🍥｜anime

[nova remnant]

ahh ok, i think its close to the currrent version im running

[gentle musk]

Hang on, I'm using the Turbo Lora because I cba waiting for gens, I'll turn it off and do a proper one at higher steps.

[nova remnant]

I use unaestheticXL_Sky3.1 negative embedding and a single token in negative "deviantart" for bluepencil lol

[gentle musk]

Yeah still does it without turbo

This isn't a good generation this time, but counterfeit keeps the anime style on the face with the same prompts

[nova remnant]

wildcards are just the best thing ever

same prompt but totally different look to character

running a 5 batch with same prompt on the 'anime-style' of BluePencilXL 2.5 to see if there is any difference.

[gentle musk]

Bluepencil 2.5 works well though without weird ipadapters and the Turbo Lora.

6 Steps

[nova remnant]

dang that's impressive

[gentle musk]

Yeah the turbo Lora is really good

[nova remnant]

oh my 'anime-style' of v2.5 is much more traditional looking

[gentle musk]

And here's with an IPAdapter Image of Mercy lol

[nova remnant]

I much perfer the base 2.5

[gentle musk]

Yeah it looks really clean

[stable crane]

what's the tldr of "what is ipadapter"?

[gentle musk]

You give it an image, it applies elements from that image to the generation

[stable crane]

oh interesting

sorta like a controlnet?

[stable crane]

so this isn't a mercy lora, but a gen of a girl with a pic of mercy as input?

[gentle musk]

Yes

[stable crane]

I see

intriguing

it captured most of the vibes, definitely recognizable

[gentle musk]

Yeah it depends on the prompting and the different IPAdapter models, it can give you vibes of it, or sometimes a bit closer.

[untold glacier]

Had to step away. I recommend something like:
(photo, photorealistic. 3d:1.3). disney. realism
In the negatives to counteract non-anime gens on an anime model. That with something like
(anime artwork:1.2) of or similar to start your positive prompt.

[gentle musk]

@stable crane For example these two are the "same image" but one has the Mercy Image IPAdapter applied to it

So it can change it a lot

Or if I tell it not to start until a 3rd of the way through, it's closer to the original but less mercy like

Or there's one that tries to do "similar" faces

[untold glacier]

When I've used BPN or similar models that have both anime and not in them, this helped me: #🍥｜anime message
It's definitely important to help things along with the prompt however you can with SDXL.

[gentle musk]

Or you can mask part of an image and get a sort of outfit swap

[gentle musk]

I'll give it a try, it likely won't work whislt I'm using the Turbo stuff as the low CFG stops negatives from working properly.

[nova remnant]

for turbo they just ignore negative altogether.

https://huggingface.co/stabilityai/sdxl-turbo "SDXL-Turbo does not make use of guidance_scale or negative_prompt, we disable it with guidance_scale=0.0"

[untold glacier]

Turbo is sounding worse and worse, lol.

[gentle musk]

It does have an effect, just not much of one

[nova remnant]

the cool part of it is that you can gen an image in sub 1 second and you can have your prompt update the images as you type

[gentle musk]

And I'm using a lora which lets the main model influence more, which means that those things do work

[untold glacier]

What does that matter if you can't get what you want though?

[nova remnant]

I can make like 10k attempts in a short time lol

[untold glacier]

I'd rather spend 10x the amount of time generating something if it means that the end resulting quality is great and it follows my prompt correctly.

[nova remnant]

I don't use the turbo model tho. Waiting for a 1024x1024 version down the line

eventually as Turbo matures it will enable the possibility for real time video animation on consumer hardware, that's where it'll get good.

[gentle musk]

@stable crane This is a fun thing I like doing in IPAdapter I generated the first image, used it as a mask to select only the outfit and then ran the generation again with an image of Vegeta

If I don't use the mask and just do the whole image

[untold glacier]

Well, that's my hope for SVD. If we can just have tech advancements or improvements to enable it for <10GB of VRAM.

[nova remnant]

@stable crane https://www.nextdiffusion.ai/tutorials/how-to-use-ip-adapter-models-for-image-prompting-a1111

two links to ip adapter models for 1.5 and xl
https://huggingface.co/h94/IP-Adapter/tree/main/models
https://huggingface.co/h94/IP-Adapter/tree/main/sdxl_models

[stable crane]

interesting ty

[nova remnant]

you see @native halo playing with Miku? I think he does that with IPAdapter, gotta ask him for the steps

[nova remnant]

just a basic idea... you can enable the ip adapter and its correct models and load up a pic and prompt for a portrait of a woman and the output will try to match the image you loaded

@stable crane ip adapter pic and prompt for a portrait of a woman

dont take the wrong way im just giving a basic example of ipadapter

you could ofc do amazing work by tweaking the settings

im gonna try use angelina jolie's face pic on a anime portrait

[nova remnant]

@stable crane @nova remnant hello

what you guys up to today

[nova remnant]

hey

sup?

[nova remnant]

I see strong resemblance of kawaiiPencilXL with v2.5 bluepencil anime, don't think i will be keeping this, the color is too bright and kinda makes the lines a bit blur

downloading blue_pencil-XL-v2.5.0.safetensors now

this is blue pencil anime version

this is bluepencil regular v2.5

using the same prompt

cold

[nova remnant]

what happened cat? weren't you asking for snow 2 weeks ago?

[nova remnant]

It got colder since then

But it's hot again

@gritty spruce just in case this needs attention...

🙂

[silk narwhal]

failed preview image of a video 2 video gen (gen errored after 97 minutes)

[nova remnant]

people are merging celeb faces with other celebs to create new attractions lol

[nova remnant]

Is this 1.5 or SDXL

SDXL dont like the artificial facial look tho

also it bothers me how most SDXL models depict 1girl as underaged girls

totally ruins the incentive to spend time on sdxl

[stable crane]

lol

[stable crane]

yo nm u

[stable crane]

oh so it is a controlnet

wonder how it is different than Reference mode in CN

[nova remnant]

doesn't reference try to reference the entire image composition?

[silk narwhal]

lol..gotta love these generation estimates : 5%|▌ | 1/20 [03:09<59:54, 189.19s/it]

[nova remnant]

i actually delved into two new things today, ipadapter and now reference

[silk narwhal]

wonder how ClipVision can do with that?

[stable crane]

as opposed to? I guess ipadapters are aiming to extract certain info?

[nova remnant]

well i can't give expert opinion on it yet, but here is what i see, with ipadapter you can take certain elements of the image into your prompt

or say create a hybrid entity

dont think reference does any of that

[untold glacier]

Okay I love it. BluePencil 2.5 Anime is my new favorite flat-style SDXL anime model. I've gotta dive into this to see how stylistic it can get. It may be the AnimeIllustDiffusion-type model that I've been looking for.

[nova remnant]

these are your own gens with it?

[untold glacier]

Yep. It's my standard anime testing suite. Nearly all of them are super cohesive and very, very good flat style by default without extra prompting for it.

[nova remnant]

very nice

i wonder how you managed to keep the colors reasonable

the few images i generated the colors were bleeding

[untold glacier]

Hmm. Were you upscaling? It could've been the upscale model, for instance, if you used lollypop.

[nova remnant]

no i wasn't using any refiner and the hi res fix i did was the same model

and i kinda ended up deleting the model and download the 2.5 regular blue pencil

[untold glacier]

Not refiner, upscale model. Was the hires fix latent or an upscale model?

[nova remnant]

its an upscaler lollypop which uses the same mode as ESRGAN

[untold glacier]

Yeah, that's the issue. Don't use lollypop, it'll do that.

[nova remnant]

oh man

which do you suggest i use?

[untold glacier]

Any of these should work. Though my favorites are fatal_anime and ultrasharp.

Fatal: https://openmodeldb.info/models/4x-Fatal-Anime
Ultrasharp: https://openmodeldb.info/models/4x-UltraSharp
Animesharp: https://openmodeldb.info/models/4x-AnimeSharp
Remacri: https://openmodeldb.info/models/4x-Remacri
Hugepaint: https://openmodeldb.info/models/8x-HugePaint
Anime_6B: https://openmodeldb.info/models/4x-realesrgan-x4plus-anime-6b

[nova remnant]

thanks, i'll grab something that can work for both anime and realism

[untold glacier]

Fatal, ultrasharp, and Remacri keep details well. So use one of those then.

[nova remnant]

any suggestion which might do that from that list?

[untold glacier]

Beat ya to it. Heh.

[supple raptor]

fatal slightly darkens images so it would probably be the best choice if the colors are getting blown out normally.

[nova remnant]

i'll get the remacri the detail of it on the page sounds good

[nova remnant]

not normally i was using lollypop

[untold glacier]

To be clear, fatal increases contrast slightly. It'll darken/brighten contrast on images.

[supple raptor]

This is my preferred upscaler for hires fix normally. Fatal is the only other one that I occasionally use. https://openmodeldb.info/models/4x-NMKD-YandereNeo-XL

I'd assume it works fine for realistic images in hires fix, but I don't really make realistic so I don't know for sure.

[untold glacier]

• Ultrasharp may cause very small artifacts with tiny details (a moiré effect or a small grid pattern) with gentle shading/color gradations or other smooth features.
• Anime_6B is fantastic for a smooth, clean finish for 2d, Flat looks to anime stuff. But if you have noise or tiny details you want in the end image, it may remove them.
• Remacri is decent all around, but I've found that a very small amount of the fine details are lost compared to something like Fatal or Ultrasharp.

[untold glacier]

I've used it before, but sadly haven't done too much testing with it. What do you get out of it that you don't with other upscale models?

[supple raptor]

leaves just enough noise to give the second diffusion pass something to bite on for additional details without added noise in the final image. very little to no change in color/brightness/saturation.

[untold glacier]

Nice. I'll give it a more thorough look then.

[supple raptor]

@gritty spruce you got your hammer handy?

[nova remnant]

downloading from openmodeldb is a pain

[untold glacier]

@finite herald Can you snip this spammer for us? #🍥｜anime message

I brought you Miku in trade.

[supple raptor]

It's better than the mess of links to various google drives and megaupload that the old wiki had at least. I preferred the organization of the wiki though.

[nova remnant]

yeah but why cant they host with civitai lol

or huggingface

these cloud servers openmodedb hosts at are always busy

[untold glacier]

Thanks fruit!

[supple raptor]

hf discoverability is ass. and I'm not sure if civitai supports upscalers. they probably wanted a separate site because the upscalers are widely used outside of the AI space though. most of the people who make the upscalers probably don't want people to have to wade through the heaps of porn on civitai to find their stuff

[untold glacier]

"most of the people who make the upscalers probably don't want people to have to wade through the heaps of porn on civitai"

This is true. They really need to fix that.

[nova remnant]

i got this 4x-UltraSharp

going to restart sd

@supple raptor the one you suggested is great for anime specifically but i need something for realism too hence i chose the 4x-UltraSharp

[supple raptor]

I'm not even sure that it can be fixed at this point without outright banning it. It's too ingrained in the culture of the community. Even if they banned NSFW images, at this point people would probably just move on to another site that still allows it.

[untold glacier]

What I wish most is that the filters actually worked. It catches a lot of things, but if I can still scroll down the home page of images and see explicit material, something is terribly wrong.

[nova remnant]

you can't ban or eliminate that fundamental aspect, best thing to do is bring a sense of aesthetics into it instead of having distorted inclinations

[supple raptor]

Yep, the sites a mess on that front.

[merry tendon]

anyone wanna try my new LLM model?

[nova remnant]

the filter works though, i use it often

[untold glacier]

What? I use the filter all the time and it misses so many things. Try filtering everything out and scroll down the main page then tell me that it works properly.

[nova remnant]

i use it when i have to blur out the nsfw content

[supple raptor]

The blur filters do mostly work for me, but like it's still a little awkward to scroll down a page and have 50% of the images blurred out in most situations that it would really be necessary

[nova remnant]

yes that is a bit odd thing, they need to sort it better

[untold glacier]

I'm not talking blurring out everything, I'm talking filtering (removing completely) what is nsfw and part of the list. These are my settings. If I'm still seeing swimwear, panties, and 'sexy attire', then it's not working properly.

[supple raptor]

If it's up on HF, I'll take the link to try it out when I get back around to running LLMs.

[nova remnant]

you have lots of filters there, i use a simple approach to just blur out the nsfw content.

would be hard to filter out everything based on tags and keywords

[merry tendon]

https://huggingface.co/NeverSleep/FlatOrcamaid-13b-v0.2

[nova remnant]

[native halo]

if u use comfy u can use HAT upscalers they are better and newer than old esrgan models

[nova remnant]

@native halo got 20 mins to spare for ipadapter session? 🙂

[nova remnant]

ah but i dont use comfy

[native halo]

i dont use it no more 😔

[nova remnant]

why not? i thought that was a nice feature to mix match content

[native halo]

yea but it works better on comfy and i dont use comfy

[nova remnant]

you can use it on a1111 too

im using it now for both sd1.5 and sdxl

but im new to the whole ipadpater thing which you prolly know more about

[native halo]

yea but its not the same

speaking of the upscaler this is the good one https://openmodeldb.info/models/4x-LexicaHAT

[nova remnant]

i got the 4x ultrasharp

[native halo]

thats esrgan too 😔

[nova remnant]

yes

oh i see this one is HAT

its better than ESRGAN models?

[native halo]

yea but doesnt work on a1111

[nova remnant]

i see, i've already changed the upscaler to ultrasharp i guess ill have to redownload bluepencil anime version

@native halo You didn't reply to her message for 30 seconds

[native halo]

lick

[untold glacier]

Yes? And? That's what I was talking about to begin with. Their filters don't work properly.

[nova remnant]

can't blame you, civitai site lots of script bugs that can get in the way from doing simple things like uploading images and have it generate data

[native halo]

theres a workaround to make it work on a1111 but its not the same because it needs certain python files to make it work properly

[nova remnant]

im eager to learn how you do the things you do with miku

lol not joking, i thought that was the power of IPAdapter

[native halo]

no its just a neco lora

[nova remnant]

sigh lol

[nova remnant]

but so far im getting interesting results with ipadapter, something along the line of loading a portrait image and write a basic prompt to generate interesting pic from the image input

[native halo]

yea its great to add certain objects to an image that u cant get through promptin

[nova remnant]

but im kinda limited to using it for faces only atm

not that im replicating but im getting interesting variations what the model alone can't do

yet keeping resemblance to the image loaded ... the hair the facial tone

[nova remnant]

what you mean

[native halo]

needs more (skin tight,bodysuit:1.5) 🤠

[native halo]

https://tenor.com/view/neco-arc-saul-goodman3d-neco-saul-gif-26298783

[untold glacier]

You are all getting coal in your stockings

[finite herald]

APPLE OF MY EYE

i mean, cool

[nova remnant]

santa needs to go easy with coal, fossil fuel will add to more global warming

[untold glacier]

[native halo]

its ok we love coal

[nova remnant]

how on earth did you do that?

[native halo]

dalle

[nova remnant]

1girl in cute santa costume, bare legs, white fur boots, walking outdoors in the snowy evening, xmas trees decorated with lights

not getting a good pic i wanna share on civitai

[native halo]

speakin of cute

[nova remnant]

fruit is watching

you are gong into the naughty list this xmas

[native halo]

👲

[finite herald]

any last words

[native halo]

sata andagi

https://tenor.com/view/osaka-sata-andagi-azumanga-daioh-anime-funny-gif-1128998042656657107

[nova remnant]

what does this mean tho :p

[finite herald]

https://tenor.com/view/ban-bonk-bic-gif-25226816

[nova remnant]

im burnt out from making images

need sleep, need break 🙂

[finite herald]

sleeb!

[finite herald]

format fail im banning myself

u win

[native halo]

its fried buns of dough

[nova remnant]

oh sounds tasty

i had fun with ipadapter today

very interesting results with it

[brittle rampart]

Guys....Can anyone tell me if it is possible to somehow speed up the work with ControlNet?

[cold talon]

Nice watermelon girl

[steady grail]

good painty style, looks a little like soushiki

[brittle rampart]

I will hope that someone here will like the illustrations I received.

[steady grail]

received?

[brittle rampart]

because in my opinion they are quite far from the classic manga style

[steady grail]

yep, they are more painty style, but they are not generated by you?

[brittle rampart]

well, I mean, i an generate them

[steady grail]

the brush is pretty noticiable

oh...

[brittle rampart]

everything I show here is exclusively mine.

[steady grail]

but that style, adds a vintage effect on the situation

i like it

[brittle rampart]

i like it too!

[steady grail]

what's the model? it's a style i like very much

[brittle rampart]

https://civitai.com/models/228564/helloart-oil from Aji1. but here some LoRa add unique style

[steady grail]

oh, that one, i have it too, i love his touch

[brittle rampart]

agree

[brittle rampart]

In its pure form, this model, in my humble opinion, is not so cute.

[steady grail]

i used a mix of realism and oily panty lora to improve some models

[brittle rampart]

"realism" like a words in prompt?

[steady grail]

no, a lora

this is an example, the model has a oily style, but with a good mix of lora, the result improved in this

[brittle rampart]

oh.. i am not a using any realistic LoRa`s...

[steady grail]

with a right mix, they just sharp some details and enanche some colors

[brittle rampart]

understand

[steady grail]

i'll look for the mix i used

[steady grail]

usually i back up some interesting metadata... usually

cute

very polished and sharp

[wooden coral]

Trying XL for anime now

[brittle rampart]

what this is mean?

[steady grail]

the metadata are the the data used to generate images

i usually same somewhere surprising results

[brittle rampart]

do you mean text files with generation parameters?

[steady grail]

yep, screen shots

text

everything

anything that can satisfy my curiosity

[brittle rampart]

It’s enough for me that A1111 makes it possible to create and save parameters into a text file with the same name as the resulting image itself.

By the way, Comfy also has such nodes

That’s how I remember when I tried to generate using Comfy

[steady grail]

they even leave metadata in generated images

but i often delete stuff

expecially wieh n i manually fix details

[brittle rampart]

for example - from Comfy img

[steady grail]

cute

[brittle rampart]

txt file with parametrs for this is img

some more frome my Comfy

[wooden coral]

Random prompts are fun

[brittle rampart]

just for fun....

[brittle rampart]

halk childhood?

[wooden coral]

Getting the weird stuff now

and stuff I cant post

[nova remnant]

are those wildcards?

[wooden coral]

Its the prompt file that was posted here some time ago together with an auto prompter

Looks like this

[nova remnant]

i see, not familiar with that

[steady grail]

this is without lora

[steady grail]

and this with lora

and is burned down, a little lol

i knew it!

i need to tune the weight, because the new workflow XD

[nova remnant]

i wonder if its possible to generate similar images with ipadapter

[brittle rampart]

a hight CFG?

[steady grail]

no, the weight of the lora, burned the image

reducing cfg will change the result too much, instead, reducing the lora weight, reduces changes to the minimuium

[brittle rampart]

or changing Upscaler or VAE will soften the picture

[steady grail]

anyway i used https://civitai.com/models/204997/digital-painting-style https://civitai.com/models/25694?modelVersionId=143906 https://civitai.com/models/196752?modelVersionId=229782 1 1 0.5

using a latent upscaler

[steady grail]

they just offset the details a little, without changing the image

[brittle rampart]

I decided on this. https://civitai.com/models/58390

[nova remnant]

sd1.5 quality

[steady grail]

aqua?

that nut head can be cool? lol

[brittle rampart]

yes

[brittle rampart]

Well, the task is to clean the lake... and even on a dark night...

[steady grail]

in the anime she was just seeping in the water with a more than dumb expression lol

[brittle rampart]

in the novel same... but

[brittle rampart]

cool "flowers"!

[brittle rampart]

That's why my interrupt generation buttons don't work!?!?!!?!?

[steady grail]

memory full to execute it

[brittle rampart]

really?!

[steady grail]

just press it, when memory is free, it will execute... unless is bugged

[brittle rampart]

a second before Aqua gets the heck out of Darkness

[steady grail]

wow, everything is so coherent. incredible

[wooden coral]

[stable crane]

nice

[dusty siren]

https://www.tiktok.com/t/ZT8Hwo2sn/

[brittle rampart]

I wonder... what will be the conclusion?

[nova remnant]

i will keep using 2m sde exponential for brighter color tone

[steady grail]

interesting, maybe i should try it too

[brittle rampart]

exp... - more contrast for me

[rich tartan]

i got comfyUI now.. but realized i don't know why.. What can i even do with it, except sdxl turbo + live paint?

[steady grail]

generate faster, assemble complex workflows

for example, you can influence some precise steps with lora, controlnet, or ip adapter, and other steps, with different stuff

@brittle rampartplayed with your idea a little, too bad i failed to add people in the back ground

[brittle rampart]

people, this is oddly enough the idea of SD itself

not my

[steady grail]

i suppose, my workflow need some improvement, adding concat, should became easier to add such details

but that's another thing that needs finetuning to get an acceptable result

[rich tartan]

uh, any docs or something? It's like russian to me

[steady grail]

oh

i send you in pvt the comfy discord server

ask there

you will find there even already ready workflow

[rich tartan]

thanks

[brittle rampart]

heh... russian - not traditional chinese... 😂

[steady grail]

i really like this style

[brittle rampart]

totally agree!

[nova remnant]

BluePencilXL v2.9 dropped https://huggingface.co/bluepen5805/blue_pencil-XL/tree/main

as well a blue_pencil-XL-v2.5.0-photo-style which should be interesting to test out

[nova remnant]

BluePencilXL v2.9 + unaestheticXL_Sky3.1

I think the lightning and crispness of lines is better in 2.9

1920x1080 native render

[gentle musk]

"Landscape" from that model

[nova remnant]

2.9?

[gentle musk]

Yeah

[nova remnant]

interesting

lots of missed on 1920x1080 renders but when it hits it comes out real nice.

[nova remnant]

running a 10 batch for 2.5 photo style to see what that is all about.

[gentle musk]

oh I didn't even notice that model in the list

[nova remnant]

just dropped today which is odd cause same as the 2.9

oh photo style is for realism gonna test without my anime modifiers

with anime modifer for 2.5 photo style

this model is very artifacty for me it might be my negative embedding

[gentle musk]

I can have a quick look

Lets see how slow huggingface is today

[nova remnant]

no idea why this model gives so much weird output lol

maybe too low of CFG? I use 30 steps and 6 CFG

okay I took out the negative embedding and used no negative prompt much better output

for an anime/realism look it seems like a decent SDXL model choice.

[gentle musk]

It seems ok, goes for the rounder smoother faces

[nova remnant]

full body shots get real artifacty

[gentle musk]

Looks like a jellyfish got her

[nova remnant]

the color bleed in the hair is so weird

[gentle musk]

And yeah it seems to occasionally go a bit crazy

[nova remnant]

disco knight

wow an actually good render finally.

dem hands tho

[gentle musk]

Hmm the Anime style model does it occasionally as well, just did the same gen but with the anime model

[nova remnant]

v2.9 has been consistenly not artifacty for me

other than the usual hand shenanigans

[rich tartan]

uh, wtf.. i accidentally generated an 71mb file with comfyui... Linux png viewer crashes when opening that thing, had to open in browser

it took 347.91s

[nova remnant]

lol whats the resolution on it?

[rich tartan]

what tha.. 15360 × 10496

[nova remnant]

lol did you run it through a couple 4x upscalers or something?

[rich tartan]

seems like yes

[nova remnant]

that'll do it lol

[rich tartan]

i thought it wouldn't as it didn't give any factor

messed up asf

[nova remnant]

yikes lol

[rich tartan]

the quality is fatal..

[nova remnant]

that's a very high pixel count for a chin

[rich tartan]

dam, comfyui is actually fast upscaling for 3840 × 2624

[rich tartan]

yh

not even the preview loads it

looks fine, just the faces are bad.. I can't manage to fix this

[nova remnant]

what model are you using and what settings?

[rich tartan]

nvm, alr. Sampler issue

[rich tartan]

dam, comfyui can generate better as auto1111 ig..

couldn't imagine generating with 6080 × 3264 upscale at such speed

[nova remnant]

yeah there is alot of advanced stuff you can do with comfy too

[rich tartan]

This is.. Why is it literally sideways?

[rich tartan]

Interesting picture frame

[supple raptor]

[rich tartan]

can you even open that thing without bluescreen?

[supple raptor]

image editors can open it

[rich tartan]

Interesting, my browser lags and my discord was about to crash for a 71mb png file(cancelled upload)

[supple raptor]

Yeah, browsers don't like my monstrosity. Photoshop and Krita can both open it though.

[steady grail]

i would need weeks to generate such an horror

and i'm pretty sure the result wouldn't lack weird mutations

[supple raptor]

It's a prompt matrix of around 2k 512x768 portraits

It's completely useless between being generated with WD1.2 and just being too damn big to reasonably reference anything on it. I only keep it for the memes.

[steady grail]

lol

you could print it a glue it on a wall

[rich tartan]

Generated it with a 4090?

[supple raptor]

RTX 2070. It took around 2 hours.

[rich tartan]

2h? That's good

[supple raptor]

yeah, roughly 3.5-4 seconds per image. IIRC it was 35 steps euler a so it would have been roughly 10 it/s

[rich tartan]

2000 series were the best ones ig, they are still so good nowdays

I do, if i remember correctly 15it/s with my 4070

[supple raptor]

3000 series were the best value if you don't look at the inflated prices that scalping caused.

3060 12gb is the best value

[rich tartan]

Amd and it's support traumatized me, that's why i bought the latest gen..

[nova remnant]

holy.... how is the model?

so the artifacting issue with bluepencil v2.9 is due to sampler?

[nova remnant]

v2.9 .. no artifacts, image looks very crisp and defined

gonna do a visual comparison ... this is bluepencil xl

sd1.5

[dusty siren]

https://www.tiktok.com/t/ZT89NWtN2/

[nova remnant]

why can't sdxl be a little more subtle on the faces

[brittle rampart]

SD looks warmer and in my opinion more stylized and detailed

[nova remnant]

yep, another example of sd1.5 using canny and openpose

[vital raptor]

That's why a lot of people stay with 1.5 instead of using sdxl. Everything has to be trained from scratch so the models will never be the same, and the plugins and everything work a bit different so they'll have different results.

It's hard to find something that matches what you're used to

Not to mention most people don't have the hardware to use sdxl properly, so when you balance quality and speed, 1.5 tends to be the winner

[brittle rampart]

I completely agree.

But as for habit, the question is of course debatable.

I think everyone is trying to get closer to realism. but sometimes such an approximation is quite unnecessary.

[nova remnant]

i kinda aim for hyper realism

[brittle rampart]

More than once I caught myself thinking that we all take millions of photographs and admire and love hand-drawn paintings.

[nova remnant]

soft edge and canny is another good combination

[brittle rampart]

and yet, using the latest versions of controlNet on SD1.5 in A1111 , my generation is very slow.

it's sad

[bitter latch]

Miku Hatsune by TIm Burton

haha

a tweak here , pull here, and.....

[brittle rampart]

👍🏽 now try adding her “scissorhands” from Edward and a hairstyle from the Queen of Hearts

[bitter latch]

thats better

[brittle rampart]

listen to me, pathetic Earthlings! I came to capture your BRAIN!

[bitter latch]

AH HA!

Poster style

[bitter latch]

FINALLY

[brittle rampart]

Well, it’s already like “Hellraiser”

[bitter latch]

that took a turn

[brittle rampart]

Crap! This is already psychedelic!

[bitter latch]

and... more so!

[brittle rampart]

Where Miku!??!

[brittle rampart]

it worked out well. emotionally

[bitter latch]

goes to bed. thud

[brittle rampart]

naked butt in the snow - this is very, very unpleasant!

[brittle rampart]

I have always been tormented by the question... why do they have 4 ears!?! 😂

[vital raptor]

It has something to do with the traditional Japanese stories of youkai/oni (monsters) where there's a difference between a cat turning into a human (having both) versus a cat spirit with a human form (cat ears only), kitsune (fox) spirits usually only have the fox ears, and no human ears.

Artists tended to use one or the other depending on the character's origins, but these days it's just whatever the person likes

[brittle rampart]

Then, contrary to all tradition, Youkai come out quite cute

[vital raptor]

youkai were often cute and harmless, even helpful, there's a lot of different types

they're not all meant to be scary or evil

[brittle rampart]

exactly like the brownies in Eastern Slavic mythology

[vital raptor]

the wiki page for kitsune says this:

In Japanese folklore, kitsune (狐, きつね, IPA: [kʲi̥t͡sɯne̞] ⓘ) are foxes that possess paranormal abilities that increase as they get older and wiser. According to folklore, the kitsune-foxes (or perhaps the "fox spirits") can bewitch people, just like the tanuki[a] they have the ability to shapeshift into human or other forms, and to trick or fool human beings. While some folktales speak of kitsune employing this ability to trick others—as foxes in folklore often do—other stories portray them as faithful guardians, friends, and lovers.

Multiple kitsune temples still exist in japan where they're revered as village protectors and fertility dieties

[brittle rampart]

about it i know

[brittle rampart]

Is it dangerous! cuteness multiplies

[vital raptor]

yeah, it just gets more and more dangerous the more there are

[brittle rampart]

cuteness will take over the world

5 can!?

[vital raptor]

I for one, welcome our cuteness overlords

[brittle rampart]

7 !!?

they not a rabbit!!!! how!?!

[vital raptor]

lol

[brittle rampart]

the back rows were jumping up and down for a group photo!

[vital raptor]

wow, adetailer actually found and cleaned up all 17 faces

[brittle rampart]

interesting if you take the sky-high width of the image

[brittle rampart]

why not?!

[vital raptor]

never tried it with so many, i dont usually use it

[vital raptor]

add in a bit of color 😄

[brittle rampart]

😂

[vital raptor]

ok last one, time for work

[rich tartan]

ComfyUI is actually extremly good..

[wispy canopy]

new model merge int he workings!

[stable crane]

cute lol

[wispy canopy]

new model merge in the works ^^

[stable crane]

ooo

looking forward to it

sdxl to 1.5 pipeline

maybe needs more denoise

[wispy canopy]

you could also try as refiner

convert the latents to the appropriate formate, resume sampling from step (x)

[stable crane]

I guess I'm kinda doing this in a way?

mine is basically img2img with the 1.5 model

there's some charm to it

though maybe it actually needs a downscale in there

[stable crane]

ok making some progress

progression of 3 samples

[nova remnant]

@supple raptor any chance you might release an anime model that has bit mature face and not all sugary cute?

thats one thing im still searching for all over civitai

[supple raptor]

Maybe some day. The goal of my merges is basically always just to make what I feel like using at the time when I find a gap in the available models or something I feel that I could improve on. I make them for myself and then share them in case other people also want to use them.

[nova remnant]

yes thanks for sharing those 2 models you have up, however from all the anime models i have used very few reflect on facial tone, what im looking for is sharper eyes and pronounced lips unlike those that are labeled 2.5D or 3D anime

[nova remnant]

for example this has perfect anime look, not 2.5d or 3d, but i would prefer if her face was bit more mature

its difficult for me to express as it is hard to find

[supple raptor]

I think I know what you mean. The flatter art style mostly, closer to a screencap from an actual anime, but with less rounded faces than that one has. That one in particular was from Soushiki using a lora that contains like 600+ artist styles and was using the artist Okota Mikan.

[nova remnant]

yeah flatter images but wth right balance of shades and shadows along the shapes, i think most models that try to bring maturity in the look go beyond anime and into the territory of realistic look, which kinda basically loses the touch of anime look

[brittle rampart]

@supple raptor do not give in to provocations. Your model is unique!

[nova remnant]

don't twist the words like that, its nothing to do with provocation of any sort, those modes he has up are good too and i use them with specific purpose, but what im asking for his skill to fill in what seems to be missing with other anime versions

[unreal ridge]

😐

[brittle rampart]

This is not a distortion of words... it was just a joke on my part. I apologize if it seemed otherwise.

but it’s true that there are so few models on this topic...

[untold glacier]

It's too cold in here

[nova remnant]

lol that looks like a good fireplace

[steady grail]

cute

new model?

[brittle rampart]

i think this is V3

[nova remnant]

yeah v3 is somewhat different from v2, they both have distinct differences which is good

[brittle rampart]

of course, as @Nubby himself answered me, the third version is closer to real anatomy

[nova remnant]

ah thats cool

im looking for bit more maturity in facial tone in anime theme

[brittle rampart]

loli - save the world!

[nova remnant]

but for anatomy the glitches are far less in his models

thats the thing, i dont want loli

those are not applicable to the ideas i want to express

[steady grail]

just use a lora

[brittle rampart]

but on the contrary, I really like using the appearance of younger characters

I like their certain spontaneity. expressing my thoughts through them

[nova remnant]

the way i see it, maturity and immaturity are different traits from young and old

[steady grail]

cool

[brittle rampart]

Damn am I old? or.... maybe i am too young?!

[nova remnant]

dunno

age is different from maturity and immaturity

[brittle rampart]

rather just the perception of things

[nova remnant]

thats the reality but yes it depends on having the perception to see it

[steady grail]

adults have more pronounced details, meanwhile in young characters, they are more curvilinear

[nova remnant]

damn duality... the mind and the body, makes it hard to reflect

[brittle rampart]

What kind of pronounced adult details are we talking about? 😁

[steady grail]

a little more pointy chin and cheekbones

more pronounced eye sockets and nose

[nova remnant]

in other words the symmetry between mind & body sexiness

[brittle rampart]

and I was thinking about women's arguments 😂

[steady grail]

eheheh

[brittle rampart]

joke.... jusy joke

[steady grail]

i'm more of a physiognomist

i like to study it

[steady grail]

expecially Da Vinci's stuff

[brittle rampart]

oh!!! classical art this is COOL! i really like it!

[steady grail]

i dig classic art like a crazy man

[brittle rampart]

I'm very sorry that I can't visit the world's museums. especially the Hermitage in St. Petersburg

what to say about the Louvre

[steady grail]

me too, i just visited Rome, Milan, Naples historical sites.

[brittle rampart]

I kindly envy you

[steady grail]

never had the chance to visit France, Greece, or Spain

[brittle rampart]

me too

[steady grail]

i can say that experiencing historic sites live, storngly influenced my tastes

the experience is shocking, to say little

[brittle rampart]

totally agree!

[steady grail]

i was blown by the majesty and disconnection between present and past time, that i could feel from such unique and irreplaceable treasures

humanity is so lucky to have preserved so well such pieces of the past

[brittle rampart]

in my humble opinion, it is by studying such pieces that one can learn such things as composition, perspective, lighting. and when you understand this, you yourself try to pay attention to it in your generations.

[steady grail]

yeah!

[brittle rampart]

you have a completely different approach to the very idea of expressing what you want

[steady grail]

in the past, they knew what true beauty was. they had the time and true care to invest in it!

[brittle rampart]

invest for posterity?

Or maybe it’s just the fashion of that time?

but still I’m 100% sure everyone needs to study their work at least a little

[steady grail]

well, they didn't have internet at that time. they had to invest their free time(or life) on something lol

[brittle rampart]

of course, if the purpose of generating images is not just their quantity and not the ability to show skill from a technical point of view

I don’t know how to draw at all and all my attempts were simply failures.... but one person suggested this path to me and I am very grateful to him

[steady grail]

eheheh

i learned the base of drawing by retrace and teorical study, with time, i lost my hand

now i'm more invested in music, but the love for any form of art is still there

[brittle rampart]

understand....

for me - music and musicians are fantastic people!

[steady grail]

i feel the same admiration for sculptors and chisellers

[brittle rampart]

simple 7 notes and such variety!

[steady grail]

i couldn't absolutely grasp how people can sculpt a 3d form from a block

i tried and is insane, my brain simply isn't set to do it

i suppose, some people can, and some other can't

[brittle rampart]

ah.. for me it’s quite simple... but to combine 7 notes into something that touches you to the core is mysticism!

[steady grail]

in theory, i can't see some symmetry in how a music is composed, but when i play, i just don't bother with theory

i try and if it clicks, i simply know it, when it doesn't, i try again XD

[brittle rampart]

I think this approach is acceptable for any activity

[steady grail]

yep uahahah

this is the nature of jazz and blues lol

[brittle rampart]

but music - magic!

same manga and anime!

[steady grail]

well nippo culture is pretty introspective, when well done

[brittle rampart]

It’s so brilliant to convey emotions in fairly simple forms!

[steady grail]

even if nowadays, it's all about boobs and moe

[brittle rampart]

Low bow to all Japanese who stand at the origins of manga culture!

[steady grail]

the last recent anime i truly loved, was tokyo goul

i wonder what's the difference between the various concat of comfyui

[brittle rampart]

you know, I don’t really highlight anything... there are, of course, things that really touch me and they are surprisingly far from serious works, according to critics. I always evaluate a novel, or manga, or anime for myself by how much the work “hooked” me

and not only the essence. as well as a picture, and most importantly the transfer of emotions.

but I’ll say this for sure: these are the works of Mr. Miyazaki and his studio... these are truly masterpieces and classics of Japanese anime culture!

[steady grail]

i agree

the today's stories, to me, feels too much cliché

[brittle rampart]

We scared all the people with our communication...😁

SOORRYY PEOPLE!!!

[steady grail]

eheheh

[unreal ridge]

It's not like this channel has ever been alive 😭

[brittle rampart]

Well, really... I apologized!

peace, friendship, chewing gum!!!! 😂

[untold glacier]

Nah, if anything chatting is good to see in the channel. And those who wish to can join in if they want.

[brittle rampart]

we were chatting about art as such. and about Japanese anime in particular

[untold glacier]

Yeah, I've been reading it here and there.

[brittle rampart]

and what nodes are between?

[stable crane]

it's a sdxl starting gen -> img2img-esque process to sd1.5

[brittle rampart]

👍🏽

[stable crane]

results are getting a bit interesting

[brittle rampart]

THIS!?

ah

look like a stylization

[stable crane]

a bit

[untold glacier]

Like so many of those ANIME YOU style apps out there, heh.

[brittle rampart]

yeap!

[stable crane]

I think I need some controlnet in there so I can really raise denoising

[brittle rampart]

I tried this trick too

got bored quickly

[stable crane]

in my case I'm using sdxl to try to get the before pic too

I want the sdxl consistency with 1.5 style

[brittle rampart]

doesn't change the essence. In my opinion

[stable crane]

not of the tech, but it does of the final result

[brittle rampart]

what a final ?

[stable crane]

can do different scenes where stock photo does not exist

[brittle rampart]

controlet like a reference img?

[stable crane]

no, I VAE encode and load up 1.5 and resample

[brittle rampart]

interesting..... 👍🏽

[stable crane]

so the idea is it's entirely AI generated, I benefit from SDXL composition, then use 1.5 to get the anime back

[brittle rampart]

I take my words back and take my hat off!

[stable crane]

for cases when you have no reference to actually img2img

but I think I still need controlnet

because I'm losing the finger detail

[brittle rampart]

lucky! I can't use controlnet at all

[stable crane]

need to test it on more concepts

[brittle rampart]

Well, yes... the topic of boobs has not been fully covered...😂 sorry.. but today I’m talking about something I don’t understand ☺️

[stable crane]

though it can't do some proportions

[brittle rampart]

but the idea is quite interesting. It’s so tempting to start and work in Comfу again

[stable crane]

it's nice to be able to compose arbitrary flows

[brittle rampart]

She has a serious argument!

[steady grail]

ok, assembled concat, now i should just test it

[brittle rampart]

now do not want fix hands....

but yuri for me look allways cute

[untold glacier]

This is fantastic. Great flow and excellent quality. Well done.

[brittle rampart]

Yes. it came out quite emotional

[steady grail]

looks like the concat from inspire is bugged 😦

[brittle rampart]

you can do it!

[steady grail]

changed node, too bad, this one have no weight setting 😦

[serene spoke]

My best result no inpainting

[serene spoke]

Suggest me how to improve it

[brittle rampart]

depending on what you want to get in the end? alas, there was no button with the inscription “make it beautiful”, no, and never will be.

[serene spoke]

No but if you have an advice for the prompt i appreciate

Or any other things where i should work

[brittle rampart]

I'm not a wizard. I'm just learning. in my humble opinion, it's just flat. there is not a shadow. no light. no emotions. she is, as it were, not alive!

[stable crane]

contrast is a bit harsh

[steady grail]

try to add (make it beautiful:1.3) in the prompt

[serene spoke]

Ok, thanks i still learn it I'll note it and work on that

[serene spoke]

how can i fix it?

[brittle rampart]

what is this for SD?

[steady grail]

weight works on comfy too

[serene spoke]

Generate only beautiful images 😆

[steady grail]

in my experience, 1.3 is the max, before things starts breaking down or glitch in a weird way

[brittle rampart]

🤪

[brittle rampart]

ummmm

[brittle rampart]

look... I just “stole” your prompt - as is

Do you understand what I was talking about?

[serene spoke]

Yes but i was using AnyLoRA Anime Mix to make images most similar possible to anime

And i used controlnet to forced a specific pose

[brittle rampart]

oh.. in this case, as an artist, that’s exactly how you see it. Why then ask?

[serene spoke]

maybe someone has an advice, or a tip for make it better

[brittle rampart]

this reminds me of something

better - HOW!?

[nova remnant]

trying to do water color effect 🙂

those image you see with midjourney i love that style

[serene spoke]

similar as much i could to character and anime style

[stable crane]

my take on your prompt, only slightly modified

[brittle rampart]

midjourney - the style has rather soft shades and color transitions. also soft shadows

[nova remnant]

sd hasnt achieved that style yet

kinda close but not quite

[brittle rampart]

sd is unique in its own way

[steady grail]

when you fail epically

[serene spoke]

I see many people make images in this style i tried to replicate it but without result

[stable crane]

this one? I feel like this is a popular artist lora

[steady grail]

too much exposed

[brittle rampart]

why?

[serene spoke]

Yes it has even the watermark but it's just for example

[steady grail]

this is not what i intended at all.

[brittle rampart]

of course, if we are talking about the initial....

[steady grail]

bud damn, the effect is epic

[steady grail]

yep

that's the idea i'm experimenting with

[serene spoke]

Sorry i don't understand do you change the prompt? Even if can you share the new one?

[stable crane]

I changed it slightly

[steady grail]

what happens if i upscale this? lol

[stable crane]

mostly I changed lora and used a different model

[brittle rampart]

(dramatic praying pose:1.3)

[brittle rampart]

(victorian nun praying in church)

[steady grail]

starting with explonential and ending with karras, things gets psychedelic

[nova remnant]

anyone want to come up with prompt ideas with basic set of words can try this link www.thomas.io its good for generating those words into detailed prompt

[serene spoke]

Anyway thanks to all of you guys

[stable crane]

practice and it will improve

and inspect other ppl's gens

[serene spoke]

Yes it seems 10 times better then 5 mouth ago

[brittle rampart]

from my firsts

my first attempts at SD pen

[stable crane]

not too bad from first time

[brittle rampart]

maybe.. but...

[steady grail]

another setting

[stable crane]

can you throw some doves in there

[steady grail]

doves?

[stable crane]

the flying bird

[steady grail]

oh

it's an idea

for now, i just try an upscale and than i go to sleep, tomorrow i try with the white doves

[brittle rampart]

now i can do she more and more alive!

[native halo]

my bad i thought we were talking about serbian anime

[nova remnant]

hey steven

i realized, using negative prompts even on sdxl can make a huge difference

[brittle rampart]

Well, I can’t repeat the laugh of Victorica de Blois 😁

[native halo]

(human,girl,arms,face:2.5) in negative

[nova remnant]

i used this whole bit here for that image above Poorly rendered woman in an unflattering outfit, against a poorly executed stormy sky during sunset. The scene lacks detail, with dull colors and mediocre lighting. The artwork is of low resolution, failing to capture the essence of the medieval setting.

[serene spoke]

A question but for expression should i use loras expressions?

[nova remnant]

if i take off the negative prompt that will change the output dramantically

[native halo]

depends,some expressions cant be created without loras

[serene spoke]

I thought that maybe the output can be better

Nagatoro like smile need lora?

[native halo]

me tryin the spiral eyes without lora

[brittle rampart]

well!!! it`s - WOW!

[nova remnant]

so i asked google bard ....

generate a txt2img prompt based on the following keywords:
pretty woman, cyberpunk setting, short casual outfit, evening glow. scenic
and i got this ...
A photorealistic portrait of a stunning woman with sharp cheekbones and piercing eyes, her face illuminated by the neon glow of a cyberpunk cityscape. She's dressed in a chic, casual outfit - ripped denim shorts, a cropped leather jacket, and a holographic tank top that shimmers with every movement. Her hair is windswept and wild, framing her face like a cyberpunk halo. The background is a bustling panorama of towering skyscrapers adorned with flashing advertisements and holographic projections. The sky is a vibrant blend of deep purples and oranges, reflecting the city's vibrant energy. The overall mood is one of gritty coolness and futuristic allure.

that's a pretty big prompt but also sounds very relevant

[brittle rampart]

really...

interesting

[nova remnant]

hold on let me show you what i got

i dont think that prompt played out 100% lol

more of a technical limitation on how the image is generated

https://youtu.be/RoPZSIT0uTo

[brittle rampart]

but it can be simplified by taking keywords and semantic words

[nova remnant]

yes ofc

[brittle rampart]

i am try using "promptgen" extention...

[nova remnant]

im looking for convenience with prompting, it can get exhausting to brain storm all the time especially when you want to just releax late night 🙂

[brittle rampart]

then I just let my imagination run wild...

any thought can be the beginning of interesting results

[nova remnant]

yeah that basic idea has to come from the user, and you can use additional tools to assist along the process

[steady grail]

ok, the upscaling froze

going to sleep

see ya

[nova remnant]

epic

[nova remnant]

[neat jackal]

Absolutely my finding topic

[nova remnant]

👍

[nova remnant]

Lmao her weapon is a bamboo piece

Dangerous

[vital raptor]

junior squad 😄

[steady grail]

wow, sdxl?

[native halo]

1.5

[steady grail]

i like it, what model is this?

[native halo]

its my own mix + freeu + lcm + ogipote style lora + dragon lora + shinobu lora

[steady grail]

wow, it's very cool, it's already online?

[native halo]

no 😔

[steady grail]

oh ok, good job anyway

the bg is a little rough, but the focus is very cool

[native halo]

yea the artifacts are probably because of the 3 loras

[steady grail]

this time we have a fantastic focus on the dragon's face and the girl

maybe cadapter controlnet can help in this

[native halo]

weird dragon 🐲

[steady grail]

that's the magic of ai lol

you added more face details of the dragon to the promp i suppose lol

or increase their weight

[native halo]

no all of them with the same prompt

[steady grail]

some results are really insane

[brittle rampart]

Hi hi... dragon lora you own or not?

[steady grail]

ehy Oyhr

@ФуНя-Нянодесу leave me a curiosity, how i spell your nick? lol

[brittle rampart]

is an untranslatable play on Russian words with a hint of Neko slang...

[steady grail]

eheheh

must be a sound not present in english

[native halo]

real dragons from civitai

[brittle rampart]

based on ''nano desu"

[steady grail]

oh, so i can call you desu XD

[brittle rampart]

you can call me just Elizabeth or Liza

[steady grail]

k Liz

is insane how ai add unpredictable but still so cool details

every gen is a new surprise

[brittle rampart]

agree

[rich tartan]

is that made with sd?

[native halo]

yes

[steady grail]

wow, they improved face adetailer on comfy