-20895163

[bitter latch]

no clue what mappa is

[junior quest]

[native halo]

https://tenor.com/view/steve-buscemi-how-do-you-do-fellow-kids-old-gif-11451463

[junior quest]

https://huggingface.co/destitech/controlnet-inpaint-dreamer-sdxl/tree/main/models
found a inpaint model for controlnet sdxl

5GB

[terse ore]

I have a 2048x3072 image and am using SegmentAnything+Grounding Dino to generate a mask to inpaint the hair of the character. However, I get a OOM error. How can I generate inpaint without getting OOM for a large image? It should in theory only inpaint and thus use not too much VRAM because it's only the hair part (and I have a 3060 with 12gb)

[native halo]

sdxl inpainting is laggy and gives u OOM on a1111

[terse ore]

I was using a 1.5 model to inpaint though

[native halo]

try on forge its better optimized for controlnet

[terse ore]

Maybe I'm not doing it correctly somehow

[native halo]

if nothing else works try to enable the -medvram flag maybe it will help with such a large image

[sonic bolt]

how do i get rid of the all the ribbon like things

[vital raptor]

depending on what's causing it, it could be as simple as lowering your CFG Scale

[native halo]

probably the model,a lot of those 1.5 pastel models have these artifacts like confetti,ribbons,flowers,butterflies etc...

[sonic bolt]

ight

[sonic bolt]

gotcha

[junior quest]

is anyone using svd?

[vital raptor]

i've tried it

[junior quest]

I see

btw do we have a add detailer for sdxl?

https://civitai.com/models/122359?modelVersionId=135867
this one?

[vital raptor]

it's the same one

adetailer

[junior quest]

add detailer works?

nice

how do I fix triton?

[vital raptor]

you don't

[junior quest]

its a python optimizer or smthing right?

[vital raptor]

supposedly it's linux only

[junior quest]

I see

[supple raptor]

Inpainting is done at the resolution of the image unless specified otherwise. You need to crop the image, inpaint, then stitch the inpainted area back into the original image. In A1111 this is done by selecting the inpaint only masked button which will do that process but also adds rescaling the cropped area to the resolution you set in the UI, before inpainting and then rescale it again to the original size before stitching it back into the original image.

If you're using Comfy, the steps are the same, but you'll have to assemble the nodes to do all of the steps yourself.

[junior quest]

release her

NOW

[junior quest]

shes like her kinda

[junior quest]

waittt it needs to be a pleated skirt, tight skirts doesnt fit

[junior quest]

oh i forgot bout her

[junior quest]

is it okay to use normal anime images as regularization images?

[nova remnant]

No

[nova remnant]

Needs a bigger sword

And tail

[supple raptor]

yesh

[supple raptor]

nyo

FrierenSmug

[nova remnant]

Thick tail

👍

[native halo]

yes

https://tenor.com/view/dagashi-kashi-hotaru-shidare-anime-cute-gif-13267876

[nova remnant]

Anime gif

[native halo]

https://tenor.com/view/gojo-satoru-fortnite-dance-anime-dance-gojo-dance-gojo-satoru-fortnite-gif-9635138092968300656

[junior quest]

@supple raptor for regularizaion images what if we use normal anime images?

[supple raptor]

I haven't really messed with regularization images in a long time (I don't train much), but from what I recall it's best practice to generate the regularization images on the model you're training on as they should represent what the model already knows the represented concepts to look like.

[junior quest]

dah, hmm...

[native halo]

https://tenor.com/view/travis-scott-excited-birthday-yelling-shouting-gif-11756534

[junior quest]

found sm info from pixiv

Model: 0.7_0.7_ghostmix_v20Bakedvae_+0.3_divineelegancemix_V5__+0.3_cetusMix_Whalefall2_+shinyOiledSkin_v1fp16,

now how do we do dat?

[native halo]

supermerger

or sd-webui-model-mixer both works

[junior quest]

kay thanks

[nova remnant]

cx

[junior quest]

where yellow hair?

wait is dat her tail or ?

pikachu?

[nova remnant]

Shes pikachu human form

[junior quest]

then where is dat YELLOW hair

[nova remnant]

[junior quest]

oh Its so annoying to do inpaint

[nova remnant]

then dont

Just generate a more perfect image

[junior quest]

I see

[nova remnant]

@junior quest I put blonde it didn't work

Pikachu girl is brown haired

[junior quest]

sadge

[nova remnant]

Grabby tail

[lyric zealot]

no grabby

[nova remnant]

@cerulean bane

[junior quest]

can anyone generate a girl with a confidene face?

[junior quest]

similar to this kinda

[nova remnant]

Put looking at viewer

And it might generate that

[junior quest]

hmm like expressionless, confidence, from below?

[junior quest]

I need that head tilt

[nova remnant]

She's literally doing nothing

[nova remnant]

@primal wharf the p mouth

[lyric zealot]

[junior quest]

hmm it looks goood with those texts

[nova remnant]

Armpit

[near elbow]

random anime result 🙂

[vagrant falcon]

what upscaler you guys are using?

[steady grail]

i like animesharp

[lyric zealot]

Im using mostly 4xfatal

[junior quest]

deym novel ai can generate hands pretty accurately

I didnt find any image that has deformed fingers and stuff from pixiv

NOVELAI

[nova remnant]

grab

[nova remnant]

.>

people posting images that aren't inpainted 40 times on pixiv?!

[junior quest]

hmm nah I dun think

maybe 2 or 3 times

[vital raptor]

4x-UltraSharp

[terse ore]

I'm trying to use the Segment Anything+ Grounding Dino tool but after I generate and select the mask I want, where do I click to actually generate the inpaint?

[vital raptor]

anyone have an interesting idea for a SDXL lora? I was going to do some lora training this weekend just for the hell of it, but I don't really have a need for anything so I don't know what to train

[terse ore]

I don't know if it's possible but something that "cleans up" generated images? Like if we use img2img to make realistic trees into anime trees, you'll notice that a lot of the time there are floating leaf artifacts around the trees for example. And other messed up parts.

[vital raptor]

that's not generally what loras are for, you could fix that with better checkpoints or generation settings

or inpainting

[terse ore]

Maybe a gymnastics/flexibility lora? Most generations can't do stuff like backbends, scorpion pose, etc. I saw a model like that for 1.5 a while back but it wasn't ever released. Just an idea

[vital raptor]

that might be a good one actually

could be some work finding good images to train on

one reason it may not be popular is that the majority of reference images are going to be children based on a quick google search

yoga poses will have more adults

might be easier to just use openpose or something to generate the images since you could have a number of different angles for any given pose and the model is just going to end up blending them unless you only train one or two

loras are best for very specific things

[terse ore]

Oh. I tried using the openpose for that a few times and it gets confused sometimes about disjointed legs and feet and stuff. But you're probably right

[vital raptor]

this is with canny using a reference image, it's not great but i don't really know much about controlnet

this is the reference i grabbed off google

could do better with some tweaking and maybe a better reference

[terse ore]

Hey that's pretty good! I think when it's like a leg going behind the back and covered by the body, sometimes it thinks it's an arm and puts a hand there and stuff. I guess it all depends on the pose.

But it looks like her chest and back got flipped around lol

[vital raptor]

this is just from prompting gymnastics pose and scorpion pose

it messes it up most times, but it knows what it is

[terse ore]

Yeah 3 feet lol. But it does have an idea of those things. It will perhaps work with more controlnets and a little more refinement, stuff like that

[vital raptor]

just using 'gymnastics pose' comes up with some pretty good ones if you're not after a specific pose

extra feet and the floating tail is a simple inpaint fix

you're never getting a perfect image first try

[terse ore]

True.

[vital raptor]

another simple one, but it messed up the arm

probably a better model to use too, i'm just using SugoiXL to see what it can do

pose wise, this is damn near perfect

after running a batch of 100 images, it looks like it's confusing the gymnastics scorpion pose with the yoga scorpion pose

[terse ore]

Didn't know there was a difference tbh. Probably can skew things if you add "standing" or something like that. There are lots of ways to change it I guess

[vital raptor]

this is the yoga one

versus gymnastics/cheerleading one

but yeah, some extra details in the prompt might help

[nova remnant]

Went out for a few hours

Come back to this

[vital raptor]

(gymnastics pose:1.1), standing, (scorpion pose:1.1), from side, arms up,

[nova remnant]

(gymnastics pose, dynamic pose, action pose:1.3)

[nova remnant]

https://tenor.com/view/cat-running-away-escape-getaway-bye-gif-16631286

[vital raptor]

but also this one 😄

[nova remnant]

She angy

[vital raptor]

figures, i got that one image of a proper scorpion pose and i can't get it to do it again

[nova remnant]

(Scorpion pose:1.8)

Surely

[vital raptor]

surely

[nova remnant]

I love tripod girls

[vital raptor]

just using (gymnastics pose:1.1), (standing:1.1), from side, (arms up:1.1), actually gets closer more often than when i include scorpion pose

too many conflicting terms i'm guessing

[nova remnant]

What if you only try scorpion pose

Or make it gymnastics scorpion pose

[vital raptor]

it's all basically the same, the prompt i have was best results

a different model might do better

[nova remnant]

We need the Olympic model

[vital raptor]

this all started because i was trying to think of something to train a lora on

gymnastic/yoga poses was suggested

i was trying to figure out if it's worth the trouble of finding training images if a normal model could already do it

it might be worth the trouble, even if it's a different lora for each pose, just have to get the training data collected

[bitter latch]

hmm gymnastics

so far no success they come out umm very weird

[vital raptor]

it definitely works better to just prompt gymnastics pose, from side and then let the model do what it wants 😄

[nova remnant]

👍 very athletic

[bitter latch]

almost

[vital raptor]

ugh, why is img2img such garbage, can't use xformers or hires fix

have to either inpaint a pre-upscaled image, or use worse upscaling afterwards

[terse ore]

I used inpaint upload to upload an image and it's mask generated by Segment Anything and Grounding Dino. Then I selected Inpaint area Only masked. But when I try to generate, it says out of memory. (Image is 2048*3072 and is a 3060 with 12gb vram)

[vital raptor]

12gb usually isnt enough for an image that size

[steady grail]

well, he can cut a portion of the image and inpaint it

[terse ore]

I thought the inpaint means it only does that for the masked area so it would in fact just be a small part of the image and not need so much VRAM

[vital raptor]

it still has to put it back in the full image

and blend it

using dynamic prompts to pick random artists is interesting 😄

[terse ore]

How does one inpaint a high resolution image?

Or I should just Google it?

[vital raptor]

probably the easiest to to downscale the image, do the inpaint, then use a tiled upscaler if you don't have the vram

[nova remnant]

cut the part what you'll inpaint I guess

[vital raptor]

H.R. Giger's style works so well with my model

[nova remnant]

to a reasonable size then stitch together

[terse ore]

So there are two options. 1. Downscale, inpaint, then upscale. 2. Crop, inpaint, then overlay on the old image

[supple raptor]

Are you using a1111 or comfy?

[supple raptor]

You got it almost right but you also need to set the image resolution to a size that you can actually generate. It will crop the image to the mask + padding pixels (in the aspect ratio of the set image dimension) then upscale the cropped image to the resolution you have set, inpaint, downscale and stitch back together. With inpaint only masked, the resolution you set it determining the size during the inpaint, not the final output image.

[terse ore]

I am using a1111 but only for inpainting. I see your other message, I'll save it and try that

[terse ore]

Woo I did it! Thank you so much! I can be a pro now

[supple raptor]

happy to help

[ionic goblet]

whats a good Upscaler for anime/cartoon ( not semi-realistic)

[supple raptor]

depends, do you want it for standalone upscaling or for using with hires fix/tiled diffusion upscaling?

[ionic goblet]

not sure, would just like for it to look like the cartoon show

[warm oracle]

generating cute stuff for my friend and this was one of them

this too

[native halo]

[supple raptor]

An upscaler shouldn't significantly change the style of the image. I think you should probably look into using img2img and/or controlnet assuming you already have an image that you want to "animeify".

[junior quest]

its a win

[gusty palm]

hii i saw a reddit post for auto111 update

ran update bat but says im up to date is that ok

im asking here cuz yall are smart

[supple raptor]

last update to the main branch was Dec 15th. It could have been talking about the dev branch or possibly the forge fork

[junior quest]

still 1.7.0 for me

[gusty palm]

ooh

hum

giess i didnt check the date lol

it just popped up in my google news feed

[junior quest]

Model: 0.7_0.7_ghostmix_v20Bakedvae_+0.3_divineelegancemix_V5__+0.3_cetusMix_Whalefall2_+shinyOiledSkin_v1fp16,
um..so bout that, there is a lora called "shinyOiledSkin_v1" not a checkpoint model, so is it still possible to merge lora with models?

and also lora name is "shinyOiledSkin_v1" not "shinyOiledSkin_v1fp16"

[native halo]

yes u can merge a lora into a checkpoint with the extensions i linked earlier

[junior quest]

nice

[junior quest]

is sdxl inpaint works better than sd1.5?

[gusty palm]

for the infill thingy.. do you take the image and change the canvas size in gimp with transparency and then put it into img2img

like this

do i set the img2img tab to inpaint or

[supple raptor]

Here's an even better method that helps control colors better: https://www.reddit.com/r/StableDiffusion/comments/1aexch9/using_mosaic_tiles_to_outpaint_expand_images_3/

[gusty palm]

ooo

i am trying something similar useing clone tool

n then smudging bluring it all up

what kind of denoise do i need

not sure what ti put for the 'masked content' portion

ooo its kinda working

just got some funky looking edges

ok after upscaling it a bit the edges evened out thats so cool

[acoustic locust]

drake? is that you?

[acoustic locust]

what model?

[junior quest]

@native halo Model: 0.7_0.7_ghostmix_v20Bakedvae_+0.3_divineelegancemix_V5__+0.3_cetusMix_Whalefall2_+shinyOiledSkin_v1fp16,

guide me

I cant find 3places to enter those 3 values

[native halo]

sorry but this recipe Model: 0.70.7_ghostmix_v20Bakedvae+0.3divineelegancemixV5_+0.3_cetusMix_Whalefall2+shinyOiledSkin_v1fp16 doesnt have enough info,idk if he used a weight sum or a triple sum or if he enabled MBW or whether he used normal mode or cosine or smooth etc

[junior quest]

hmm I see lets try your other extention what it has

[native halo]

supermerger

[junior quest]

other one? mixer?

[native halo]

this one? https://github.com/wkpark/sd-webui-model-mixer

[junior quest]

yah

seems like this is the one

but I dunno now how to load a lora with this

[native halo]

i think its in second tab but only for extracting loras

[junior quest]

lora:add_detail:0.5:1,1,1,1,1,1,0.2,1,0.2,0,0,0.8,1,1,1,1,1
lora:yjk:0.5

Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 20872439, Size: 512x768, Model hash: a074c6ba71, Model: 0.7_0.7_ghostmix_v20Bakedvae_+0.3_divineelegancemix_V5__+0.3_cetusMix_Whalefall2_+shinyOiledSkin_v1fp16, Denoising strength: 0.4, Clip skip: 2, ENSD: 31337, ADetailer model: face_yolov8n.pt, ADetailer conf: 30, ADetailer dilate/erode: 32, ADetailer x offset: 0, ADetailer y offset: 0, ADetailer mask blur: 4, ADetailer denoising strength: 0.25, ADetailer inpaint full: True, ADetailer inpaint padding: 32, ADetailer use inpaint width/height: False, ADetailer use separate CFG scale: False, ADetailer version: 23.5.10, Hires upscale: 2, Hires upscaler: R-ESRGAN 4x+, Dynamic thresholding enabled: True, Mimic scale: 13, Threshold percentile: 100, Mimic mode: Half Cosine Up, Mimic scale minimum: 0, Eta: 0.2

[native halo]

thats not the recipe of the model,thats the lora blockweights used in prompt and the cfg fixer settings

[junior quest]

[native halo]

just try with a weight sum in suppermerger using that halfassed recipe maybe it will work

[junior quest]

we cant get recipe from png info right?

[native halo]

no just the name of the model

my guess would be he just did a simple weight sum

[nova remnant]

[lyric zealot]

based on the name, it seems like the person did two merged, first: ghostmix at 0.7 and divineelegancemix at 0.3 = output1 then output1 at 0.7 and cetusmix at 0.3

[junior quest]

yeah I thought bout that, but if they created a output1 then it need to be renamed right? so if it renamed, then that name is the one showing in png info right?

so it needs to do smthing with generating image?

[lyric zealot]

no, if the person used auto1111 Checkpoint Merger, and he didn't specify a custom name, then it will be auto generated name, exactly like the one you see with the 0.7 or 0.3

[junior quest]

oh I didnt know dat

nice

[lyric zealot]

that's why i kinda guess how they make that

you can try if you want, just get two models, and merge them, but don't include any custom name

[junior quest]

okay okay

lora:adddetail:0.5:1,1,1,1,1,1,0.2,1,0.2,0,0,0.8,1,1,1,1,1

and how do I do I enter those values?

just type in the promp?

[gloomy scroll]

You also need lora block weight extension

[junior quest]

like this?

[gloomy scroll]

Just let that extension be default and only add <lora:adddetail:0.5:1,1,1,1,1,1,0.2,1,0.2,0,0,0.8,1,1,1,1,1> to prompt

[junior quest]

good?

[gloomy scroll]

Don't edit anything in that image

just use default

[nova remnant]

7 grabby

[gloomy scroll]

You see this behind your window. Do oyu let her in?

[lyric zealot]

https://tenor.com/view/menhera-chan-chibi-angry-anime-gif-25268544

[lyric zealot]

https://tenor.com/view/yesssss-agree-anime-cute-point-gif-15932028

[gloomy scroll]

Yue thanks you for snack

[nova remnant]

Small

[gloomy scroll]

Here take smol

[nova remnant]

Too small

[junior quest]

yabe

[junior quest]

ohh maaa god

[nova remnant]

Pikachu like

[junior quest]

pikachu: slap hard
she: goes to sleep

[wooden coral]

Cascade can do anime

[junior quest]

hmm interesting

[wide geyser]

whats cascade?

[wooden coral]

https://discord.com/channels/1002292111942635562/1207078178510872636

[wide geyser]

guys can I request a generation please?

[junior quest]

wdym?

[gloomy scroll]

Don't do anything with the extension just use default values it had there is no need to even open it since it's enabled by default and only change prompt

[nova remnant]

with that name

yes

xD

[wide geyser]

Yayyyy

So I used to use google collab

and I promised my girl a meme

cause we both are huge spiderverse fans

and shes like my twin like whilst were talking we have everything in common

[junior quest]

souka souka thanks

[nova remnant]

🥢 sushi talk

[wide geyser]

so could you make a comic book cover, gwen stacy in her apartment, chatting on her phone, pointing at the viewer, laughing. (and if u can title the cover can you title it "Us moment"

[junior quest]

oh im doing it again

habit

[nova remnant]

https://media.discordapp.net/attachments/931925914470711346/1061076535525785680/321666B2-CC36-467F-B254-56055DEB1C8C.gif

You know what

I think you need dalle for that

[gloomy scroll]

Here chopsticks for you

[wide geyser]

I have a lora for it that I used to use but I cant use it anymore

[wide geyser]

https://civitai.com/models/210095/the-wizards-vintage-comic-book-cover

[nova remnant]

I can't download loras, I don't use my pc to generate cause its too slow and bricks often

[wide geyser]

oh what do u use?

[nova remnant]

dalle would've been able to do that but for some reason its blocking that prompt

[wide geyser]

Can you pretty please just make a girl looking at her phone and laughing at it

[nova remnant]

@wide geyser https://www.bing.com/images/create try this

Its reaally good at a bunch of things

it just censors a lot of things

@wide geyser

a literal the wizard's comic book lmao

wiziard

[wide geyser]

how did u make this

[nova remnant]

The wizard's vintage comic book cover style, A comic book cover featuring a blonde girl with pale skin, talking on her phone and laughing, she is pointing towards the camera

[wide geyser]

could you please try this:

comic book cover titled "us moment", 1 girl, sitting in her apartment, laughing at her phone

[junior quest]

[nova remnant]

US MURICA!!

[wide geyser]

ACTUALLY

THANK U

SO MUCH

[nova remnant]

can't type "thighs", too nsfw

[junior quest]

then put stocky

[wide geyser]

ONE LAST THING PLEASE 😭😭

could you make one exactly like this but with a boy instead of a girl

[junior quest]

or plump??? not sure

[nova remnant]

any more description

Dalle will literally make a strong african dude

[wide geyser]

uhhh

[nova remnant]

or skinny asian wearing rainbow shirt

[wide geyser]

curly hair

wearing headphones

[junior quest]

[wide geyser]

pale

[nova remnant]

Gotta force all types of people in random generations 👍

[nova remnant]

grabby

[junior quest]

[wide geyser]

comic book cover titled "us moment", 1 boy, black loose curly hair, wearing blue headphones, sitting in his apartment, laughing at his phone

[nova remnant]

@wide geyser hair length ?

eye color

[wide geyser]

uhhh medium ish

[wide geyser]

brown

like this

[nova remnant]

Is that gud

[junior quest]

need a vampire girl

[wide geyser]

its perfect but can u just remove the emojis

[junior quest]

oni horns, pointy ears, fangs

[nova remnant]

@wide geyser editing is not possible, I just generated more

this one has 2 views

for some reason it likes heart emojis

[junior quest]

I like her

[nova remnant]

I wonder why

she will suck your blood

[junior quest]

hmm I'll give my blood after I tame her

not by force

[nova remnant]

shes not a pal

[wide geyser]

could you like make him a tad bit younger

[nova remnant]

[junior quest]

anyone know how to get a 1horn?

center of the forehead?

[wide geyser]

😭😭

[nova remnant]

type unicorn

[junior quest]

oh good one lets try

[nova remnant]

@wide geyser

weird hand

[wide geyser]

thanks

what software are u using?

dalle?

is that paid?

[nova remnant]

you have 15 tries free every day

@wide geyser all the best ones I could try

[wide geyser]

I appreciate u so much man

[wide geyser]

where can I use it

[pure oyster]

like this?

[nova remnant]

the link I sent

bing create

[wide geyser]

this is dalle?

[nova remnant]

yes

[pure oyster]

definitely higher quality than SD 1.5

[nova remnant]

Its not that its higher quality

its just muuuch smarter

holy booba

[junior quest]

shes going to transform, deym its not wrking

oh thanks, wooow

[lyric zealot]

https://tenor.com/view/anya-forger-rub-eyes-stare-oomfie-spy-x-family-gif-6959524118542060626

[nova remnant]

no looking

[junior quest]

yeah you're seeing illusions

[lyric zealot]

[pure oyster]

More dommy mommy?

[junior quest]

its jut not create a single horn

@vestal crypt how is novel AI?

what artstyles it has?

[pure oyster]

Imagine stabbing yourself in the thigh every time you tie your shoes.

0/10

[nova remnant]

pretty much infinite

[lyric zealot]

[pure oyster]

model btw?

[lyric zealot]

asking me?

[pure oyster]

yeppers

[nova remnant]

smol is autism

[lyric zealot]

[pure oyster]

reply function is for losers

[lyric zealot]

like cat said, it's autismmix

[pure oyster]

fuck, it's sdxl, probably can't run it.

[lyric zealot]

im running on my 8gb vram gpu

so if you have 8gb vram, might be possible

[pure oyster]

AMD card

[lyric zealot]

tho there's a new thing the mimic having cuda for AMD card

wtf is this, a new scam?

@wooden coral #🍥｜anime message

[nova remnant]

if you're on linux its doable

unless its a bad card

[pure oyster]

I would probably nuke my computer attempting to make a dual boot.

[junior quest]

whaa

thats too good

prompt?

[nova remnant]

theres an official thing you can download in windows to do it, not super complicated

[lyric zealot]

1girl, red hair, blue eyes, single horn,

source_anime
Negative prompt: (worst quality, low quality:1.2), 3d, censored, abdominal muscle, artistic glow
Steps: 35, Sampler: Euler a, CFG scale: 7, Seed: 1677137885, Size: 768x1024, Model hash: ac006fdd7e, Model: autismmixSDXL_autismmixConfetti, VAE hash: 66f78ad136, VAE: sdxl_vae-fp16fix-blessed.safetensors, Denoising strength: 0.4, Clip skip: 2, ADetailer model: face_yolov8n.pt, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer mask blur: 32, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True, ADetailer inpaint padding: 32, ADetailer version: 24.1.2, Hires upscale: 1.5, Hires steps: 10, Hires upscaler: 4x_fatal_Anime_500000_G, Mask blur: 32, Inpaint area: Only masked, Masked area padding: 32, Version: f0.0.14v1.8.0rc-latest-184-g43c9e3b5```

[junior quest]

whats dat score meaning

[lyric zealot]

@nova remnant please explain, im eating

[junior quest]

deym I clicked dat link

[nova remnant]

special tag that makes quality better

[junior quest]

f me

[nova remnant]

good luck

👍

[gloomy scroll]

Pony based model quality tags

[junior quest]

so are they work for all of em?

[gloomy scroll]

English?

[junior quest]

eeh?

[gloomy scroll]

What all?

[junior quest]

models models

sd all

sd 1 2 xl

all

oh wait

you said pony based

[gloomy scroll]

Only for Pony and mixes based on that

[junior quest]

right

yah yah my bad

thanks

[lyric zealot]

it's a tag trained on ponyXL, but they made some mistake when training it, so instead of just score_9 you need to do score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up

therefore, any mixed that contained ponyXL can use that tag

[vital raptor]

probably doesn't need them though

mine doesn't

[lyric zealot]

it's kinda like those quality tags

kinda like masterpiece, best quality

[gloomy scroll]

Yue teaching you magic

[lyric zealot]

with/without

[nova remnant]

:oh:

[lyric zealot]

[junior quest]

oh now thats why my images not good, I see I see thanks

I stopped using it

because of dat

[nova remnant]

not like I explained it

[junior quest]

[signal panther]

Anyone checked this?

https://perchance.org/ai-text-to-image-generator

Great thing.

All of my images are from there.

[gloomy scroll]

[lyric zealot]

https://tenor.com/view/onimai-anime-sleep-mihari-oyama-gif-27609322

[vital raptor]

civitai is free too

and soon will have all checkpoints available that are hosted on there

[signal panther]

Hmmm...

[gloomy scroll]

NAI isn't

[signal panther]

Never heard)

[vital raptor]

it's the site everyone uses to get models and loras

for generating locally

[signal panther]

Okay. Thx for info)

https://media.discordapp.net/attachments/1025266140445933648/1205363441871753266/ezgif-5-2b156a8e89.gif

[nova remnant]

stupid

[vital raptor]

with and without all the score tags on a model with ponyXL mixed in

[nova remnant]

no more pony

[vital raptor]

it's there, it just doesn't need all those stupid tags

dunno why they don't just fix it

too late now i guess, the word salad has flooded civit

[nova remnant]

cause you can use it

or you can not use it

its made as an option

[vital raptor]

tbh, i haven't really messed with the base pony model much, it got really annoying to not get even kind of decent results from a normal (for me) prompt

but I don't like writing books to make images, so i specifically make models that don't need it 😄

this is the same seed/prompt as the previous 2 images on the base pony model

[junior quest]

hmm its not bad for a free one

[junior quest]

okay thats pretty good I'll say

[nova remnant]

pony is really good

[vital raptor]

it would have to be to get that many downloads

unfortunately, it just doesn't seem to work for me

just copied gen data from one of their sample images on civit and got this

🤷‍♂️

[nova remnant]

what is the prompt

[vital raptor]

score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, rating_safe, Zipp Storm, ((cute, little, fuzzy pony, fur)), (high quality, detailed, beautiful), shiny, adorable face, detailed beautiful eyes, diadema, sunlight, realistic, outstanding, countershading, detailed soft lighting, ear fluff, hoof on face, cinematic vintage photography

[nova remnant]

[vital raptor]

hmm

seems to be a token problem

if i delete most of the garbage it works

[nova remnant]

worked for me

that prompt

[gloomy scroll]

This pony model?

[vital raptor]

apparently having the pad prompt/neg length option in a1111 completely breaks pony if you go over 75 tokens

which you have to because of the quality tags

[gloomy scroll]

Try autismmix_confetti

[vital raptor]

my own mix works fine, i was just trying to figure out why the base model wouldnt work

i imagine if i was generating specific anime characters it might be useful

[nova remnant]

Is there any reason you put pony in the mix

[vital raptor]

for testing

i wanted to see what it did to my existing model since it was supposedly super awesome

i think the only reason i haven't played with the mix more is because something in the pony model breaks tensorRT, and sometimes i dont want to half my gen speed

[nova remnant]

[vital raptor]

and then i made SugoiXL so i've been testing that

[nova remnant]

It's so hot

I need to generate some snow

[vital raptor]

you can have mine

it's around 1C here right now

[nova remnant]

Lmao

[vital raptor]

lol, i tried eevee in a hoodie

[nova remnant]

[nova remnant]

Too cursed

[junior quest]

it req a name

[lyric zealot]

What is your settings in your checkpoint merger tab

[vital raptor]

set model c the same as a and see if it fixes the error

[lyric zealot]

It’s been so long since I used the checkpoint merger, and I’m not on computer, so can’t check

[junior quest]

nah its not doing anything, its clearly saying need a name

[vital raptor]

so put one in?

[lyric zealot]

So you meant if you put a name in custom name, it works?

[junior quest]

yeah no problem

[junior quest]

yah

ofc

[nova remnant]

Looks so unsettling

[lyric zealot]

Then go ahead with that then, I mean even you can see that they display optional over there

So maybe your thing is kinda buggy?

Or they updated it and now it didn’t work without custom name, but they forgot to remove it?

[vital raptor]

could be worse..

[nova remnant]

Basically Isabella

The dog from the farming game

x3

[vital raptor]

😄

[nova remnant]

[nova remnant]

@vital raptor

[nova remnant]

[acoustic locust]

what model?

[supple raptor]

autismmix_confetti
also used regional prompting to separate the two character prompts.

[lyric zealot]

[nova remnant]

you're not online

[supple raptor]

SMUGGERED

[lyric zealot]

Just like you,

[nova remnant]

I don't have a friend anymore

hes never online to talk to me

[lyric zealot]

I never have that friend

[craggy turret]

Hmm

I actually kinda like it

[acoustic locust]

That’s cool, what model?

[craggy turret]

AnimagineXLV3

[craggy turret]

FUCK NOT AGAIN

AAAAAAAAAAAAAA

[supple raptor]

?? What's wrong. It's just installing requirements.

[craggy turret]

It took like half an hour to fully install them again 💀

[supple raptor]

Oh, I guess your internet is not very fast. Yeah it's like 7+GB of stuff that it has to download if nothing is cached.

[spice shard]

sorry I lost for few month, would you tell me where I can find dream bots list?

[native halo]

bots are offline

https://tenor.com/view/dono-wall-talking-wall-bricks-gif-17741481

[supple raptor]

#1047610792226340935

[craggy turret]

Hm, interesting style

[wheat falcon]

Hey does someone know what
score_9 or score_8_up, means in a prompt?
Which extension does that?
Wildcards?

[vital raptor]

it's quality tags for pony based models

instead of "masterpiece, absurdres" etc

[wheat falcon]

So its references to stuff trained on these specific models? Without using a lora etc

[vital raptor]

yeah

it's specifically for ponyXL based models

[wheat falcon]

Ahh very interesting

Thanks for the explanation!

[vital raptor]

i'm actually testing them right now with a couple of mixes based on pony

for mixed models the affects are pretty minimal, and probably aren't needed at all, but for the base model it's basically required

[wheat falcon]

Is AutismMix_confetti base?

Or merge

[vital raptor]

both of those are merges

autismmix_confetti is currently the most popular

the other one is my personal mix

[wheat falcon]

Ah okay, yea I looked at images of confetti where I noticed the scores

But like you said the effect in the test looks minor

[vital raptor]

i noticed for my particular prompting, score_8_up seems to be forcing a specific style on everything, ran this test to find which one was affecting it and if it even mattered

i'm running it again with the base model included for science sake

not that it can usually produce anything decent with such a simple prompt, being a base model it needs a lot more help from the prompt

prompt is:

BREAK
source_anime, 1girl, child with short black hair and (purple:0.7) eyes, cat ears, wearing a choker, portrait,```

and before anyone mentions "you have to use all of them", here's:

BREAK
source_anime, 1girl, child with short black hair and (purple:0.7) eyes, cat ears, wearing a choker, portrait,```

obviously, the prompt can change the results a lot, along with CFG scale and everything else, but it seems pretty obvious that using all of the tags isn't a great idea unless you want that particular style shift

score_8_up in particular is very heavy compared to the others

[wheat falcon]

True score 8 up is a very noticeable change.
Will test stuff out with confetti later today.

[junior quest]

model?

[junior quest]

do we have good sdxl inpaint?

[native halo]

in foocus is decent

[supple raptor]

A1111 with the Krita extension is my go-to for any inpainting. Tough to beat doing your inpainting in layers inside of actual dedicated image editing software.

[nova remnant]

Stupid wings

It's not Christmas

[supple raptor]

[nova remnant]

Stupid comparison

[junior quest]

oh ma god

full face blush is....

I never thought it can do that much difference

[gloomy scroll]

[vital raptor]

Which part is stupid about it?

I’m happy to do more in depth comparisons, I only started with a simple prompt to see what effect those keywords were having

[woven sparrow]

What is ProtoLarkXL?

[native halo]

a model

[woven sparrow]

who could have guessed

[native halo]

https://tenor.com/view/anime-azumanga-daioh-shocked-chiyo-gif-18944897

[gloomy scroll]

[signal panther]

Yeah, it's great. But some prompts with many '_' or numbers break it... Thx for testing.

@junior quest BTW prompt for those girls?

[junior quest]

hmm I dun remember now

I prompted bout raiden from genshin

[nova remnant]

https://media.discordapp.net/attachments/857955668829011988/961381752805425222/kok.gif

[vital raptor]

It’s a personal mix

Probably one that won’t get released

kinda wish i would have made better notes on what i merged into it before merging my custom face lora so i could reproduce it for release

[native halo]

https://tenor.com/view/free-money-meme-fast-gif-gif-16935040457953192433

[vital raptor]

more like free ban from every server they're on 😄

[native halo]

https://tenor.com/view/gambling-jambling-gamblecore-burning-in-hellfire-screaming-gif-17846281577290593447

[vital raptor]

so, anyone have a good prompt that really shows off why ponyXL is so good?

[junior quest]

scam?

[native halo]

100% real swear on my yeezy's

[junior quest]

whos yeezy?

[native halo]

https://tenor.com/view/yeezy-dancing-giant-shoes-gif-13188626

[ionic goblet]

as real as my 2d gf

[junior quest]

seems like a scam

url

[ionic goblet]

it is lol

[vital raptor]

of course it's a scam

[junior quest]

[native halo]

real like my yeezy's

[junior quest]

@wooden coral

[ionic goblet]

look so good

[gloomy scroll]

Totally not scam when link in the message is tinyurl shown as steamcommunity link

[junior quest]

ahh no problem for me

I didnt log from my pc

[native halo]

i dont even have a steam acc and he still scammed me

[stable nimbus]

what model

[wooden coral]

Dont click those links

[junior quest]

bonk

https://tenor.com/view/yes-si-peachcat-cat-kitten-gif-15951852

[signal panther]

https://tenor.com/view/bonk-doge-gif-24837098

[bitter latch]

trying some things diff 😄

[bitter latch]

falling apart

[bitter latch]

well that came out odd

[junior quest]

deym thats so good

[vagrant falcon]

do you guys know how to run SDXL models on A1111 ?

do i select it in checkpoints like others models?

[vital raptor]

yes.

[bitter latch]

yep but need a xl vae as well

breaking thru

[vagrant falcon]

3 min to generate a 15steps with sdxl model D: