🍥｜anime

[nova remnant]

I think there was a very bad one

and then one that was like 1.5 quality

[rough oxide]

dall-e 2 was... ok

yea

dall-e 2 was like the 1.5 quality one

[nova remnant]

I was hyped about dalle 2 okay

[rough oxide]

hey it was good i think

then sd started making advancements

when dall-e 3 came out i legitimately thought it was one of the best things for anime, but so many prompts got denied Lol

but now we're far past that even

i still think dall-e 3 in concept is really good

cuz your prompt can have like

so many different things, and it'll usually get almost all of it

[nova remnant]

dalle 3 is super fine tuned

so much it can't make bad images lmao

its quite too much

[rough oxide]

hate the like automatic diversity stuff in dall-e 3 tho

[nova remnant]

it also has some weird grainy look to it that looks horrible

in the more HD version its better

from API whatever

[nova remnant]

You don't like randomly getting black asian blonde elves ?!

I find those quite funny

[rough oxide]

like

it's just that

it always like gives them an afro n stuff too

out of nowhere

like it doesn't fit the theme at all

lmao

[nova remnant]

it does break your prompt to include that ye

[unkempt cape]

me too lol it was peak when it didnt censor everything

[rough oxide]

yea like first week

Lol

now i stopped using it entirely

it will just deny EVERYTHING

in the first week u could do pepe the frog even

[unkempt cape]

Bro i made some fire pics like peak highest level quality

was so good when it lasted

[nova remnant]

here I go take screenshots of random video clip frames to use as vibe transfer

[rough oxide]

i still think, for the longest time

it wasn't as censored as ppl claimed

u could have blood, guns

etc

even the occasional accidental nudity

but you could speciify like blood/shotguns/etc

For some reason though

it would always deny anime elves xD

also dragon girls.

[unkempt cape]

we're getting there brother

[rough oxide]

pfff. i just use https://www.pixiv.net/discovery to find random stuff for init img

or danbooru's random button

which... i'm definitely not linking

xD

but it's on the front page of danbooru.

[nova remnant]

Nothing can be better than my music video clips

[unkempt cape]

1girl, by makoto shinkai, Your Name, anime coloring, double buns,purple and black dual-toned hair, twin buns hairstyle, lay on grassland, looking at comets in night sky, amazing quality, intricate details,

depends honestly but i will say, no neta doesnt do poses well

[nova remnant]

[rough oxide]

"lay on grassland"

can be "grass, lying"?

or like "grass, on stomach" "grass, on back"

etc

"lay" doesn't work usually

[unkempt cape]

clearly they were trying to more natural sentence prompt (which admitted doesnt do as well)

[rough oxide]

wat is this in reference to tho

here's a random grass, lying

lol

better qual

but wat is that prompt?

almost looks like something i'd write too tho xD

[rough oxide]

i love mixing nat language and tags

but i know when i need to use tags

[nova remnant]

Me when @lyric zealot bot doesn't work

[rough oxide]

and instead of dual-toned hair should be two-tone

[nova remnant]

"cute and pretty hair with two opposing colored sides"

👍

[rough oxide]

maybe if ur in sd3

xD

[rough oxide]

Ohhh

okay

the tihng is

i tried to do prompts like that, nat lang and all

but they didn't come out well in v2

like i tried it the hour it came out

[unkempt cape]

yeah just dont bother

[rough oxide]

it's like sort of similar to kohaku xl

but i could pull off that prompt in anima just fine

fi i change some of the words

wildcard:character, <wildcard[2-4]:artist>, makoto shinkai, Your Name, anime screencap, double buns, two-tone hair, purple hair, black hair, twin buns hairstyle, lying, grass, looking at comets in night sky, amazing quality, intricate details

still using my wildcard artist tho

xD

2 lazy to write a char desc right now

[unkempt cape]

it doesnt do poses as well, the main selling point is just is just the styles, poses maybe later

[nova remnant]

[unkempt cape]

sword hand six fingers

[nova remnant]

no

probably a microphone

[unkempt cape]

oh fuck its a mic

[rough oxide]

at least the lying, grass part

[unkempt cape]

im blind bro half my screen is purpled ):

[rough oxide]

comes out like every time

xD

so yeaaaa iunno, i dun rly have a problem with poses

or anything

[nova remnant]

cause of death: generation overload

[rough oxide]

with curr model i use

[rough oxide]

lol only like 8 secs per imageee

and i do batches of 4 xD

last one. from this prompt. i think xD

[nova remnant]

youtube started playing dragula

some weird vibe transfers coming

[rough oxide]

ya, last batch wasn't gud

so that's enuff of that prompt now lol

[rough oxide]

actual heat overload. kinda. but luckily...

i live somewhere it's like never more than like 65 degrees.

it's also never less than 65 degrees.

it's just.. always 65 degrees roughly xD

[nova remnant]

65 is not bad

[rough oxide]

so my pc heating up muh room is hot but not too hot

[nova remnant]

90+ is where things start to heat depends on the pc

[rough oxide]

yeeehhh

i'm glad this place has no summer.

and no winter xD

my pc gets to abouttt

78-80c i think

but never over 80c

i probly need to clean out the insides, used to just be like 76

[nova remnant]

imma steal this 1st picture face

[rough oxide]

shur

when u say u use

random video screenshot

i thought of turning hardware accell off on chrome so i could like screencap crunchy/hidive/etc xD

how well does that even work tho cuz

i think it reacts better if you use images that are already art/have good variety of details/etc

[nova remnant]

still looks dumb

[nova remnant]

depends on how much you want that specific face

I'm doing random stuff together so not gonna copy it as much

[rough oxide]

o ur trying to get the face

i usually increase creativity a lot

or rather

decrease the influence the image has xD

like in the 0.2ish ranges sometimes with the vibe transfer

and use multiple

[rough oxide]

that one came out decentlyish

now try to get a result like that w/o vibe transfer xD

[nova remnant]

I'm using 1/0.5 now

[rough oxide]

cuz that's the problem i was having

[nova remnant]

Can you give me the prompt you used for that

[nova remnant]

this girl

[mental marten]

#🍥｜anime message

[nova remnant]

text

[rough oxide]

lol

it's formatted thooo

1girl, jururu, waktaverse, 234 \(1234!\) bun \(bbb81bun\), makoto shinkai, Your Name, anime screencap, double buns, two-tone hair, purple hair, black hair, twin buns hairstyle, lying, grass, looking at comets in night sky, amazing quality, intricate details

i do still have sum nai credits left

might be gud in landscape but idk

[nova remnant]

[rough oxide]

only got one ok one out of the 4

but i know if i used vibe transfer

i could get really good ones

iunno

[nova remnant]

wtf

[rough oxide]

i feel like nai is already surpassed, as a base model at least

but the vibe transfer magic

[nova remnant]

one has only the legs lmao

[rough oxide]

is really good

whatever they do to make vibe transfer work

it works

[nova remnant]

this one is cute

[nova remnant]

Its similar to controlnet stuff

[rough oxide]

yeaaaa

i can't really figure out the controlnet/ipadapter stuff yet

i'm pretty sure swarmui supports ipadater stuff but idk how to use it

and controlnet for me always output like

way too close to the src image?

like i'd expect the pose to be similar n all that but, like, waaaay too close to src image

[nova remnant]

lmao its so blurry because of the video screenshot

[nova remnant]

[rough oxide]

pretty shur

we need a naidv4 now xD

i would've been impressed by that image like half a year ago

[nova remnant]

I mean

Other than the model being dumb

what else do you need from nai x3

[rough oxide]

hold on i'll gen a full body with my model and see if it's any better

i have a crapton of full body prompts for animagine lol

or i guess

that's waist up

i'll do that

[rough oxide]

owo

so here's a batch

more detailed eyes, the hands gen better

particularly that 2nd one

is probly a decentish example

naidv3 was impressive the moment it came out and for like the next 5 months

but now i think we went past it

to be fair i can try this prompt in naiv3, should be exact same syntax xD

even in the bad ones tho

i like the eyes a lot more

like

the eyes r bad in all of these

hands r bad in like 3/4 of them

so nai can do a lot better xD

but personal preference probly plays into it too

i'll do a set with quality tags too to be fair

...not that much better

[rough oxide]

this one is okayyy

but the other 3/4 are all bad

for reference prompt is "1girl, shikimori, kawaii dake ja nai shikimori-san, blue eyes, pink hair, long hair, hair between eyes, school uniform, black sweater, shite shirt, collared shirt, long sleeves, spaghetti strap, red ribbon, grey skirt, pleated skirt, miniskirt, medium breasts, fingernails
,(smile:0.7), (cowboy shot:0.7), standing, staight-on, skinny"

so it doesn't describe bg, so it's fine nai has no bg, anima kinda overfits a bit xD

but every anima output was far better than the nai outputs for this one imo

[nova remnant]

[rough oxide]

anyways, still waiting for a model to understand "caramelldansen"

one day.

cuz right now it just looks like loli god's requiem xD

[nova remnant]

sleepy

[zinc flax]

oo cute

[rough oxide]

tying hairrr i wanna try that with my oc prompt xD

[nova remnant]

I hate that clothing down the shoulder thing

[nova remnant]

what is the tag for it

[rough oxide]

detached sleeves?

[nova remnant]

no

the hoodie falling down

[rough oxide]

if ur tryna neg prompt it, then bare shoulders, detached sleeves?

[nova remnant]

@zinc flax you know it

I know you know it

[rough oxide]

._.

[zinc flax]

I don't know what you're talking about tbh lol

[rough oxide]

hao r my suggestions wrongggg xD

[zinc flax]

What kind of clothing are you talking about

[nova remnant]

It's something smol prompts for sometimes

[nova remnant]

This hoodie falling from the shoulders

[rough oxide]

are u trying to add it to ur prompt?

it's off shoulder

[nova remnant]

I'll put on negative

[rough oxide]

if ur trying to add it

[zinc flax]

It's just open jacket no?

[nova remnant]

[rough oxide]

but if ur trying to neg it, i feel like the other 2 would work moreee? not sure.

lol

it should be off shoulder tho

[nova remnant]

I'll try that

I don't want it at all lmao

[rough oxide]

example of an off-shoulder https://danbooru.donmai.us/posts/7897907?q=off_shoulder+

[nova remnant]

The bare shoulders is fine

Bare everything

Wait no

[rough oxide]

but that's single off shoulder

which is also a tag

this is sfw enuff right xD

https://danbooru.donmai.us/posts/7897465?q=off_shoulder+

this is off shoulder

should be exactly what ur looking for tho

i think?

this is just with +jacket +off shoulder

[nova remnant]

Pout

[rough oxide]

ignore the 6 fingers

xD

[nova remnant]

[rough oxide]

i remember i was so surprised when i saw models were able to do heart hands for the first time

was like "damn this is the future"

xD

[unkempt cape]

uh better hand

[ivory pebble]

Yea it does seem like it improved a bit when it comes to artist styles, however, all of the preview images on the civitai page look very low quality, which is why I said it looks fried. Other people who've used it also said the same thing. Guess we'll have to wait and see what the next version is gonna look like

Most people who use NAI don't really generate using higher resolutions because it costs anlas, same thing with upscaling. That's why it would be so much better if we had it locally cause we'd actually have access to other tools like ADetailer, controlnet or a better upscale to get the best results

[rough oxide]

yeh but even with that said, i think naidv3 is now just

naidv3 is a good base model but we have really good merges that surpass its core

maybe not as accurate with poses & such tho

it would be cool if they released it once they release like v4 or something

like trinart used to do

then people could make finetunes and it'd be amazing

[rough oxide]

been out for a while though is adetailer the same as it's always been? do we just use the basic adetailer model that's existed forever or are there new ones just for anime or something

[vital raptor]

adetailer uses whatever model you’re using for the main image, the only model it has is face/hand detection

[rough oxide]

ahhh, maybe i should get that setup again

[ivory pebble]

Because people are too lazy to come up with new artist mixes, that's one of the most fun parts of NAI. People just usually use other people's prompts for mixes and they just prompt basic stuff. Which is understandable, even I get lazy and just prompt 1girl standing sometimes

[vital raptor]

You can also generate a base low res image on nai and put it through stable diffusion for upscaling and adetailer. If you get the settings right it can make a good image better

[ivory pebble]

I've been using the same ADetailer model since SD1.5, works pretty good on Pony so I assume it would work well with other SDXL models as well

[rough oxide]

i might have absolutely no idea how to set it up now xD

dunno if swarmui has a setting for it

[vital raptor]

Pretty sure it does, but I’m not at my pc to look. If not you can do a custom comfy workflow in it

[ivory pebble]

Haven't really seen anything like that tbh, all of the local models have that same NAIV1 style to it that they just can't seem to get rid of, it just looks boring to me

[vital raptor]

With the artist tags I generally use I never need it

[rough oxide]

ooo yea i just noticed u can easily import ur generate settings to comfyui so it'd be possible to set it up if i figure out where to put it n what not

[unkempt cape]

You know the creators themselves never know what their model can do rite?

Also what do you mean other people say its fried where, no one says that (because no one uses this yet)

Act like NAI preview was any different

[unkempt cape]

like imagine having good previews

i was on their discord and they stayed silent about it 😭

this is no NOVEL AI KILLER but it's signalling one

[ivory pebble]

Probably true, but I'd also like to see some images that actually look high quality for once 😭, it's not bad but it still needs to improve a lot of things

[unkempt cape]

idk i think i would say, when i look at the imgs i post here if it was me a week ago i would mistake it for nai

like maybe it shouldve been darker

but other than that i would mistake fr

[ivory pebble]

Oh yeah it certainly does have that NAI feel to it, probably because it actually has artist tags and it's a model that's trained on just anime images unlike Pony, that also helps the base model style a lot

[unkempt cape]

this one's the most convincing, if someone asked me i would say in a heartbeat nai
just need better poses, some missing artists that are really good (wanke for example) and improve character knowledge, and goodbye NAI

[vital raptor]

quality is subjective, if the model does what the creator wants then the 'quality' is fine, just not what you're looking for

that's why there are so many models

[brittle rampart]

Gold words! I would add that the “type of instrument” does not matter.

[unkempt cape]

um... okay?

[nova remnant]

no, never okay

https://tenor.com/view/rrane-battal-rrane-battal-kızgın-battal-kızgın-gif-25355985

[zinc flax]

So true

[vital raptor]

in general people are too lazy for anything
"omg, i pushed the button and everything is garbage AI sucks, sdxl sucks, only the one I won't pay for is good because other people post good things"

everyone so busy chasing the 'best' model, different software, someone elses workflows and prompts, no one can bother to just take some time and play with it for awhile, I'm glad tiktok brain is not my form of adhd 😄

[nova remnant]

[zinc flax]

My form of adhd is lora mixing until I'm burnt out of it lol

[zinc flax]

I'm just waiting for resonance model at this point

[nova remnant]

Too many words going on, I want cute mouse girls

👍

[ivory pebble]

I can't blame them tho cause artist mixing can be really difficult, changing the weight on a single artist tag can change the style quite a lot, not to mention there's hundreds of different artists to mix so you can never decide which one to use 😭

[zinc flax]

Yeah, some style works a tad bit too hard. I'm content with my prompt fortunately lol

[ivory pebble]

Yup, that's why when I'm like "hmm, maybe I should add a bit of hews or quasarcake here" it just ends up overpowering the other artist tags so I have to play with the weights 😭

[brittle rampart]

But there are also circumstances that limit it. this does not refer to laziness.

[brittle rampart]

although I agree regarding laziness

[ivory pebble]

Now make her smile

[nova remnant]

(sharp teeth, insane, tongue out, smile, crazy:1.6)

👍

Recommended prompt for smile

[ivory pebble]

I mean it is a smile to be fair 😭

[zinc flax]

Gotta love 1.5

[nova remnant]

She's cute see

[ivory pebble]

Why did it make the tongue so big 😭

[nova remnant]

[nova remnant]

What is this pose called

[brittle rampart]

Grrrr 😂

[nova remnant]

[unkempt cape]

God forbids people have taste like no way

"GRR, People don't play with the generic anime mix v4.7. Meanwhile, I am content with my shit so therefore obviously, I AM BETTER THAN EVERYONE."

[ivory pebble]

I think the tag for it is "claw pose", not 100% sure tho

[vital raptor]

paw pose works as well, or just 'claws'

depending on the model

[zinc flax]

Claw pose

[nova remnant]

@zinc flax

[zinc flax]

O what's that artist combination

[ivory pebble]

artist:opossumachine, artist:wamudraws, artist:mizumizuni, artist:healthyman, {{{{artist:sumiyao_(amam)}}}}, artist_dino_(dinoartforame), ciloranko, traditional media, {{painterly}}

[vital raptor]

Obviously you didnt read what i said (or understand it), I didn't say anything about being satisfied with a model.

The whole point is that most people just copy/paste someone elses prompts into a model and expect perfect results with a single click, even when the model doesn't match the style they want. It takes time and effort to get the best results from any model, it doesn't matter which one it is.

If you don't like that model, fine, find a different one, nothing wrong with that. Searching for my posts in this channel will show that I change styles and models pretty frequently along with making my own merges. I even post my test results because using different samplers and step/cfg values also changes things even on the same model.

But, saying models, or an entire model architecture (like sdxl) is bad, is simply wrong and people need to stop blaming the tool for their own inability to make what they want.

[zinc flax]

Istg I see healthyman on every nai gen lol

Including me.

[ivory pebble]

It just gives the gens those soft colors and thin lines, not sure how to explain it

The "traditional media, painterly" also helps a lot

[ivory pebble]

Not sure why discord removed the underscore between dino and dinoartforame but it's supposed to be there btw

And the "artist:" as well

[brittle rampart]

and I'm not late at all... 🤣

[nova remnant]

Biig ears

[unkempt cape]

everyone so busy chasing the 'best' model, different software, someone elses workflows and prompts, no one can bother to just take some time and play with it for awhile, I'm glad tiktok brain is not my form of adhd 😄

in general people are too lazy for anything
"omg, i pushed the button and everything is garbage AI sucks, sdxl sucks, only the one I won't pay for is good because other people post good things"

Everybody knows it doesn't work that way. Different models different results.

It's like the most common sense thing ever

[vital raptor]

except a lot of people don't, they try a model or 2 and then come here to whine about "sdxl sucks" or whatever

or complain that NAI3 is the best but they can't afford it, so everything else sucks

my point is there is no 'best' model, but people are always trying to find it

[unkempt cape]

no way, opinions???? Oh my lord how DARE people have opinions?

Look you can get obnoxious with it, sure, but WHO

No one whines THAT much

No one goes:

"ALL SDXL SUCKS"
they go
"Eh currently its not my thing"

[ivory pebble]

It's funny cause people still don't realize NAI is an SDXL model 😭

[vital raptor]

and pony too

[vital raptor]

that's kind of my point though, there's no way anyone has tried every single option out there to accurately say "not my thing"

it CAN do almost anything, you just have to find the right model, settings, and prompts

[unkempt cape]

Or, they aren't SO into SD to try every model currently exist, try every Lora, try everything...

[vital raptor]

on the rare occasion that the literal thousands of sdxl models out there actually can't do that one specific thing, then it needs to be trained, but realistically that's not likely

and yeah, if you don't have the time or desire to search for it, then that's fine, but don't act like "sdxl can't do it"

[unkempt cape]

Yeah the average joe will go 'hmm, i want to draw giantic man with tires on their face' and go train a model. No they won't, it's not their fault if they won't

[unkempt cape]

no one truly thinks like that evar

[vital raptor]

there's a huge difference between "sdxl can't do it" and "i don't know how" but no one wants to admit they don't know how

[vital raptor]

like 90% of people in this server think that way

[ivory pebble]

People just love to complain, you could give them the most perfect model and they would still say "oh it's bad because it can't do this and that", that's just how it is

[vital raptor]

not the majority of us that are here a lot, but they come through all the time

[brittle rampart]

or NAI - that's cool! who cares!? The main thing is to enjoy the result.

[unkempt cape]

No that's false asf

[ivory pebble]

You'd be surprised, the amount of people I've seen on 4chan that claim SD1.5 is better than SDXL is unreal. There's genuinely people out there that think that way just because they're not able to run SDXL

[unkempt cape]

It's 4chan bro like expect more

[ivory pebble]

It takes then 10x longer to generate an image before figuring out what the model is capable of and they'll claim that the model is garbage

[vital raptor]

i mean.. just saying

[unkempt cape]

I mean... tell me I'm wrong

[vital raptor]

if you want examples of the "no one does this"

i AM telling you you're wrong

[ivory pebble]

You said "no one truly thinks like that ever", clearly there are a lot of people who think that way

[vital raptor]

if NAI can do it, something else can too

[unkempt cape]

First of it's more or less a joke, second of all, I never said 'SDXL' can't do it

This is an example I talk about NAI

[vital raptor]

and if other people are trying to replicate NAI 'style' (whatever the fuck that means) then other people think the same way

[unkempt cape]

And I KNOW NAI is SDXL

[unkempt cape]

Right... not like I am complimenting an SDXL model

No this is negative not a positive comment

[vital raptor]

'NAI' isn't a style, it's a SDXL model trained by a large corporation on a huge dataset

[ivory pebble]

Yes but you also can't say the same about other people like you did earlier, there are still people who think SDXL models are bad 😭

[wooden sail]

Just a simple style transfer aka referencing

[unkempt cape]

Oh now you switch to something else? Know what I was comparing how NAI handle style to n Neta

[ivory pebble]

Cute

[unkempt cape]

sorry maybe I didn't reach far enough into the dark side

[vital raptor]

besides, i wasnt talking about you specifically, but you took offense, so 🤷‍♂️

[ivory pebble]

Huh? 😭

[unkempt cape]

Like I never ventured into 4chan

[unkempt cape]

Yeah I'd like to kick people off their high horse 👍. No one attacks you, don't attack them and go tiktok adhd or whatever

[ivory pebble]

Oh it's not just 4chan, back when Pony first released there were a lot of people in the Civitai comment section having debates how SD1.5 models are better than Pony at anatomy and poses, you just can't really argue with these people cause they just can't run SDXL models

[unkempt cape]

Wow that sucks

[ivory pebble]

What's worse is, they were using multiple loras with their models and saying "Well Pony can't do this out the box so it's bad" 😭

It's the stupidest debate ever

[unkempt cape]

Welp guess you have a point there

[vital raptor]

I'm not trying to be on a high horse, I just like to complain about things like everyone else. If you're going to complain about X being bad or X is better than Y, then my complaint is about them being straight up wrong most of the time

I used to just generate images to prove people wrong "sxdl/pony/whatever can't even do this!" so i post it being done. But they don't care, they blame their hardware or lack of time or whatever

one of the most recent was "pony models can't do wings" so I made this

[unkempt cape]

Sure but next time do try to make your point more clear

[vital raptor]

I always try to make my point clear, there's always someone that doesn't get it no matter what I say. I have a lot of customer service experience, I'm used to it

[steady grail]

pony can't make bossy coquettish milfs lol

[vital raptor]

nice try 😏

[steady grail]

don't blame me 🥳

[ivory pebble]

Does anyone know what the best settings for ADetailer are? I feel like it doesn't really work that well when the character is looking back

[steady grail]

with her face facing the camera?

[steady grail]

not at all... (nailed 😍 )

[ivory pebble]

No, I'm talking about when you have something like "from behind, looking back" in your prompt

[steady grail]

you should try again coff coff

[steady grail]

i don't have a fetish for pantyhose <.<

[vital raptor]

some people like feet 🤷‍♂️

[ivory pebble]

I wonder who those people are

[steady grail]

better don't push too far, we hare almost rule breaking borderline lol

gulp

i had a teacher like this... i blame her!

[vital raptor]

lol

last one, was worth upscaling

[steady grail]

i want to work with her >.<

[vital raptor]

i put an extra 7.4s into that

[steady grail]

2704 is too much for me wihout tiled

i can gen max 1600x1600 <.<

[ivory pebble]

One thing that I have to say about Pony, it is honestly really impressive how good it is with feet. On NAI I basically have to inpaint every single gen that has feet in it

Well, I guess that's AutismMix but still

[vital raptor]

yeah, it's pretty obvious what the source dataset was when you start figuring out what it's good at

it knows all the positions and poses... for certain things

[ivory pebble]

Having both danbooru and e621 datasets in a model really is the best for anatomy

Or one of the booru's, might be gelbooru actually

[vital raptor]

so, for adetailer settings...

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 model 2nd: hand_yolov8n.pt
ADetailer confidence 2nd: 0.3
ADetailer dilate erode 2nd: 4
ADetailer mask blur 2nd: 4
ADetailer denoising strength 2nd: 0.4
ADetailer inpaint only masked 2nd: True
ADetailer inpaint padding 2nd: 32
ADetailer version: 24.5.1```

[ivory pebble]

Thank you so much!

[vital raptor]

i think they're mostly default (in a1111) usually the only things i change are confidence to 0.7 ish if it's detecting non-faces as faces

and the denoise strength, lower for less changes, higher more more (like img2img)

usually 0.2-0.4 work best for me, but it's more of a slight cleanup i'm looking for

haven't used it at all since i switched to swarm, and don't usually need it with the styles/artists i use

[ivory pebble]

I've usually kept the denoise strength around 0.2, maybe having a higher denoise is better for like wide shots or when character is looking back

[vital raptor]

yeah, it's worth playing with it, use the same seed and only change the adetailer settings to compare

I had to turn it down for background faces because it was making them too clear compared to their bodies

[ivory pebble]

Should I use the "ADetailer prompt" box at all or? I usually just leave that empty

[vital raptor]

i leave it empty so it uses the same prompt as the image

once or twice i've done special prompting for the hand model, but i was trying to get a specific gesture

it's usually not worth the effort, inpainting is faster at that point

[steady grail]

i usually add a expression wildcard, to get different faces on every character

[vital raptor]

that's a fun idea

[brittle rampart]

do not quarrel. Get some ice cream!

[brittle rampart]

eh.. I wish my ears were that size! 😁

[brittle rampart]

this is already a baby elephant

[vital raptor]

😄

it's always cuter when their ears are huge

[brittle rampart]

and why doesn't anyone generate baby elephant girls!?

[brittle rampart]

and this is a black fennec

[vital raptor]

elephant ears look really weird without the right skin tone

[brittle rampart]

they look weird without the hobbot! 😂

[brittle rampart]

oh bunny!

[unkempt cape]

Nope your wording was bad

[brittle rampart]

she misbehaved very early in life and was often pulled by the ears!

[nova remnant]

Copyrighted ears by comfy

[brittle rampart]

thinks this is something familiar

[nova remnant]

cosplayer mouse

[nova remnant]

[formal shoal]

i like it

[junior quest]

eh?

[zinc flax]

I think I like this combination

[nova remnant]

I like her combination

[ivory pebble]

Didn't take that long to combine either, I just went to danbooru and searched for artists with the most amount of images. I like the ones who draw the jitome style eyes

Try using the Heavy UC preset with that style, it might look better

[zinc flax]

Yeah I'm using it

[ivory pebble]

Best girl frfr

[nova remnant]

Ugly eyes

[wooden sail]

Made this when experimenting with darker images

[ivory pebble]

You're ugly

[ivory pebble]

I love it! I'll have to make something like that later

[nova remnant]

Less ugly than ugly eyes girl

[ivory pebble]

Nuh uh

[pale oar]

Damn pfp material

[glacial bane]

Hello alex

https://cdn.discordapp.com/emojis/992326967879811152.png?quality=lossless&name=MiyabiKill_seia&size=48

[unkempt cape]

its better at hands than v1 lol

not sure if it goes for feet (or foot wtf i forgor)

[zinc flax]

It's quite unfortunate that the model quality is so dang crusty

[ivory pebble]

"skibidi doom doom geats geats" 😭, do not tell me this is an actual thing in the model

[ivory pebble]

Hi

Apparently there will be a new article on the new PonyV6.9 in 2-3 days so that's exciting

[zinc flax]

Can't wait to see the newest hash

That surely won't mess with the prompts at all

[green temple]

\

[glacial bane]

Can i request something if its okay with you rn

https://cdn.discordapp.com/emojis/1174355453958946816.png?quality=lossless&name=smartpepe&size=48

[ivory pebble]

No! (Yes )

[ivory pebble]

It will be fineee (we're doomed)

[zinc flax]

It hasn't even begun and it's already over

[glacial bane]

Dms🥹

[green temple]

[fluid torrent]

dont mind the filename of the last image, it was not animaPencilXL but Cute Lunar instead

[viscid sapphire]

any lora similar to this style?

[lyric zealot]

@dull jackal

[viscid sapphire]

Does anyone know about a similar style lora? Or the prompt i should use?

I use pony diffusion

[lyric zealot]

@wooden coral are you here

[rough oxide]

i used report to staff on all 3, they might be afk lol

o

[wooden coral]

I am now

[lyric zealot]

scroll up

it's on #🆕｜sd3 as well

[rough oxide]

https://tenor.com/view/月光下的異世界之旅-tsukimichi--moonlit-fantasy-月が導く異世界道中-醜男-gif-5150862452407527927

[lyric zealot]

https://tenor.com/view/ban-banned-sakura-anime-spray-gif-22585378

[wooden coral]

Should be all gone now

[lyric zealot]

[wooden coral]

https://tenor.com/view/cant-unsee-how-to-unsee-blind-homer-simpson-simpsons-gif-17613101

[lyric zealot]

not sure about lora, but lineart and 2d might get close

[wooden coral]

If I were you id remove that

[fluid torrent]

😭 no more yoga ... ?!

[viscid sapphire]

i will try it, thanks

I have entered stable diffusion after months, all the things I had have been deleted except the models, how was it to be able to change the vae and the clip from above?

[lyric zealot]

settings > User Interface > Quicksettings list and add it from there

you wanna add sd_vae and CLIP_stop_at_last_layers

[viscid sapphire]

thanks

[viscid sapphire]

when i put "hand on face" another hand is appearing

how can i avoid this?

I want the hand of the character appearing

not a random hand

[fluid torrent]

what is your prompt?

[nova remnant]

Try own hand on face

And solo

[viscid sapphire]

i will try to add solo

[wooden sail]

hand on own face

[nova remnant]

"own" is usually what you want for many things

[viscid sapphire]

not working

[wooden sail]

also touching own face

[viscid sapphire]

i will try

[rough oxide]

aida are u using a model that recognizes danbooru tags

hand on own face should work just fine

[fluid torrent]

nope, cant make it work. my fractal lora really wants symmetry. just 1 hand is nearly impossible!

[rough oxide]

now that i'm freed up a bit, this is hand on own face

it just works™️

to be fair, maybe 50% of them gen with 2 hands on face

idk if "hands on own face" in neg might help with that or confuse it

o

seems like it maybe worked tho?

sort of but maybe same amount where hand is floating in air now

putting hands on own face in neg might help. there's never 2 hands on face now at least xD

but a lot of floating hands or not touching

[left nova]

if this counts nsfw deleet but it fits her character

[nova remnant]

ugly

always this girl

[fluid torrent]

hey! dont mess with Makima.

[formal shoal]

i have one cool generation of makima but i think its a bit too nsfw

[left nova]

send me dm lol

[fluid torrent]

me too

[formal shoal]

i cant dm the image

discord dont let me

[rough oxide]

it triggers the person needs to enable nsfw dms or the image is completely stopped from being uploaded anywhere?

[formal shoal]

only dm

[rough oxide]

i have my nsfw dms enabled. u could try dming me

lol

[formal shoal]

okey

[formal shoal]

i cant either

[rough oxide]

that means uhhh

it's triggering like the child filter or something like that

[lyric zealot]

[formal shoal]

i dont think that

in other discord server i can

[lyric zealot]

[formal shoal]

i cant try to upload here its only cl3avage

[rough oxide]

what message does the dm give?

[formal shoal]

a bit sugestive

[lyric zealot]

i think it's the default settings of discord to disable dm from non-friend

[rough oxide]

o no i have dms enabled here

and my age filter stuff all off

[formal shoal]

its make sense

[lyric zealot]

try me

totally not because i wanna see that

[rough oxide]

lol

[lyric zealot]

yup

i see it

[rough oxide]

my settings r such that it should've worked thoooo

[lyric zealot]

[formal shoal]

yep it worked

[rough oxide]

;_; that's weird

what message did u get when u tried to send it to me?

[formal shoal]

failed in upload

[rough oxide]

o then it wasn't the age restricted block or anything

...u just failed to upload

xD

[formal shoal]

i dont know

[lyric zealot]

i think might be this setting

[rough oxide]

i have all that

[lyric zealot]

[rough oxide]

i just got a random dm from someone in this server yesterday

[formal shoal]

[rough oxide]

who made like 2 posts (and not even in this section). seems like a spam/scambot

xD

[lyric zealot]

all this?

[rough oxide]

yeh i double checked all that and it's all checked

[lyric zealot]

then i have no idea

[rough oxide]

anyways was just troubleshooting kekw. if they could send 2 u then it isn't discord blocking the image

cuz i kno discord blocks some images from like

ever being uploaded

[lyric zealot]

rare but yeah

mostly cheese pizza

[rough oxide]

yeh

but it mis like

completely seiso image it'd just miscategorize it

cuz of the combination of colors/pixels

or something

[lyric zealot]

i think what discord is doing right now, is they have a database of such stuff, but not as the base picture, but like their bytes, then if the bytes is the same, then = auto banned

not sure if there's any AI auto looking for cheese pizza tho

but well, we should move on from such topic as well

https://cdn.discordapp.com/attachments/1067638175478071307/1250543398591398001/GIF_2024-5-31_0-44-06.gif

[rough oxide]

of courz

lol

[rough oxide]

oooo nice

i still have trouble doing ghostly transparency

but moreso on actually making them phase through things i guess

that looks nice aesthetically tho

[fluid torrent]

thank you for the compliment! 😄

[nova remnant]

tupid gif

[lyric zealot]

https://cdn.discordapp.com/attachments/1067638175478071307/1255606727940902972/1718759370298183.gif

[nova remnant]

discord is censoring anything suggestive if you're not friends with that person

[rough oxide]

is that how it works?

[nova remnant]

even stupidier wtf

[nova remnant]

ye

[lyric zealot]

https://media.discordapp.net/attachments/1226718125337350237/1226718126452899870/pockygame2.gif

[nova remnant]

https://tenor.com/view/delete-erase-not-worth-forget-it-clean-slate-gif-8510937

[nova remnant]

https://tenor.com/view/gun-db-shotgun-gif-21597283

[lyric zealot]

https://cdn.discordapp.com/attachments/133606697519415297/1263946107411496960/evil_neuro-shika1.gif

[nova remnant]

big

[green temple]

Ugly indeed

[nova remnant]

sooma knows

smols only

[nova remnant]

everything looks so great but I am going to have to trash this image cuz the hands will never, ever look right

[rough oxide]

just inpaint

[nova remnant]

been doing that for 2 hours

[rough oxide]

and let it run for a long time

o

[nova remnant]

lol

[rough oxide]

it'll eventually probably get fixed.

but might have to run over night

and then check over everything

...only worth it if you're trying to make like a complete image to post on pixiv or something

i think there's a couple other things with that img that would need fixing to me at least too

like the eyes kindaaa

[nova remnant]

yeah the eyes are easy enough though

[rough oxide]

then like the groin area of both pants

r like visibly ai

other than that seems solid

o and the pocket here

lol i use to put so much effort into making images to post to pixiv

[nova remnant]

that's just photoshop work

[rough oxide]

i'd fix every little thing

ye mix of some hand-drawn stuff and ai for more complex parts

[nova remnant]

I don't think the local version of stable diffusion is going to ever have a hand fix. I believe they want to make the best stuff paid only

bing ai art seems to make hands no problem

but microsoft

[rough oxide]

was also simpler when the AI was a lot worse LOL

but now that it's capable of doing really good looking things... must... fix... everything

dall-e 3 is at least free to use for 1024x1024

but uhhh

stuff's gotten a lot better now

the thing is

dall-e 3 does regional stuff really good, it was made based on the fact that it was going to take super complex prompts and include like 'everything' in the prompt

[nova remnant]

so everything is improving except the one we get to use locally. I kinda figured the best AI stuff would never be allowed into our hands

[rough oxide]

nuuuu

i really feel like we've surpassed naidv3 in local already too

[nova remnant]

Ever since character AI disaster my trust in AI corporations is below 0

they start with the lewd stuff and you're like oaky I get it

next thing you know everything is censored

[rough oxide]

i think at least for now we're back to the best image stuff being local. but ye, specific details like hands

do need like a bit more attention

but the overall aesthetics otherwise surpass the paid things & we can do so much more

[nova remnant]

yes but man the hands... I'm tired of having to only make things with hands hidden

maybe this hand holding concept lora will help

[rough oxide]

if u make hands the focus of the prompt

it'd come out well

well in the top right one i didn't do that

sometimes they just come out

althoughhhh yea

clasped hands

is a little bit harder

definitely confirm, clasped hands always come out a mess xD

i've only gotten it to work like 3 times

...one day

[nova remnant]

damn guess it isn't just me...

[rough oxide]

it's just clasped hands

like doing hand poses like heart hands n all that

usually just works perfectly fine

[nova remnant]

I just wanted her on a bench with a boy in a sunset overlooking the ocean, the AI shot out the pose

I was like wow that's good

[rough oxide]

ye. i unfortunately can't get a single good clasped hand

even if rest of composition is good

out of like 20 now

no good clasped hands xD

[nova remnant]

chibi is best

hands don't need to look right

[rough oxide]

aftr 32 i giv up

xD

even chibis have to be drawn right thooo

i mean they can be messed up too

anyways, as long as they're not clasped or overlapping in that way u can usually get fine poses with hands

it's just that one specific thing

[nova remnant]

a little bit of photoshopping

it still sucks but it isn't THAT bad

I'll just roll with it, wheneve rI post I say it's AI art and I suck

lol

[terse ore]

Is there Lora for war related injuries like limbs by themselves or inside organs coming out and stuff?

[nova remnant]

wtf

you said it first

[zinc flax]

Uh huh

I think you could easily cook it yourself no

Though back then civit got some guro lora but it got deleted iirc

[nova remnant]

spend 15 minutes rerolling this inpaint, wondering why it's never changing. Adjust everything

turns out a seed got inserted

I think seeds get applied if you click on a certain button

From the gen tab

yeah I had to refresh the page and I clicked on the use last prompt

I forgot it inserts seed

[nova remnant]

balloon

[nova remnant]

grabby

[brittle rampart]

interesting... why the round eyewear token also draws glasses.. and much better than the round glasses token

[nova remnant]

one is better trained than the other then

[nova remnant]

actually "round glasses" is not a thing

I searched on danboru and it redirects to "round eyewear"

[lyric zealot]

you are not a thing

[nova remnant]

[lyric zealot]

https://cdn.discordapp.com/attachments/1072697081443131476/1259527311389233162/1720262526856102.gif

[nova remnant]

no

[lyric zealot]

https://cdn.discordapp.com/attachments/1067638175478071307/1265736640530223197/DuoNodders.gif

[brittle rampart]

damn my eyes!

[brittle rampart]

someone is in a very good and playful mood! 😂

[nova remnant]

[brittle rampart]

It turns out that you must always specify the round eyewear token in order to accurately display round glasses?

[nova remnant]

Probably o:

They are very round if you use that tag

[brittle rampart]

exactly. and that's what you need

[brittle rampart]

YES!

[nova remnant]

ye,

the library girl glasses

[brittle rampart]

where is the library or at least a book?!

[vital raptor]

same seed/prompt, first one is the base prompt followed by 'glasses', 'round glasses', 'round eyewear', and 'eyewear'

[vital raptor]

this is with autismmix specifically though

[brittle rampart]

in my opinion glassess and eyewear are the same

[vital raptor]

so are spectacles

[vital raptor]

danbooru tags are (usually) generalized so everything gets tagged with the same thing, eyewear is a better general term since it could include other things

[lyric zealot]

https://media.discordapp.net/attachments/1072697081443131476/1206242040052973598/0D682744-A633-4215-B958-55637CFEE0D3.gif

[nova remnant]

How to hide a body in the woods, book

[lyric zealot]

how to hide my wood that's on my body, book

[zinc flax]

vine boom

[nova remnant]

sleepti is here run

[lyric zealot]

[brittle rampart]

Understood

[vital raptor]

the way the images get tagged is usually "general term > specific term" so as an example, "hat" and "cowboy hat" would be on the same image, and image with a "bowler hat" also has the tag "hat", so "hat" is tagged many more times than anything specific, then when training AI it just sees 100 things as 'hat' so it's very well trained (or confused sometimes) for what a hat is

eyewear is probably similar, round glasses was probably trained just fine, but 'eyewear' was on every glasses picture, so round eyewear ends up being more accurate

[nova remnant]

big book

[nova remnant]

no :3

unplayable

[brittle rampart]

@vital raptor tell me how, with the same seed, adding tokens, you actually get the same image? except for small details

[vital raptor]

depends on the tokens, where they're placed in the prompt, length of prompt, etc

but really, i didnt do anything special, just happened to keep everything mostly the same in that example

usually you get the best results if whatever you change is right at the end of the prompt

and little to no negatives or loras or anything to affect the token weights

also seems to work more often with comfy/swarmui since a1111 (and i think forge) apply weight normalization over the entire prompt, but i used to do this all the time with a1111 as well

back when SD1.5 was new, one of the first things I learned for improving an image was to bulk generate a prompt, then find the seed that was closest to what you wanted and then use that same seed with small changes to the tokens and/or the steps to get slightly different results

most likely if i took the images from that example and changed the steps on some of them i could get the backgrounds to be more similar

[brittle rampart]

simply adding Refiner has already changed the image.

[vital raptor]

that's what it's supposed to do

[brittle rampart]

In what sense?

The refiner is launched at a certain generation step

[vital raptor]

yeah, and it generates more on top

it modifies the results

[brittle rampart]

that's it. that's why it's surprising

crap! I was wrong! this is not the seed!

not the same seed as it was.

[nova remnant]

[brittle rampart]

Topic of the day on the Anime Channel - Round glasses as an attribute of super cuteness and kawaii!!😍

[brittle rampart]

about the Topic of the day - this is of course just a joke.

[desert island]

saw @zinc flax's image and was like 'ipadapter and a prompt tweak' 😆

[desert island]

I scrolled up and now I get it 🤦‍♂️

[brittle rampart]

😁

[nova remnant]

grabby

[brittle rampart]

I understood! She's at a séance!

spiritualism

[desert island]

Could stand to do a little inpainting but I like my new prompt and work flow 🥹

[desert island]

Oooooh I love this! 👍

[brittle rampart]

it turned out quite dynamic

[desert island]

https://tenor.com/view/v-a-sly-racoon-gif-24009134

[nova remnant]

doesn't look safe

Did you post the same image twice

Or is this a find the difference game

[formal shoal]

a mistake

[brittle rampart]

I found a new image of Dantalian

[brittle rampart]

king of mice

This reminds me of a fairy tale - The Nutcracker

[desert island]

Okay I'm done posting for now. This was fun. 😄 cheers!

damn it I lied one last one 😁

[brittle rampart]

How much time did you have to spend on vids?!

incredible

[desert island]

I haven't done videos yet but each still takes about 70 seconds to render. The workflow and the dynamic prompt are static once I find something I like and then it's just seed changes between each iteration.

[brittle rampart]

but I'm talking about the video on your channel. In my opinion, this is quite labor-intensive.

[desert island]

oh most of those I do in about a day... the last one took longer since I migrated to basically all on my home system.

[brittle rampart]

Yes, you apparently have a whole studio at home.

[desert island]

lol nope... just a 4080 super... other than that no real wizardry other then prompt and comfyui workflow stuff. I'm currently building an app to streamline the process even further.

[brittle rampart]

4080... hmm... I don’t have any more questions!

[desert island]

well it's no h100 😄 but it does the trick. I love it though... funny thing is I still feel under-powered. God I love ai.

[dull jackal]

Do I sense... that there was a PING!?

[brittle rampart]

no, no one called for help

[fluid torrent]

Guess I need to click Photos.exe in the VIDEO folders to get started??

[unkempt cape]

what a eva elfie

[fluid torrent]

perhaps not known here, since it's a "real" life character

Hmm, I get "Dangerous File Blocked

The file you attempted to download was determined to be dangerous. For your protection, MediaFire does not enable distribution of dangerous files."

Hint for creator: if you want your shit installed, write a good install manual to go with the files, to prevent users from running into bumps like this.

[unkempt cape]

I'm pretty sure you're supposed to add 0s to make the file unscannable

but idk

[fluid torrent]

LOL, thats pretty much what CrowdStrike had done

[zinc flax]

wait

Ain't no way you actually downloaded it

[unkempt cape]

ew 3d people

[fluid torrent]

I've downloaded the .dll file but it's no use trying to run it here

[zinc flax]

Damn, you got some massive balls to download probably a virus lol

[fluid torrent]

This is just utter trash, image sizes are below 50 KB

[zinc flax]

Taking one for the team

[unkempt cape]

windows 🤮

[fluid torrent]

low life amateur script kiddie work. Even if I wanted to nuke my windows install (which I dont have) , I couldnt make this work. ThunderH, go back and do you homework!

btw, this is her

[zinc flax]

jeez

They probably got hacked

[unkempt cape]

maybe they wanted to be the next edgy hacklord

and downloaded weird command prompt apps

and is probably being watched live by another group of creeps in a discord server, who will prevent them from factory resetting the pc

and their webcam is likely hacked too

[fluid torrent]

blue dresses always work nice with dark hair

wait, her skin needs to be darker

[fluid torrent]

"Care for a dance here in the forest?"

[rough oxide]

i think there was one yesterday when someone was posting very nsfw harry potter images, but that person is banned now i think. (maybe?)

[fluid torrent]

oh yeah i remember that one, unfortunately ...

When the music makes you feel so good

[nova remnant]

Anime

[nova remnant]

[rough oxide]

sum less anniemay ones.

[green temple]

Yoimiya 😍

[rough oxide]

but with elf earzzzz

lol

cuz i was doing like an elf ear prompt

this also came out of that

[nova remnant]

Look like anime characters

Is what I mean