🍥｜anime

[elfin geode]

Please keep in mind #✍🏼｜rules-and-tos when posting in this channel.

Rule #4. Absolutely no NSFW

• Nudity, excessive gore/blood
• Anything that exists to suggest sex including sexualized poses
• No illicit imagery - this includes NSFW content involving or depicting underage characters such as schoolgirls, lolicon, etc.

Sexualized poses, or borderline underage sexualized characters in particular cannot be shared in this server. Heavier actions such as time-outs or bans will be considered on repeated offences.

Rule #7. Do not attempt to bypass Automod

• Trying to utilize workarounds, chopping your links or messages into segments, and or trying to deliberately break the rules in craftier ways will not be tolerated.

Community members are encouraged to report rule-breaking images they may encounter by flagging them with ⚠️ or right click > apps > report to staff.

If you're new here or would like to know where to start, here are some quick links around the server:

#1080946152318443610 - Official & community resources.
#🤝｜tech-support - Community support channel for local/cloud/colab installations.

#👥｜roles - All server roles are available here including archives.
#1010934719455707218 - Contact the mod team directly.

#1073085702927024128 - Forums for themed art galleries.
#1047197565365538826 - Share or browse through community models!

[elfin geode]

Welcome to the new NEW anime channel!

[lyric zealot]

[elfin geode]

it's alive!

[supple raptor]

And now cabbage chibi shall forever be the first non staff message in here.

[elfin geode]

wouldn't have it any other way lmao

[hearty heart]

Hello!

[elfin geode]

Hi!

[finite herald]

[wind girder]

WHERE ARE WE!

[halcyon karma]

on discord

[gloomy scroll]

Isekai

[supple raptor]

Sir, this is a Wendy's...

[mellow otter]

We are in need of celebration animu posts

[true iris]

ghibli lora?

[finite herald]

So good i cri

[halcyon karma]

uraraka

[gloomy scroll]

I'm currently restocking my Mikus since I have probably used all already

[supple raptor]

Now I can bribe @finite herald @elfin geode and @wind girder all at the same time

[halcyon karma]

yooo that one looks good bro

[elfin geode]

do you guys prefer to be called #1072013871730131004 or #🍥｜anime

im partial to the weebaloo

[subtle crown]

#1072013871730131004

[supple raptor]

I like weebaloo, but I'm concerned that new people might be confused and go looking for anime 1

[hearty heart]

anime

[acoustic agate]

[lyric zealot]

I'm fine with either, just happy to have a place of home

[acoustic agate]

Trexdel's Anime Palace

[elfin geode]

will keep the weebaloo for another day as a migration event and then switch to #anime!

[wind girder]

let's rename the previous ones to "anime archive" and "anime achive 2" to disperse any linking confusions ? (in a day, sure ^^)

[upbeat bramble]

Mikasa ¯_(ツ)_/¯

[finite herald]

nubby i am married to Vil pls u musnt do this

[tropic wraith]

FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝｜tech-support for any problems you encounter while installing it locally!

[supple raptor]

Can't stop, wont stop

[acoustic agate]

So let's talk about how when the channels transitioned to forums, server activity took a nosedive, and the anime community is on the cutting-edge of technology, bringing this server back to life by reverting to older ways

[finite herald]

Anyways

[acoustic agate]

Turtle

[finite herald]

tru

[acoustic agate]

No, in fact, false

[finite herald]

No inv to u

[supple raptor]

Turtle dog

[acoustic agate]

[acoustic agate]

https://tenor.com/view/you-wo-joker-joaqu-gif-23924191

[mellow otter]

I mean, I think of it more as it was moved to a forum since the format was thought it was going to work better, you all wanted to go back to a channel, and we listened to what you wanted

Butheythat'sjustme

[finite herald]

outerside spicy today

Get em

💅

[mellow otter]

What If I decided to go with blue today?

[acoustic agate]

Oh... oh my apologies, I didn't mean for you to take me seriously. Oh waow lel
I was trolling

[finite herald]

https://tenor.com/view/king-bluprint-drunk-falling-gif-21429804

[acoustic agate]

[desert arch]

Omg anime

[finite herald]

Badabee daba da

[desert arch]

Letsgooo

[finite herald]

So true

[acoustic agate]

Wait, there are no blue roles 👀

Stable Society Blue?

[gloomy scroll]

Well Miku role was deleted

[elfin geode]

[finite herald]

There used to be...

[acoustic agate]

P h o t o g r a p h e r

Who needs it

Axe the role

[finite herald]

Remember miku role

That was crazy

[acoustic agate]

https://tenor.com/view/falling-drunk-shocked-surprised-gif-15546336

[elfin geode]

top ten saddest anime moments

[finite herald]

sometimes when i close my eyes i can still hear her

[supple raptor]

All ten of em in fact.

[finite herald]

Singing levan polka...

[finite herald]

Ty for my life vil

[acoustic agate]

Yoooo fr no cap touch grass

[finite herald]

Can't. Allergic.

[supple raptor]

Miku when: no Miku role

[mellow otter]

Don't even start

[finite herald]

Brother in arms

[acoustic agate]

👀👀👀👀👀👀

[stone onyx]

Finally..

[supple raptor]

But FR... 2b role when?

[nova remnant]

are tofu people welcome here?

[acoustic agate]

How long must we wait for anime pow game night

Admins are all antisocial butterflies with triggerhappy banhammers

[finite herald]

im all about it

U guys are scared

[mellow otter]

Fruit can ban with three keyboards at once

[finite herald]

Bold words coming from the anime channel

[acoustic agate]

I swore I wouldn't screencap my dms, and I still won't. Respect.

[elfin geode]

lmao literally

[finite herald]

Minecraft fell through smh

[acoustic agate]

9/10 players smh

[finite herald]

~f e a r m e~

[rich tartan]

Sup, how you guys create at such high resolution without bugs? Upscaler ok, but that doesn't give it those details

[lyric zealot]

Hey buddy, are you using hires fix?

[acoustic agate]

Upscaler does give details tho

[proper trellis]

why did anime move again 😦 lol

[rich tartan]

Error: out of vram, sadly i cant with 8gb vram, so i am using upscaler like waifu2x

[gloomy scroll]

Guys no need to fear anymore I have taken care of the situation

[lyric zealot]

because we rich, we buying new house left right center

[supple raptor]

Hires fix, followed by img2img using tiled diffusion extension is my current workflow. If needed I inpaint before tiled diffusion.

[rich tartan]

Those lighting details and so on too?

[lyric zealot]

what about your upscale by amount?

[finite herald]

UNLEASH HER

[rich tartan]

Trying to 2x with 5 hires steps

At 512x812

[rich tartan]

Thanks i'll note that down

[gloomy scroll]

Okay, Here's cool Miku instead

[lyric zealot]

try lowering if running out of vram, then if good enough, you can send that pic to extra, and upscale it

[supple raptor]

are you using xformers? that will help a decent amount with vram usage.

[acoustic agate]

Lighting details could be a matter of prompting - upscaling can definitely help add detail to the entire image though - including lighting.

[rich tartan]

Oh, that's another option, thanks, i'll try definitely out

[rigid coyote]

I've heard of Danbooru tags for anime prompts, but can't seem to find what differs a danbooru tag from a regular one... What are Danbooru prompts?

[rich tartan]

Wasn't that a upscaling method? But def i didn't use that, i used a method with 4x at the end, can't recall the exact name

[rich tartan]

Didn't know that, thank you for the information

[gloomy scroll]

Here's example of Safebooru image tags

[supple raptor]

xformers is an optimization if you're using Automatic1111's WebUI, you enable it by editing your webui-user.bat file and adding --xformers to the COMMANDLINE_ARGS line.

[rigid coyote]

Oh so they're basically tags that relate to anime?

[rich tartan]

Good to know, i'll note that too, gonna try it out tomorrow, thanks for all those new tips

[gloomy scroll]

Here's my current prompt for chibi images I'm doing
(chibi:1.3), hatsune miku, cute aesthetic illustration, stylized digital art, abstract and colorful paint style, happy, (bride:1.4)

@finite herald

[rigid coyote]

That would be a yes then

Ty!

Cool image btw

[supple raptor]

Yeah, danbooru is an image board for anime images that uses a very good tagging system to describe what is in the images. Anime models are pretty much all trained using the same tagging system which makes those tags work well for prompting. Danbooru comes with a very hefty NSFW warning so browse at your own risk. Safebooru is a similar site that uses basically the same tags, but doesn't allow as much lewd content as Danbooru.

[somber oasis]

Ooo nice art Vil

[rigid coyote]

Very thorough explanation, tyvm!

[supple raptor]

Here's a very useful extension that adds autocompletion for danbooru tags (as well as any embeddings and LoRAs that you have) to WebUI. https://github.com/DominikDoom/a1111-sd-webui-tagcomplete.git

[rigid coyote]

Yeah I just installed it

I'm currently loking through this guide: https://huggingface.co/hollowstrawberry/stable-diffusion-guide#promptneg

[gloomy scroll]

@rigid coyote You can also use normal phrases with anime models not everything needs to be like booru tags

[supple raptor]

yep, usually a mix of some booru tags and some natural language prompting gives the best results.

[gloomy scroll]

Now nothing is going on here

[proper trellis]

testing out my lora with different style, think it turned out pretty nice, apart from it the fact it has a bad name lol

[proper trellis]

this is so cute

[proper trellis]

i need the model & prompt for research purposes lol

[gloomy scroll]

@proper trellis This one should have pnginfo and for model https://huggingface.co/Surficialruby/RandomMixes/tree/main

[proper trellis]

thanks 😄

[stone onyx]

👁️

[magic wave]

Ayyy, animoo channel is back, I just realized

[lyric zealot]

[acoustic agate]

Ghibli Miku

[lyric zealot]

[pliant epoch]

We're back

[proper trellis]

hmm maybe my lora didn't turn out as well as i had hoped. owell back to the drawing board

[gloomy scroll]

Clyde got cuteness overload so you get only one image

[solid ember]

bro expmixline has been deleted again

[supple raptor]

They can't delete it from my HDD

[gloomy scroll]

All here

[supple raptor]

BTW for anyone who likes UniPC sampler. It's even better if you modify processing.py to make it use DPM++ 2M Karras for img2img (hires fix) instead of DDIM.

[pliant epoch]

How many steps of unipc do you find good?

[supple raptor]

I like it between 20-30 steps (usually 20). Doesn't work as well for all prompts. but it can look quite nice when it does work right.

[proper trellis]

where were these results for when i needed a cow submission lol

[pliant epoch]

ok ty for the info, I'll give it a test

[supple raptor]

processing.py is in the modules directory BTW if you want to change the img2img sampler. You'll know it worked because it won't say that it's using DDIM in console when it does the hires pass. Not sure if a restart after changing is required or not.

[magic wave]

I still dont understand UniPC :o

[supple raptor]

I'm not sure how it works, but it generally gives a look closer to brush strokes than most of the other samplers. I have no idea why, but it isn't compatible with img2img so it has to use a different sampler for the img2img pass in Hires fix which Auto1111 decided to set to DDIM. That works fine, but I find 2M Karras works better with it.

[magic wave]

Fair fair :o

[merry tendon]

whats the best chad scorer for anime?

[merry tendon]

does precision: 64 exist?

[umbral depot]

ANIME 2 ELECTRIC WEEBALOO

[merry tendon]

why is it named like that?

does precision: 64 exist?

[umbral depot]

the chat?

[merry tendon]

yes

[umbral depot]

no idea this is anim3

anime 3

[robust vector]

Nahuel huapi lake with futuristic city on the shore, pedestrian streets and lots of parks between buildings in patagonia

[hearty heart]

are you argentine?

[tropic wraith]

FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝｜tech-support for any problems you encounter while installing it locally!

[supple raptor]

It's temporary. They're changing the name to just anime tomorrow

[umbral depot]

but it is anime 3

[autumn violet]

I mean it is funny

[merry tendon]

yea

[umbral depot]

petition to have it named anim3 instead

[merry tendon]

why that name tho

[umbral depot]

you are familiar with the electric boogaloo meme right?

[merry tendon]

no

[supple raptor]

https://knowyourmeme.com/memes/electric-boogaloo

[keen hollow]

new channel oooo oooo

[autumn violet]

I guess one was the prequel

[marsh surge]

"tennis shoes"

[desert arch]

Good morning

[umbral depot]

good morning space cowboy

[umbral depot]

[acoustic agate]

The Aestpaiid brand is all the rage these days

[lyric zealot]

all the weebs know about that brand

[desert arch]

Civit's down, can't use the web ui

[acoustic agate]

Can't use webui? Y tho?

[desert arch]

I mean the google colab mb

My gpu doodoo so i use the colab version

[acoustic agate]

So you're drawing by hand? Gross.

You could at least steal someone's art and take the credit for it

[desert arch]

This is from yesterday

[acoustic agate]

Is it ai

[desert arch]

Yeah helltaker artstyle lora

Can't draw so yeah

[supple raptor]

You can just grab models from HF instead.

[desert arch]

Right haven't really been looking at hf recently

[lyric zealot]

never really use collab, but can't you download the model to a drive and load that?

or like, to your google drive or something

[supple raptor]

That's also an option to do once CivitAI is back up. I wouldn't do it for more than 1-2 of your favorite models though since that means eating into your google drive space.

[desert arch]

Yeah i don't know how so i didn't really bother to think about that option until now

Guess i'll just upload my older images

[empty chasm]

How to generate ai image i can't find any channel bruh

[tropic wraith]

FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝｜tech-support for any problems you encounter while installing it locally!

[dull jackal]

I would recommend starting with #1072220168534642768

[sand geode]

based, what lora and model did you use to get bocchi?

[desert arch]

Model's cetus mix, forgot the exact name of the lora but it's the most downloaded one on civit, also ligne claire anime lora

[sand geode]

alright

[sand geode]

did you use a bocchi lora?

[desert arch]

Yeah the most downloaded one

[sand geode]

alr

[desert arch]

Tried to make more of those and got whatever this is

[sand geode]

literally my dreams at night lmao

[desert arch]

Basically just their first song title

[dull jackal]

I'm ascending on pickapic with cat fashion I NEED this as a purse

[lyric zealot]

[dull jackal]

I need a cat shaped purse

[desert arch]

The catgirl is the purse

Just bring her everywhere

[dull jackal]

Well. I guess I can ask someone else to hold all my stuff

This is true

The one on the right even has a fish.

Or fish shape--take your p i c.

[dull jackal]

Also. I love this very much ❤️

[desert arch]

How much would you pay for this

[dull jackal]

Did you just

Are there color options here? ❤️

CAT!!!

[hollow geode]

Why did u kill a cat for a bag

[dull jackal]

I love this interstellar lounge idea

[lyric zealot]

each bag made is one cat dead

[dull jackal]

I can't-

[hollow geode]

Yeah

[lyric zealot]

[dull jackal]

That's an adorable bag

[lyric zealot]

welcome to the community btw

[lyric zealot]

for sure for sure, I am almost certain you can get something similar through amazon or something

[hollow geode]

❤️i just got here

[dull jackal]

I have to actually look

I have a panda bag

[desert arch]

Dear god it's 125$ on amazon

[dull jackal]

cat pillows

[lyric zealot]

well, enjoy yourself! If you are keen in seeing more anime stuff, you can visit here https://discord.com/channels/1002292111942635562/1072013871730131004 , as we just recently migrated here

[dull jackal]

You can make it for less--faux fur can be bought reliably

[dull jackal]

Good to meet you, Vivek! Welcome to the Anime C l u b

[lyric zealot]

https://www.youtube.com/watch?v=D_Bb_6JkC1Q

[hollow geode]

I like 3d design so i got here through a person name(literally forgot)

[lyric zealot]

no idea why, but that reminds me of that song

[lyric zealot]

like rendering?

octane rendering for example

[dull jackal]

You mean like Blender, etc?

[lyric zealot]

https://discord.com/channels/1002292111942635562/1072006938839228416 this might be interesting to you

not a whole lot of messages there, but it's something

[desert arch]

Civit's back online, now we're talking

[hollow geode]

Yeah

[hollow geode]

Yeah

[dull jackal]

For sure! If you are looking for stuff like help/info the Blender plugin, it's in #1054807328509145158

[lyric zealot]

there's a new images tab?

[dull jackal]

I'm actually working on tutorials that I'm in the process of releasing 👍

[desert arch]

Oh damn didn't see that

[lyric zealot]

if i don't know about making art, is it still possible to learn Blender?

[dull jackal]

Ofc!

There are actually a lot of 3D artists who don't do 2D art, btw.

Understanding 2D art just helps you become a better 3D artist.

But some people actually do better off the bat with 3D art--everyone is different--so it might actually be easier for some people to do 3D over 2D!

[lyric zealot]

If i ever start learning blender, I'm gonna look for you for help if i need any.

[desert arch]

Gonna need a better gpu if i want to start using blender

Tried it with my current gpu and the rendering takes way too long 💀

[dull jackal]

W a t c h the tutorials I make

Honestly, though, there's tons of tutorials on Blender on YT

[dull jackal]

Change your render settings

And watch the amount of polys in your scene

Etc

[desert arch]

Ah haven't been that deep yet

[dull jackal]

There's two ways to render--Eevee and Blender

's more realistic, but longer render, Cycles

There's a lot of ways to optimize in Blender, such as putting things on the GPU or CPU

So try lowering the render resolution, etc, or just search on YT about optimization

[desert arch]

Thanks, might give it a try later

[dull jackal]

Np! 👍

[umbral depot]

I can make a cube in blender :D

[dull jackal]

Eevee is faster. But your model could just have a lot of polys--check out decimate for that.

[dull jackal]

And delete it!?

[umbral depot]

cube is best cube

[dull jackal]

Cube needs hugs

[umbral depot]

I did make a compainion cube in blender once

[dull jackal]

I was just thinking about that

See, now you just need to take an anime girl 3d model, have her hold the cube, and then put that into SD

[umbral depot]

I think I'll start with updating to 3.5

from 3.3.1

[dull jackal]

Do i t

[umbral depot]

Done

wait can the Blender SD integration support custom models or is it just SD2.1

[dull jackal]

I haven't checked the recent updates, tbh, but I do know there's a plugin that works in with automatic as well

[umbral depot]

I will have to check into that

[dull jackal]

There's info in #1054807328509145158

So I would look there

[desert arch]

Where is the chair

[acoustic agate]

she's looking at it

I'm sorry I'll step out

[dull jackal]

It's the magician trick where it's invisible, naturally

Her angelic powers are not for us, mere mortals, to understand

[stable crane]

oh we're a real channel again?

[mossy kayak]

https://giphy.com/gifs/nba-lebron-james-cleveland-cavaliers-king-xTiTnpcaFPfJAzaP8A

[lyric zealot]

[tropic wraith]

FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝｜tech-support for any problems you encounter while installing it locally!

[stable crane]

🤔

particularly mech-y

[astral aspen]

good morning SD Heroes! classic maid style at the morning will make the day much better !

just testing vae 's for anime

and black white and face with high contrast (Redhair ,white skin) so a maid is perfect for it XD

above is the wafu diffusion vae with elysium anime modell

[acoustic agate]

My buddy, best of pals, amigo. Friend. I do recommend removing the maid on the right. It for sure counts under rule 4

[astral aspen]

no, i dont think so cause its an adult and no offensive parts are shown, so it has to be ok - only the camera angle is different... compared to the left pic

but if u get mad i will delete it, no problem at all

so now fine?

🙂

[random valley]

Guys... Why is it that people don't label some of their models for nsfw, and then not use the NSFW tag when it's literally showing some porn crap.

[tawny helm]

Because they want us to get fired if we open this up on our phones at work at the wrong moment. 😄

[random valley]

I was scrolling for a cartoon/anime -ish models in civitai when sometimes, I would stumble upon full hardcore porn on the thumbnail simple because some people don't actually label it.

[stable crane]

embrace the nsfw side

[astral aspen]

@dull jackal counts this pic as nsfw and under rule 4 ?

[lyric zealot]

can't you report the pic for mis-labeling?

[astral aspen]

maybe im wrong so i have to ask a mod but compared to other pics its a grandma, only the maid costüme triggers a bit

[random valley]

I would do so.

[lyric zealot]

that pic you linked is fine

[pliant epoch]

this one looks fine, the one you deleted was bad because it was an upskirt angle where the skirt was barely covering the panties,

[astral aspen]

ok so its was just the camera angle

[pliant epoch]

yeah, no nsfw doesn't just mean everything not nude is ok. Different things affect it like angle, pose, intent and focusing in on certain body parts

[astral aspen]

so i have learned this now^^

[pliant epoch]

Np, we're just giving friendly reminders incase it gets you bonked. Now here's a bunny

[untold glacier]

Oh sweet we're back to our own channel!

[cyan burrow]

I love multidiffusion so much and yes back home

[untold glacier]

Haven't used multidiffusion yet. Was learning openpose yesterday. How often do you use it with your gens?

[cyan burrow]

all of these images

I use it for upscaling

when I find an image I like

[cyan burrow]

kind of a before and after there

pre-mechanization

and post

alternate timeline

[stable crane]

bit of 50/50

[cyan burrow]

Cat "I didn't do it..."

[cyan burrow]

typical shounen fanservice

[cyan burrow]

I think I will multidiffusion this to make it look like something real

correct some errors

[cyan burrow]

did you want some examples of what it can do?

[stable crane]

I would

[untold glacier]

Sure. Is it very different from the default hires fix upscaling quality?

[untold glacier]

I gotta say Lewds, you make some of the best stuff I've seen on here.

[stable crane]

thanks! I appreciate that

[cyan burrow]

yeah it can be

I will be upscaling etc this

[untold glacier]

Alrighty.

[cyan burrow]

quality, design etc prompts are important, but qualifiers should be avoided unless you know exactly why you're using it

[untold glacier]

Prompting methods are the same for SD upscale so far then. I follow. 👍🏻

[cyan burrow]

denoise never go above 0.4, use the images own original size in prompt area, use the sampler and steps you used previously and maybe seed (experiment, depends what you want from it), then the rest is set up in multidiffusion

[untold glacier]

Can you use different upscaling methods? Anime6B or Ultrasharp for instance, as opposed to latent?

[supple raptor]

I find it's best to use both together. My workflow is txt2img with Hires fix 512x768 to 1024x1536, then inpaint if needed, and lastly img2img multi diffusion upscaling with tiled VAE to bring it up to 2048x3072.

[supple raptor]

Yes

[untold glacier]

I'm not seeing much difference thus far between Multidiffusion and SD Upscaling. Is there a definable difference in how it goes about the job?

[cyan burrow]

I was having issues with it turning clouds into images

but thats just a matter of prompt

[supple raptor]

That happens due to too high of denoising or steps. If you're using 2S a or SDE then you want denoising below 0.3 and probably no higher than 0.21.

[cyan burrow]

there is a huge difference when the image contains the details to be enhanced

also, it uses way less vram

[cyan burrow]

like waaaaay less

I'll upscale this

[untold glacier]

I haven't seen a large difference in memory usage from normal generating? It takes things tile by tile, which is what it sounds like multidiffusion does. For clarification purposes, I'm talking about the build-in auto1111 script for img2img.

[cyan burrow]

is one example from last night

[stable crane]

cool

[stable crane]

is it a piecewise upscale? like the robo dragon eye

do you make his eye specifically robotic?

[cyan burrow]

well its a robo eye cause the original prompt was mechanical and hidden inside a matrix

[cyan burrow]

HxH bugs there Lotta

[stable crane]

HxH?

lol

[cyan burrow]

Hunter X Hunter

[untold glacier]

That used to mean a different anime...

[supple raptor]

1. Original
2. Tiled Diffusion @ 0.41 denoising
3. SD Upscale @ 0.41 denoising
4. SD Upscale @ 0.28 denoising

[untold glacier]

Wait. I'm thinking DxD.

[untold glacier]

Oooh. Comparison chart. Thanks.

Wow. There's a definite difference between tiled (which I guess is "multidiffusion"?) and SD Upscale. Very nice.

[stable crane]

2 and 4 look pretty similar

[supple raptor]

Froggy style

[stable crane]

2 def has way more fine details

is this a new img2img script?

[untold glacier]

Yeah, way more defined details in 2.

[supple raptor]

It's an extension https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111.git

[untold glacier]

Better eyes too, but that's likely the luck of the xformers' slight differences.

[supple raptor]

Yeah some of it could definitely be due to xformers, but I do feel like eyes have been a bit better on average with Tiled Diffusion than SD Upscale. So it may be a real benefit too. Maybe I'll try a couple comparisons without xformers tomorrow for a controlled test.

[cyan burrow]

Rendering upscale

[untold glacier]

Have you tried the DPM++ 2M Karras Test changes yet? You get even better small-detail generations with the code changes.

[supple raptor]

Not yet, got a link?

[untold glacier]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/8457

[supple raptor]

Cool, I'll probably check that out tomorrow.

[wheat falcon]

did you removed the positive prompt while using sd upscale script ? looks like it has much artifacts

[untold glacier]

Sure thing. The tradeoff is very slightly less contrast in gens, but for my purposes it's great.

[untold glacier]

It was just a straight comparison.

[cyan burrow]

gonna run it aid

[supple raptor]

No change to the prompt on any of them.

[wheat falcon]

yea looks like it. for sd upscale its better to remove the positive prompt so that you dont get artifacts while upscaling and denois can be a bit higher, but it depends. (quality tags can still be used + negative)

[stable crane]

you run SD upscale with an empty positive prompt?

TIL

somehow got really good fishnets

[desert arch]

Yes haipy cer to you too

[untold glacier]

Haipy Cer, everyone.

[sly rune]

ooh, a new anime channel!

[cyan burrow]

multidiffusion is a blessing when you learn

[cyan burrow]

x§ a few results from first multi attempt

original

sometimes I just want to upscale without artifacts and with clean details that are kept and sometimes new ones reveal new details

they dun fucked up with their tongues I think

apparently Peter Pan paid me a visit

[sly rune]

"one does not simply generate waifus"

[cyan burrow]

generate one with the tagline omae wa mou shindeiru

[sly rune]

Mostly random results. Thing on the left is cool though.

[zinc flax]

Generate two anime grils instead

[cyan burrow]

so old anime already completelt abandoned huh? ;D

or is it just locked

[ocean reef]

Locked 😄

[cyan burrow]

"Aniki's ears have always been weird. One of his ears is perfectly aligned to constantly hear his own farts and mistake even the most quiet of cavern winds for an enemt attack"

we're going to war

[mild tree]

We back?

[cyan burrow]

yeah set the old place on fire

[mild tree]

EYYYY

[cyan burrow]

we've got this weirdo though, she tagged along. She's a stray

[mild tree]

MAAAN thats a sick ass character

[mild tree]

[cyan burrow]

working with a ton of images that Im preparing for examples

brb multidiffusing some cityscape

and going to store

[mild tree]

[cyan burrow]

wtf

[umbral depot]

Both :D welcome to the anim3 channel

[cyan burrow]

welcome back you mean

[mild tree]

WHAOAHDBD

What model is that

And what kind of insane lora combination

[shy totem]

Just prompts- as for the model I mentioned it before- it's a merged model that uses AbyssOrangeMix as the base- I don't quite remember which models I used for the merge

No LORAs

[mild tree]

Ah ok

Fucking crazy tho

[shy totem]

I just used this sequence of prompts and that was the result:
girl, cute, cat ears, (black hoodie:1.2), (one eye closed:1.4), (smiling:1.2), (slightly closed eyes:0.6), simple background, pink background,

As for negative:
(low quality, worst quality, bad quality, nsfw:1.4)
That's it-

I'm not particularly doing anything too crazy or mindblowing-

[mild tree]

It sure works fucking well tho

Although, what upscaler are you using?

And i suppose just highres fix

[shy totem]

I'm using Lanczos with 0.6 Denoising Strength

[mild tree]

Hmm you have an anime specific upscaler? You might wanna try that with about 0.3-0.4 denoise

[shy totem]

Uhh...

[mild tree]

There should be one called 4x anime something

[shy totem]

The only thing I see on the list is R-ESRGAN 4X + ANIME6B

[mild tree]

Yes

Try that one

[shy totem]

Should I try that?

Okay okay

[mild tree]

.35 denoise

And then adjust from there

Run 2 with the same seed but different upscalers so you can see the difference

[shy totem]

Okay-

There

Give me a bit- my GPU isn't that beefy

[mild tree]

Hmm yeah

But go lower on the denoise, that high typically creates alot of artifacts

[shy totem]

AH

Too late- sorry

Hmm... this is with the R-ESRGAN 4X+ Anime6B

With 0.6 Denoising Strength

I'll try again with 0.35

[rich tartan]

Ayo, upscaling really brings details with it

[magic wave]

I usually do 0.5

[shy totem]

For comparison- this is the original without upscaling

[mild tree]

Looks alot more juicy

yeees

Way better lighting

[magic wave]

Also you should use 4x_anime_fatal

[shy totem]

Is that another upscaler?

[mild tree]

Try img2img after at like another 1.5 upscale with the same upscaler

Yes

[magic wave]

Yeah

[shy totem]

I don't think I have that in the list of upscalers

[magic wave]

Its the best one there is

I don't think it is there xD

[shy totem]

This is with 0.35

[mild tree]

eyy you wanna upload it AGAIN to another person and wait for 20 minutes...

[magic wave]

WELP

XD

[mild tree]

Ok around 0.5 should be better

[magic wave]

I'll send the upscaler in DMs @shy totem xD

[shy totem]

Wha-

That'll help a ton- thanks

[magic wave]

The good upscaler :O

[mild tree]

What you working with?

[magic wave]

Yee

[shy totem]

1660Ti

[mild tree]

Yeeaahh

[mild tree]

Whaaa

So i should be able to run sd on my laptop then

Got a 3060

?

6 gb

[shy totem]

Yup-

[mild tree]

EYYY

Poggers

[magic wave]

I mean, if you have anything above like GTX 950 I think you can run it

When I first installed it I had a GTX 960 xD

The bare minimum, and I was able to run it xD

[shy totem]

It takes me 2 whole minutes to generate a 320x640 upscaled to 640x1280 image

[mild tree]

Well it still looks awesome

[shy totem]

With 20 Sampling Steps - DPM++ 2M Karras

[magic wave]

Fair

[mild tree]

I never realized theres isnt such a big hardware boundary

[magic wave]

On my 960 it would take me like 10 minutes to generate a 512x512 into 1024x1024 xD

[shy totem]

For CPU-only, you get an image per hour

[mild tree]

Isnt even that long tbh

[shy totem]

But I had to add Command Line Arguments

I mean- launch arguments was it?

Apart from the --medvram

[shy totem]

Uhh --no-half and

I forgot the other one-

[magic wave]

I mean, in that time I could train a LoRA xD

[magic wave]

--api? :o

[shy totem]

No no

Ah- --precision full

Or else it only makes black images

[magic wave]

Oh, I forgor what that one does :o

[shy totem]

I have no idea- I just read about it in Reddit

[magic wave]

Fair XD

[shy totem]

It helped fix the black images

[magic wave]

Fair fair :o

[proper trellis]

anyone got a nice prompt for me to try

[magic wave]

(gigantic breasts:1.4), (thick thighs:1.4), (curvy:1.4)

[proper trellis]

i mean i could do it but i wouldn't be able to share the results 😂

[magic wave]

Thats the challenge, do it, and make it sfw

[shy totem]

[proper trellis]

lol with that kind of prompting there would be no way to make it sfw

[magic wave]

I mean, I've done it xD

[proper trellis]

share it, show us 😂

[shy totem]

[magic wave]

A bit horni, but I did it

And it was actually at 1.5 weight

[astral aspen]

(gigantic breasts:1.4), (thick thighs:1.4), (curvy:1.4) seperate it, one tag in one part not all together so the curvy makes her f*t

[magic wave]

It works fine in my experience

[astral aspen]

ok 🙂

[shy totem]

Tadah- this is with the new 4X-Fatal Upscaler

[magic wave]

Ayyy

Eyes looking nice and sharp

[shy totem]

Yeah

[magic wave]

Also what was your model again? XD

[shy totem]

AbyssOrangeMix is the base

[ocean reef]

I think I need to change my UI for SD to use nice upscale 😑

[magic wave]

Well, its pretty far from the classic Aom look

[magic wave]

Can it not use hires fix?

[shy totem]

I merged it with something else way back- I... can't remember the name because I renamed it

I can give you the hash instead if that helps?

[ocean reef]

Nope, I use Easy Diffusion UI and they have only 2 upscalers which drives me mad

Can't use any other

[zinc flax]

Tried this
Got some cursed output lmao

[magic wave]

O ooof, and I don't think hash would help much xD

[magic wave]

Ohhh damn, I haven't heard of Easy Diffusion :o

[shy totem]

I tried doing this:

[magic wave]

Though I use A1111 and I've only used A1111, since it also seems to be the most flexible one :o

[shy totem]

[magic wave]

Oh damn that is pretty nice :O

[magic wave]

Oooof XD

[zinc flax]

I got exactly all those 3 lmao

Very proportional body

[ocean reef]

It has really nice interface but they are way behind in updated. Can't use lora there yet

[magic wave]

Oh ooof

Damn then yeah you should

Probably move over xD

[ocean reef]

And I always have problems with nice eyes generation

Yeah will have to

[magic wave]

Yehhh, and I mean, eye generation can be dependent on res and model, I usually use Holokuki and with hires using the 4x_anime_fatal upscaler I never have issues with eyes I'd say, unless I'm using like a overtrained or wonky LoRA

[ocean reef]

I saw that good upscaler can do magic

So yeah, gotta move out 🤣

[magic wave]

Yeah, 4x_anime_fatal is the best one I know of, I don't know where it comes from but I got it from @quick hill, and its amazing xD

[ocean reef]

Probably will take some time to get used to new UI, ugh but working gonna be so much better

[magic wave]

Yee for sure :o

[ocean reef]

And thanks for advices ❤️❤️

[magic wave]

You're welcome

Tell me when you got it set up so I can pass you the good upscaler

[ocean reef]

Okiee, gotta get back home 1st 🤣

[magic wave]

Fair fair xD

[mild tree]

😭

The dealer

[magic wave]

Yes XD

[mild tree]

Just send it here, copy msg link, and send that to ppl

[magic wave]

True, das a good point

I think it was in the anime-1 chat

But RIP anime 1 chat

Anyway, I make elf girl now

[mild tree]

Oh ma gawd... Not again

[magic wave]

XD

I made a new elf :O

Looks pretty similar cause Aroma LoCon scuffs hair tho xD

[mild tree]

Its too early for this

[magic wave]

XD

[mild tree]

And i cant make it myself

Not home

[magic wave]

I will make for you

[magic wave]

[mild tree]

Thanks

[magic wave]

Yeee

[mild tree]

[near elbow]

help, i'm stuck in the mirror!

even worse 🙂

[mild tree]

[magic wave]

Though now I think I see why the LoCon might be plagued with that haistyle, there is two repeats of the first one of these images, and like 3 of the second

[mild tree]

Lol

[shy totem]

Ooh I want to try generating an elf

[mild tree]

So would taking them out help?

[mild tree]

PLEASE

[magic wave]

I went back and cleared 33 pics

Repeats and older pics that didn't match the style

[mild tree]

Ye

[magic wave]

Though now I need to adjust my epoch amount

And go back to old settings

[mild tree]

Is it focused on the face for the style influence

[magic wave]

What you mean? Like what I'm aiming for?

[mild tree]

Ye

I feel like the style is most influential in the face

[magic wave]

Yeah, I mean here is a pic from the dataset, like thats what I aim for, the face, and the kind of soft shading look

[rich tartan]

what are loras?

[magic wave]

Uhhhh

Basically, its like DLCs but for models, you train a LoRA to make portable data that you can smack on a model and teach it how to do things it normally doesn't know how to do

[rich tartan]

Ah, thanks for the information

[magic wave]

Yeee :o

[magic wave]

I feel like I should do another attempt after this one though, with 4 repeats and 32 epochs, cause 5 repeats and 4 batch would mean that there would be 4 of the same pic in one batch, and 1 in the next batch, then 2 on the next, then 3 on the next, then 4, and the repeats, I don't know if that's good

[magic wave]

Cute

[magic wave]

[gloomy scroll]

hatsune miku, (outlines:1.3), cute aesthetic illustration, stylized digital art, abstract and colorful paint style, happy, happy easter,
And A3CAEW model from hf
https://huggingface.co/Surficialruby/RandomMixes/tree/main

[magic wave]

Though about 80 would be good to make a LoRA, so there is still a long way to go XD

[shy totem]

I'll try to generate more- though with my GPU it's very... slow-

[mild tree]

PLEASE

YEEEEEEEEEEES

I NEED THAT

[shy totem]

I'm guessing me making my own LORA is out of the question

[mild tree]

absolute genius

[shy totem]

Since I read it needs a copious amount of VRAM

[mild tree]

or just upload your mix as model to civit

[magic wave]

I mean

I think

It can be made with 8 GB

Though I dunno about 6 GB, I haven't asked :o

[shy totem]

I'm not sure if I'm okay with that- I'd rather just share it with you guys-

[mild tree]

well... i tried...

[magic wave]

Fair, I mean you could upload to drive, and I generate more for you to make a LoRA out of it, well, a LoCon, and that way you'll have that style on any model :o

[shy totem]

That does sound like a very compelling idea-

[mild tree]

hmmm

ima need that model then too

well i just mostly wanna try it with character Loras

since its based on AOM it should give some juicy results

[magic wave]

I mean, your model got a crazy style, but it do seem kinda limited with like body shapes, so that could be a way to circumvent that :o

[mild tree]

hmmm

yeah

[magic wave]

I was gonna give a example with my Sabu LoCon but I forgor I'm training XD

[shy totem]

I honestly don't know that much when it comes to the technical side of SD-

I just learned from a few Google Searches and YouTube

[magic wave]

Well, I can generate, but it will take like 2 years xD

[shy totem]

I didn't even know my version was so outdated if you guys didn't tell me

[magic wave]

I mean, you still made a pretty cool style :o

[magic wave]

Fair XD

[shy totem]

Not trying to downplay myself but I think it was just dumb luck-

[magic wave]

I mean, maybe, still good tho xD

[shy totem]

But I'll consider it for now

But I don't mind you making a LORA

[desert arch]

Good evening

[magic wave]

Fair fair

Though I mean

If you just send pics

That is enough to make a LoRA xD

THOUGH, I know LoRA extractions are a thing

But I am not too sure if that is what I am doing, or if its a different thing, I think its what I'm doing, but it is a term I have heard

Though I just looked in the scripts here and there is a LoCon extract script :o

[mild tree]

hmmm

[shy totem]

Is this similar to like- PNG Info for the Metadata?

[magic wave]

in what way

[shy totem]

The LoCon extraction?

Sorry I'm still a little lost with all those LORA and LoCon stuff

[magic wave]

Well, I am not sure how it works, since I haven't asked about it, but I have seen extractions on Civit, and they are considerably larger than normal LoRAs, I saw some that were like up to 1 GB, so I probably wont be doing those anyway

But it is a thing, and it is there

[shy totem]

Hmmm okay- but I suppose unless I get at least an 8GB card, those are beyond me anyways

[magic wave]

Probably yeah :o

[mild tree]

:)

also

:(

no lora extraction for me

[magic wave]

Ayyy, Aroma V6 finished training, now its time to test it

[shy totem]

Might be another dumb question, but, is it possible to generate two characters without making them look like twins?

[magic wave]

Yes

[shy totem]

HOW

[magic wave]

Latent Couple or Regional Prompter

Let me pull up the links

https://github.com/hako-mikan/sd-webui-regional-prompter

Regional prompter

[shy totem]

I can't thank you guys enough-

[magic wave]

https://github.com/ashen-sensored/stable-diffusion-webui-two-shot

Latent couple

[shy totem]

Oh lord this looks like another whole can of worms to learn-

It might take me a little while to fully grasp the concept

[magic wave]

I haven't used that one yet

I only used the rectangular version

And tested regional prompter a bit xD

[pliant epoch]

here's what you can do with a basic 50/50 split but you can go into more specific region control

[magic wave]

Ayyy, its better now

Less noisy, and bangs style is more malleable

It isn't just this thing all the time anymore it seem

[mild tree]

POGG

[magic wave]

Wait, I might be lying, Cyan tests coming out bad

[magic wave]

I mean it does look better

But the damn

[pale monolith]

whats wrong with hair_between_eyes ?

[magic wave]

I mean, it repeats 7 or 8 times in the dataset, but there is also other bang styles, what the AI wants to give me aneurysm

[magic wave]

Well

This

I am not asking for it

[pale monolith]

hair between eyes in negative?

[magic wave]

I want it to be versatile without being overly complex tho

[mild tree]

[proper trellis]

Clyde strikes again

[mild tree]

[magic wave]

I might make a few more closeups, and do 4 repeats

Perhaps I could also upscale the closeups

[proper trellis]

apparently all these images are explicit lol

nothing different to what ive previously shared

[zinc flax]

It's probably the blood

[magic wave]

Clyde is schizo

[gloomy scroll]

You may want this also so you can use multiple character loras with latent couple
https://github.com/opparco/stable-diffusion-webui-composable-lora

[proper trellis]

it doesn't explain why forest 1 is also explicit 😦

[ocean reef]

Kill me... I've deleted all my models by mistake....

[gloomy scroll]

Psst. check Jae's pin

[mild tree]

not restorable?

tf

[ocean reef]

nope, Shift + del

[mild tree]

NOOOOO

[compact warren]

Whats shift del do

[proper trellis]

ive shared images with even more gore and blood and those images and they've been ok 😂

[compact warren]

Get rid of em forever?