🍥｜anime

[gleaming flame]

sorry for spamming, that't probably enough for now. I should probably be doing something not-ai related anyway... xD

[gleaming flame]

ok, enough. btw all these done with one illustrious model

[gleaming flame]

ok, last ones, promise!

[gleaming flame]

...I need some serious help xD

[keen hollow]

I still think you should try z-image.

[gleaming flame]

I still might...

[keen hollow]

https://tenor.com/view/do-it-star-wars-emperor-palpatine-palpatine-gif-799657800635657398

[gleaming flame]

https://tenor.com/view/aboazoz-goat-gif-15274093712873480024

[near elbow]

great stuff, i don't mind if you post more than a few in one go 🙂

[gleaming flame]

https://tenor.com/view/dirigieren-tom-and-jerry-jerry-the-mouse-conduct-gif-15003087

I get the mouse and suite

[gleaming flame]

sisters

[gleaming flame]

it's some kind of plane with some kind of plane, but what is that flame? and what are you looking at?

[wooden sail]

Rotation is possible with verbose prompting and balanced time length

[gleaming flame]

this is wild, especially considering I painted the pose myself and started from there, wasn't sure it would work/look correct, since I suck at painting in general

my prompting attempts at "lifting off the ground", "levitating" etc ended with lifting clothes, so I gave up trying to work with that

[gleaming flame]

ok, time to update invokeAI... knowing my past experiences with this s%@t I won't be albe to run it for a week xD

[gleaming flame]

Well.. it works... but img2img produce slighgtly less satisfying results with same loras/models etc for some reason and z-image doesnt work... shows a bunch of errors which tell me absolutely nothing.

[flint bobcat]

hello, ive downloaded some checkpoint mergers from civitai, do i still need to download the checkpoint?

[gleaming flame]

my first z-image turbo images... not exactly my style, but I gues it works nice for setting the scene

[gleaming flame]

heh

[gleaming flame]

ok ok, here: illustrious -> zit

[keen hollow]

That one z-image too?

[gleaming flame]

yes, it's quite good with ghibli style

[tawny wasp]

Meowscarada - Pokemon - illustrious Apex_Ai https://civitai.com/models/2309357?modelVersionId=2598264

[nova remnant]

@gleaming flame this should interest you to start with comfyui from the scratch, have a look https://youtu.be/HkoRkNLWQzY

[gleaming flame]

thanks, but for now new version of invokeAI works fine I guess

[nova remnant]

dont have to give up on invokeai which i understand cause you have gotten used to it, but the comfyui link is for you to watch leisurely

[gleaming flame]

ok, thanks, maybe I'll check it out

[nova remnant]

sure, watch the series as entertainment

[nova remnant]

looks very good and im curious to try anime stuff on zimage

[gleaming flame]

for now at least for anime I'll stick to illu. it just works, and is few times faster than z-image

[nova remnant]

im trying to figure out how you'd describe the anime styles you are generating, they look better than ghibli, vintage anime, cel-shaded

[gleaming flame]

which one?

[nova remnant]

most of what you posted

[gleaming flame]

that's illustrious styles, many of them...

[nova remnant]

yeah but dont they fit into certain anime style?

[gleaming flame]

actually I don't know

I don't care about names, I just search for styles that work for me

[nova remnant]

ok, im trying to instruct z image to render the style but not sure atm

[gleaming flame]

good luck

[nova remnant]

asking gemini for the anime style with this reference

and this is what i got ... The image style can be described as Polymer Clay Anime or "Claymation" Illustration. This aesthetic bridges the gap between traditional 2D anime character design and the tactile, three-dimensional qualities of physical sculpting.

[gleaming flame]

the "style" was in this example a simple image made with illustrious, then evolved with z-image in "clay-mation" in prompt

original one looked like this. i was just curious if I can make pretty claymation character

[nova remnant]

they both have distinct style, i like them both

[near elbow]

"anime style cel shaded" works well

and of course speech bubble text "Whatever you want if it's not too long" and even gothic font text "goths rule"

[near elbow]

"anime claymation style female tank top smirking " - hmmmm probably not 🙂

"waifu, shiny glossy anime cel shaded style" - a bit too glossy

"waifu, shiny glossy anime style"

etc etc

but a lot of fun to play with these

[nova remnant]

i tried but haven't had much luck generating the same level of anime aesthetics with z-image as you can with SDXL variant

what i was going for are these anime styles, but z-image lacks consistency maintaining these look and feel, originally posted by @keen hollow > #🍥｜anime message

[near elbow]

well a booru tag anime model trained on shady content like illustrious will always outperform a general purpose photo focused modul like z-image 🙂

"anime cel shaded style, resolute female warrior pricess wielding comically large sword, dragon background" for fun with z-image - that's the ugly 90s anime/manga look

just "cel shaded" - i like this one the most

"cel shaded manga style", nice b&w

"cel shaded studio ghibli style" also does the job nicely

so personally, i don't miss anything in z-image... of course it doesn't know the 1001 obscure anime series and all that, but good enough for me

[nova remnant]

well yes those are doable but not the same facial aesthetics like illustrious

or should say not the same facial sensuality

[gleaming flame]

I don't care about series, but I need nice styles to work with, and that I miss in zit

[gleaming flame]

I love that I can mix and match hundreds of styles and get hundreds on new styles this way

[near elbow]

yeah, the currently released distilled model is focused on photography and needs pushing to get other styles... hoping for the base model they announced a while back

[keen hollow]

I think the meta is retained in those images so feel free to take a look!

[near elbow]

i'm saving as JPG and discord kills the EXIF so mine are without, blame them 🙂

[wooden sail]

New season is on AIR

[cyan oak]

/cinematic slow camera movement,
soft lighting changes,
smooth natural animation,
high quality, realistic motion

[bold ether]

not how it works chief.

[nova remnant]

current zit model has limitations with styles, this image is illustrious, its not possible to train zit properly until alibaba releases the base model

[near elbow]

nice but not anime 😉

[near elbow]

i'm getting a bit closer to the glossy look that makes a lot of the illustrious results so "tangible"...
this is "anime style, cute mature catgirl eating burger, highly detailed glossy surfaces sketch" with z-image. not there yet of course, but less flat than the default

[near elbow]

"sketch" yeah

[gleaming flame]

I think I found "my" style for now. that mix between anime and western comic is way to fun and dynamic

I might work a bit more with that kind of style too, because it's intriguing, but for now orange hair and fit bods ftw

[nova remnant]

the distilled model is not good for training, you'd better off training flux2 klein 4b base model that black forest lab released recently.

@near elbow https://docs.comfy.org/tutorials/flux/flux-2-klein

[near elbow]

hm, how does that compare to z-image? i love the speed and text and all

[terse ore]

Mistoon_anime does a cross between those 2 styles

[gleaming flame]

maybe, but there are many styles of both anime and western comic and many ways you can mash them, I found my preferred mix, well almost

[terse ore]

It looks really good, well done

[tawny wasp]

usa37107692 Style illustrious Apex_Ai
https://civitai.com/models/2316824?modelVersionId=2606507

[near elbow]

not sure if that is SFW enough... not complaining personally though

wholesome sketch

[near elbow]

big city lights

[nova remnant]

the publicly available Z image is a distilled version, alibaba hasn't released their base model yet, so not possible to properly train z-image yet. To compare the distilled version of zimage and flux2 klein, i'd prefer zimage for its realistic texture and quality, but if you trained the flux2 klein base model you could bring out much better quality than what their distilled or base model has to offer

[near elbow]

looking very much forward to it then... supposedly it's also quite fast, which is a big bonus. i hate the 1-minute-or-more generation times of flux2 and other big models...

[nova remnant]

yes, all turbo model are pretty fast, takes me less than 1 min on my12gb card.

[z-image]
anime style, digital illustration, 1girl, brunette, orange sundress, slender figure. scenic street view, balcony, evening scene, neon lights.

[hasty flax]

Help

[gleaming flame]

?

[tawny wasp]

JNXED_BLADE28 (vtuber) illustrious Apex_Ai
https://civitai.com/models/2319885?modelVersionId=2609877

[nova remnant]

ZIT

[nova remnant]

most of the times with z-image trying to generate anime with sensuous flavor is a struggle against the model, ZIT is very good at realism, pushing the aesthetic boundaries where other top tier, bigger models fall short.

[near elbow]

still waiting for the base model though

[tawny wasp]

ScorBunny (Pokèmon) illustrious / ApexStorm_Ai
https://civitai.com/models/2324258?modelVersionId=2614642

[tawny wasp]

Thx a lot 🙂

[near elbow]

the best demons have twice the horns

surf lords

[tawny wasp]

Till (Easygoing Territory Defense by the Optimistic Lord) illustrious / ApexStorm_Ai
https://civitai.com/models/2326905?modelVersionId=2617651

[near elbow]

brutal

cute

[bronze wolf]

/anime

Giveaways

[near elbow]

(should be "intro" not "into" but you get the idea 😉 )

[tawny wasp]

Ethereal Style illustrious ApexStorm_Ai
https://civitai.com/models/2329429?modelVersionId=2620376

[gleaming flame]

Vin.

[dire herald]

​🚀 LOOKING FOR A CREATIVE PARTNER (AI ANIME SERIES) 🚀
​The Vision: I am a 15-year-old creator building a high-quality AI Anime Web Series. The goal is to build a solid portfolio and eventually pitch to major platforms like Netflix/Crunchyroll.
​Who I am looking for:
​🎨 AI Artist: Someone who can generate consistent Anime characters (Stable Diffusion/Niji).
​🎬 Video Editor: Someone who can animate images into cinematic scenes.
​🌍 Age: Preferably 15-18 years old (Beginners are welcome!).
​The Deal: * This is a Long-term Partnership (50/50 Profit Share).
​No upfront payment, we grow and earn together as a team.
​Let’s build a masterpiece from scratch!
​Interested? Send me a DM

[near elbow]

hm, not sure if commercial postings are compliant with server rules

[tawny wasp]

117 Core Style illustrious / OtakuStorm_Ai
https://civitai.com/models/2332640?modelVersionId=2623927

[dusky linden]

Anyone messed with the flux klein models for 2d content?

[terse ore]

A bit, yeah. Can make blender scenes and then run it through to make it 2d

Anyone know of a good model to make unusual characters? Like, an anthropomorphic cloud of gas for example?

[near elbow]

"an anthropomorphic cloud of gas in the shap of a human" with z-image

"green diffuse gas haze in the shap of a human"

[terse ore]

Eh, thats more like balls of gas in the shape of a human. I mean more like, the shape of an indistinct cloud or haze, and a face and hands within it. But ill try a few things and see if I can get it

[near elbow]

chaos

[tawny wasp]

Ezo Yakumo (Mato Seihei no Slave) illustrious - OtakuStorm_Ai https://civitai.com/models/2339657?modelVersionId=2631701

[tawny wasp]

Label (Goddess of Victory Nikke) illustrious - OtakuStorm_Ai
https://civitai.com/models/2340823?modelVersionId=2632983

[near elbow]

cuuuuute

[near elbow]

manga version of #🔆｜dailies message

[pine pollen]

Hey, im maintaining a somewhat similar open-source tool (Image MetaHub), focused on local metadata indexing and filtering. If you need a replacement, the repo link is on my profile. I use it to quickly filter/search my 'keepers' by prompt/checkpoint/lora/seed, etc and safely delete the rest. give it a try, might be helpful.

[tawny wasp]

Minazuki Kyoka (A Misanthrope Teaches a Class for Demi-Humans) illustrious - OtakuStorm_Ai
https://civitai.com/models/2348908?modelVersionId=2641999

[near elbow]

i think we get that you need buzz by now 😉

[tawny wasp]

Panamera Karla Kaien

[tawny wasp]

Panamera Karla Kaien (Easygoing Territory Defense by the Optimistic Lord) illustrious - OtakuStorm_Ai
https://civitai.com/models/2350615?modelVersionId=2643785

[near elbow]

wholesome neko instead

[glacial adder]

Marín kitawaga

[near elbow]

tiefling hackers

[keen hollow]

Yep I have seen this tool before! (Tools like this are why I stopped--people came along and made much better ones than me. 😄 )

I never could get the image caching/thumbnailing/display all that optimized.

[near elbow]

it's interesting how the belly button magically pulls in all clothing 🙂

cute

[near elbow]

semireal 🙂

[tawny wasp]

Maki Zenin Post-Shibuya (Jujutsu Kaisen) illustrious - OtakuStorm_Ai
https://civitai.com/models/2354630?modelVersionId=2648247

[terse ore]

I hope z image anime fine tunes can be made soon so illustrious won't be the main go to. Since its pretty old already.

[soft parrot]

Hi

[near elbow]

a bit openhearted but happy

[near elbow]

polka dot power

[keen hollow]

i like polka dots

[keen hollow]

that was the new z-image base... I should have specified. I did not really find it easy to work with.

[near elbow]

yeah, needs really long prompts to shine, in my experience

goth angel instead

[tawny wasp]

Erza Scarlet (Fairy Tail) illustrious - OtakuStorm_Ai
https://civitai.com/models/2365135?modelVersionId=2659921

[near elbow]

contrast

[pale mural]

Andrew WK is still partying hard i see 🤣

[wispy canopy]

don't sleep on anima

[tawny wasp]

Yvonne - Arknights: Endfield (illustrious) OtakuStorm_Ai
https://civitai.com/models/2370229/yvonne-arknights-endfield-illustrious-otakustormai

[near elbow]

any good at all?

[wispy canopy]

u tell me

[near elbow]

interesting style for sure

[wispy canopy]

i did kinda prompt for it, so yeah

[near elbow]

just teasing here

[wispy canopy]

still figuring out what sechedulers/samplers work best tho

[near elbow]

black magic as usual 🙂

might give it a try, how fast is it compared to ZIT and illustrious?

[wispy canopy]

it's slower than sdxl, feels faster than flux

zit turbo is just insanely fast

yeah, this is a very badass style

[near elbow]

"faster than flux" (with flux 2 at least) is like "faster than a glacier" 🙂

but sounds cool, looks like you're playing with it now, could you post an update tomorrow with your discoveries? that would be super interesting

[wispy canopy]

well, initially i thought it was "meh"

but then i learned that artist tags are supposted to be @<actual artist name>

so i corrected it, now fucking around with some sampler combos, but i think this one is nice

this took 140s on a 3090 (batch of two)

lets see if sage attention can improve it somewhat

it's also very good with artist styles. so yeah, big thumbs up. if this is the preview, i can't wait for a proper trained release!

[wispy canopy]

and pretty nice on styles, if you have an artist sheet

still figuring out the right lowest sigma, but i'll get that figured out 😄

[keen hollow]

let's see some more anima

[wispy canopy]

smth like this?

[wispy canopy]

AI putting the world on fire

the text encoder is seriously impressive, this is a kinda hard prompt (i specefically want her placed inside the bubble)

anywya, this is one of my favorite path of exile builds :
"Relic of the pact Scion"

and you can make the weird shit on it too. it's a very good model ❤️

[wispy canopy]

also anima xD

[near elbow]

looks like i should try it as well, thanks!

deranged robots with illustrious in the meantime

[near elbow]

cuuuuute

[keen hollow]

struggled with anima.. too hard to use 😐

[near elbow]

which aspect? the comparison workflow is just a mess, using tons of other anime models and creating an image side by side matrix... but the workflow in the example image does the job

nice distinct look

and hey, it does reasonable TEXT, eat that SDXL 😉

[near elbow]

anso with anima, not bad IMHO

but that's my rather open prompting - trying to get a specific look might be harder

[wooden sail]

Good at prompt understanding, composition and styles. Bad at details

[bold ether]

can always refine it afterwads with another model then

[near elbow]

having an anime model that does some basic text is neat

[bold ether]

didn't say that I would do that :p

[near elbow]

no worries - was just an opportunity to test speech bubbles 😸

[keen hollow]

I've been making a little game and now need to make some characters for it. 🙂 Need to pick a model.

Been messing with z-image for too long... forgot what was good. 😛

[keen hollow]

I dunno.. anima was hard to use. 😛

[near elbow]

depends on the intended outcome i guess 🙂
i'm just prompting along semi-randomly and like the results most of the time

if you don't need NSFW, many will do... i wouldn't go back to SDXL based stuff (no text, mixes characters and attributes, hard to get details)
so, e.g. z-image turbo ("cel shaded anime style" will do), or even flux klein, or qwen (does mild NSFW), ...

for instant-shiny-sexy, illustrious is still great of course, but see above for SDXL limitations

[keen hollow]

part of me wants to go all the way back to pony... but I think I'm forgetting about the bad parts like hands 😄

yeah z-image turbo does ok, but I think a dedicated anime model might help. I'm trying to generate characters that are basically anime stereotypes

[near elbow]

what's wrong with anima there? getting a specific repeatable result?

[keen hollow]

just can't get it to give me quality images without a bunch of extra prompting

So here's the progress so far. Each character has their thumbnail, plus 1 image in their gallery from the get-go. More are unlocked. (Details in 'how to play'). It's just an idle clicker game. https://oozedgames.com/ag/game27.html

I've only done images like the first 7 characters, but I gotta take a break. 😛

works best on pc, but might on mobile? dont know if scaling is perfect

[near elbow]

nice game, clicked around a bit, pun intended 🙂
what makes an image "quality" for you?

mandatory non-text

[keen hollow]

Was trying to get something with a coherent and reproducible style without having to rely too much on artist tags I don't know.

How far didja get in the game?

[near elbow]

24k or so, wasn't playing for too long before other stuff came up (and i don't want the browser to do too much at once)

yeah, consistency is totally optional for me, i just care for one cool image

[keen hollow]

Yeah I need the consistency a bit since the characters have unlockables.

[near elbow]

ah, and for the US folks... enjoy the super-whatever-thingy 🙂

[pale mural]

The Superball? Lol

[near elbow]

spread the message

[near elbow]

what does the fox say?

regarding anima, i just came across this - might be helpful for styles?
https://thetacursed.github.io/Anima-Style-Explorer/

[keen hollow]

Reposting in case anyone else wants to check out this game in progress 🙂

[wispy canopy]

so, tune some anima settings, download glm-4.7-flash, feed it random song lines, and then create some coherent imagary from it -> feed to anima with well known artist styles, and get this. imma keep this on autogenerate and see what else cool stuff comes out

[near elbow]

cool stuff 🙂

random kitsune for you

quite different style

and another

last one for now

[uneven ocean]

16 emoticons of chibi orange tabby cat in 4x4 grid, Chinese New Year themed, each panel showing different festive action:
(1) cat clasping paws greeting with golden "Fu" above head,
(2) hugging big red envelope with "Bao Fu",
(3) jumping with sparkler, night fireworks,
(4) eating sticky rice cake with crumbs,
(5) spinning chasing fluffy tail,
(6) covering mouth giggling with lantern light,
(7) sticking crooked Spring Festival couplet on door,
(8) kneading on red cloth with peanuts and dates,
(9) hugging big fathead fish proudly,
(10) curled in fur ball exhaling warm breath, snowy background,
(11) shaking bronze bell with tail swaying,
(12) lying on back paws up asking for petting,
(13) holding small lantern exploring night alley,
(14) rubbing owner's neck in cozy home,
(15) carrying willow branch knocking door,
(16) three mini cats stacked with tiny cup, family silhouette,
uniform panel size, clear borders, round edges, vibrant red and gold color palette, cute cartoon style, high detail, clean lines --ar 1:1 --style cute --no text, watermark, signature

[near elbow]

you realize there is no image generation here?

[wispy canopy]

also anima

[tawny wasp]

Dark Gloss Style (illustrious) OtakuStorm_Ai
https://civitai.com/models/2381812?modelVersionId=2678396

[near elbow]

who made her angry?

[wispy canopy]

abyss of memories?

[wooden sail]

Elf repellent

[wooden sail]

Severe lack of cookies

[potent venture]

do you use MJ or SD variants?

[wooden sail]

Local Wan 2.2 model. Recent video models are quite optimized for the new GPU and generate about 16 frames per minute.

[near elbow]

nice - my favorite way is uploading to grok and having it animate it for free with really good quality, even though i could do the local thing on the 4090... too impatient 🙂

here's the default "create video on upload" result. interesting how the sword changes hands magically in the end

"she throws the purple fireball at the camera and the scene transforms into a rainbows and unicorns wholesome setting, with her as friendly queen of the realm"
(via grok in the browser)

[near elbow]

another hijacked account - folks, remember, 2FA is not optional it's a necessity

[near elbow]

good listener for sure

[tawny wasp]

Bianca (Pokemon black & white) illustrious - OtakuStorm_Ai
https://civitai.com/models/2389449?modelVersionId=2686727

[near elbow]

frankenstein labs - always a good setting since you can justify AI errors by saying "well that extra finger was freshly stitched on" 🙂
(flux klein 4b distilled, extremely fast with 1-2 seconds per 1024x1024 image, but also quite sloppy of course)

[wispy canopy]

when your favorite vibe comes on

mind = blown

[near elbow]

chili for you?

[wispy canopy]

wait wait wait.... anima knows obverse and reverse?!

it ABSOLUTELY does

anima is CRAZY smart for the size it is

[near elbow]

and fast, and has a ton of styles, and a lot on top... i'm also totally loving it. and it will only get better, this is still the preview version

[wispy canopy]

yup

[wispy canopy]

this one will supercede illustrious, and it's not even a question

[near elbow]

and the nasty pony as well

i never got warm with that one

[wispy canopy]

pony was the first real showcase that sdxl can be retrained fully

just retrain it from the ground up -> now it can do anatomy

[near elbow]

but the quality score choices were very dubious even back then

well, here's to much fun with anima now and in the future

[wispy canopy]

astralite admitted he made a fuckup when training those scores

and an unintended effect happened

(basically, it only responded to all scores at the same time, individual score never really worked)

they are using the same system in anima -> but they trained it with 50% of the images, so it can the images with or without, so it can actually be used to steer image generation

[pale mural]

This is awesome

[wispy canopy]

old men unite!

[gusty palm]

trying to improve my overall image quality.
did i do something to break discord i cant see the other categories in this category

[bold ether]

Nah you good there's just no other survivor :p. In ancient times, there used to be more.

But that might not be the best channel to post that kind of thing. #📝｜prompting-help would probably fit better.

[gusty palm]

ok thks

[wispy canopy]

"In ancient times"... i was there

XD

[near elbow]

yeah yeah, back in the days in 2022 with the SD beta....

[near elbow]

tired?

[near elbow]

manga

[near elbow]

action

[vale cedar]

anima is surprisingly good at dynamic action poses for its size

[near elbow]

actually this is flux klein 9b, simply with "anime style kitsune female ninja wielding sword slashing attacking motion action" - it's not as refined as anima and much more censored, but can do more than i thought initially

the manga hippie gal above is also with flux klein

[vale cedar]

ah yeah my bad, good catch, flux follows specific prompts way better than anima

[near elbow]

i love _un_specific prompts though - give the AI some room

neko action

assuming her right arm is in the cape somewhere 🙂

[wispy canopy]

meanwhile, i'm just seeing how far i can push anima, having a blast with it

[vale cedar]

it's honestly great, try lower cfg if it starts looking a bit crunchy

[wispy canopy]

it wants some high cfg tho

got a few artists that are good for testing -> they overburn easily

[vale cedar]

nice, which ones are you testing? try some with heavy linework to see how it holds up

[wispy canopy]

dunno what i'm looking at tho xD

[vale cedar]

haha fair, did it go full deep fried or is it just weird

[wispy canopy]

nah, that's certainly the style

and probably the quality tags

oops, had some node deactivated that caused unwanted things

[vale cedar]

haha classic comfy struggle, happens to the best of us

[wispy canopy]

@vale cedar this kinda linework?

[vale cedar]

yeah exactly, looks super clean and held up well

[wispy canopy]

i mean, depending on artist tag, you get completely different outcomes, even on same seed

[near elbow]

"cel shaded style" also gives a nice look

[near elbow]

"classical manga style, neko female"
(SFW enough?)

[wispy canopy]

can't combine artists as with illustrious tho, but eh, the price of progress

[vale cedar]

yeah illustrious spoiled us with that, tagging is just built different

[wispy canopy]

but, do we get so much more defined styles? yes.

[near elbow]

anima - you have to try the time tags as well, e.g. this is "classical 70s manga style"

[wispy canopy]

all the same seed

[near elbow]

90s

[vale cedar]

90s is the sweet spot for that clean cel shaded look honestly

[wispy canopy]

@vale cedar just find a fav artist that does cell shaded type artist 😄

[vale cedar]

for sure, combining 90s tags with specific artists is like a cheat code for that look

[near elbow]

btw, have you tried the period tags "newest, recent, mid, early, old"? also neat

[vale cedar]

haven't messed with the period tags much on anima yet but sounds sick

[near elbow]

early in the prompt - the sequence is "[quality/meta/year/safety tags] [1girl/1boy/1other etc] [character] [series] [artist] [general tags]"

i like it even more the deeper i dig

[vale cedar]

yeah recent usually hits that sweet spot for balance honestly

[vale cedar]

nicee, the composition on that is actually super clean for anima

[near elbow]

still too much luck and too little control, but i love it when things come together 🙂

[vale cedar]

yeah it really is a gacha, you tried any controlnet with anima yet

[near elbow]

it's out of control anyway 😸

[vale cedar]

lmao true, but the colors are sick, that purple is popping

[near elbow]

i'm a big fan of small prompts and semi-random results, so no use case for controlnet, or regional prompter, or... - if i want to change things, i can still run it through qwen image edit

[vale cedar]

try wildcards for more chaos, anyone else got any crazy accident prompts

[near elbow]

hm, quite impressive for simply "make them male" (yeah, semi-success on the right 🙂 )

even captures the style quite well, love qwen

[vale cedar]

hmm yeah, try adding flat chest to the edit prompt next time

[near elbow]

"ink drawing" (flux klein 9b now)

[vale cedar]

that linework is crazy clean, flux really nails the high contrast ink style

[near elbow]

"1800s etching" 🙂

[vale cedar]

the cross-hatching on the etching one is sick, definitely feels authentic

[near elbow]

nice variety with anima again

[elfin pewter]

how does one get a specific action pose like this?

[vale cedar]

controlnet is the way to go, use openpose or depth to lock it in

[elfin pewter]

so i have to uploaed a image of pose in control net?

does that work?

[vale cedar]

yeah works like a charm, it just pulls the skeleton from the pic

[elfin pewter]

alr i'll try this, should work with all the models right? and ig i would have to add some specific related prompts too

[vale cedar]

yeah just match the versions, and prompts definitely help the model along

[elfin pewter]

thanks alot brotha, i'll try this tomorrow n let u know how it goes

[wispy canopy]

Nah, that one was pure prompt actually, there is no controlnet for anima xD

[vale cedar]

ah yeah my bad, good catch

[wispy canopy]

Not at pc atm, but I can look the prompt up later

[near elbow]

with a more modern model it's super easy ... prompt for it 😎
"brutal female crusader chainmail giant comically large huge broadsword. sword in right hand raised high, left arm extended you, right leg straight, left leg bent knee up"

[vale cedar]

that prompt adherence is actually wild for no controlnet, looks sick

[near elbow]

thanks - that's anima, rather modern, got a LLM in front of it so it understands what you say

[vale cedar]

yeah it actually following those spatial tags like raised high is a vibe

[near elbow]

how about yoga instead?

[vale cedar]

yoga in chainmail is hilarious, the model handled that pose surprisingly well

[near elbow]

"ballet pirouette" (SFW enough i'd say)

so the game of controlnet and posing stick figures becomes a game of "name that pose" 🙂

[vale cedar]

it's definitely way faster than fiddling with skeletons in openpose

[smoky leaf]

what checkpoint is this? and cfg?

[vale cedar]

that's a grok api node, doesn't use local checkpoints or cfg

[smoky leaf]

and how do u know?

[vale cedar]

cuz the screenshot literally says credits per run in the corner lol

[smoky leaf]

uh no it doesnt?

[vale cedar]

it's literally in the red circle you just sent lol

[smoky leaf]

what circle dude?

[vale cedar]

you literally just drew it on the second image lol

[smoky leaf]

bro wtf r u even on about? I asked another dude about his picture

[vale cedar]

the credits tag on top of the node, that's why it's api

[smoky leaf]

again, dfq r u on about

[twilit portal]

I think you're mistaken. My artwork is made on pixai.art the one he's replied to...

Or maybe you replied to the wrong person

[smoky leaf]

bro is geeked out on that mars kush

[twilit portal]

The model is VXP_Illustrious (low cost version) v1.9

CFG Scale - 3.0

There u go

[smoky leaf]

surprisingly bright for cfg 3

[twilit portal]

Yeah that's just the model style

[smoky leaf]

i see 3 models on civit, which one would it be

[twilit portal]

This one the low cost version is exclusive to pixai.art

[smoky leaf]

what does lowcost mean

ive got a v188 one locally

[twilit portal]

Lower credit cost than the normal illustrious model

[vale cedar]

usually just means it costs fewer credits on pixai

[vale cedar]

lol missed the limit by like 30 megs, illust is a beast

[twilit portal]

@smoky leaf here's the model
https://pixai.art/model/1819999994917251788?utm_source=copy_android

[smoky leaf]

i found one on civit

[twilit portal]

The low cost one?

[smoky leaf]

nah the full one i guess

https://civitai.com/models/1662539/vxp-illustrious-20

[twilit portal]

Yeah this one is exclusive to pixai as I said. Made the 002 above with this one

[vale cedar]

grab the regular 2.0, vpred is way too picky with settings

[smoky leaf]

thats what i did

[twilit portal]

2 more variations

[vale cedar]

nice one, those illustrious colors really pop without the cfg melting everything

[twilit portal]

Yeah. This model is my fav + great low cost

[wispy canopy]

the actual prompt on anima:
best resolution, highres, score_6, score_7, action scene, @abara_heiki, sword, attack trail, jump, grimace, dust cloud, from below

[near elbow]

hard contrast black and white ink splatter grunge style, attacking sword female warrior medieval ```

[near elbow]

armed and dangerous

[wooden sail]

Anime is quite something

[wispy canopy]

pew pew, anima is really good with guns

[near elbow]

yeah, no lack of firepower 🙂

[wispy canopy]

or just making a chill image xD

[near elbow]

totally chill 😇

pausing for some meditation (pushing a photoreal/3d look makes it a bit uncanny)

[near elbow]

the shadow with their own muzzle flash is... interesting 🙂

[glass citrus]

These just sucks

[tawny wasp]

Mrs Snake (Fate Grand Order) illustrious - OtakuStorm_Ai
https://civitai.com/models/2412957?modelVersionId=2712930

[keen hollow]

Been away a while.. I miss anything revolutionary?

[gusty palm]

https://cdn.discordapp.com/attachments/641133969898733578/1475231000383455302/biva_0222_1540.gif?ex=699cbb6e&is=699b69ee&hm=3a12c63d67b65e4d59a2034cb91588a0f60ab31db00a5c1982b4e4fb3029f7a1&

im working on this frame processor thing

its broken as hell still but its like half working lol

[near elbow]

what technology are you using? nowadays my advice is "animate it with grok and done", local video is so resource intense

[near elbow]

... in case you haven't tried it yet

[gusty palm]

just sd1.5 frame by frame with some optional blending tricks

[near elbow]

ouch... that's 4 year old stuff, i wouldn't try to get anything useful from that technology anymore

[gusty palm]

its like i2i with cn options automation script

i felt like theres probably better ways now

couldnt find any tho lo

[near elbow]

yeah, nice rotscope effect but not really useful 🙂

for local video, WAN is really good, but for SFW i just go to grok for free...

[gusty palm]

wats wan

[near elbow]

this is a result for my ad hoc prompt "cyperpunk night city raining neon streets, a car chase between a hovercar police cruiser flashing lights horns, and a slee black jet powered gangster car, epic, fast cuts, driving electronic beat"
takes around 1 minute to generate, really cool results, for free. and yes, i wanted hovercrafts not hovering normal cars, and the first car just vanishes after the camera move, but hey...

[near elbow]

https://github.com/Wan-Video/Wan2.2

[gusty palm]

ok thats interesting as heck thank u

the idea for the thing im doing is im using a video and extracting some frames to be processed with CN

so that gif was from some gtav capture

i have a gtx1060 so sd1.5 is about my limit i think

[keen hollow]

Yeah I tried anima but found it difficult to prompt

[near elbow]

just start with "masterpiece, best quality, score_7" and be surprised at the super random results, natural language works great for me

[keen hollow]

I'll have to girl it another whirl.

I do all my generation on rented now, so I have to set aside time to do it because I have to recreate the environment each time. 😛

got to the point where the code was moving to fast and wasn't being reviewed quickly enough by the open source gang... when I started seeing the malware stories, I was out

and don't have a dedicated machine to run it on... need me video card for my main pc 🙂

[gusty palm]

my goal would be to be able to take any image/video and run it though i2i and have it look cool but im like too far into this to stop now lol

oroginaly i was thinking i could render just depth map animations in blender and use that for the input, which is also a feature i just havent tried it yet

i already got claude pro by accident so im just have it make crazy amounts of code

Processing [5/9] vfs_2025-01-15_22-14-01_t11_40.png...
16:24:44
ETA: ~2.8min remaining (33.1s/img avg)
16:24:44
Input: 1920x1080
16:24:44
Pre-scaled input to 512x320
16:24:44
⛓ Using previous output as input (chain mode)
16:24:44
🎞 Pre-blend: previous frame at 80% (screen) under current
16:24:44
Output size: 512x320

it automatically resizes every frame and stuff to keep vram manageable

[gusty palm]

ill stop spammin now

[gusty palm]

https://cdn.discordapp.com/attachments/1197516228743401582/1475267936712265851/biva_0222_1807.gif?ex=699cddd5&is=699b8c55&hm=0eb1b909ab5c91622bfabbd037abb79d4bc2f44d0edfc66902d4f15f8c7a2538&

sorry one more, ran it on a video of a pizza i made lol

some temporal cohesion

it blends the input frame into the output of the last frame before each new frame sent to i2i

[pale mural]

Feel like im on acid

[quasi shadow]

Hey everyone! I’m working on a little edit video for fun and I was looking for some really strong visuals, and being in this server for a while I'm aware that there are some masters here. If you have any generations you’d be okay with me using, I’d really appreciate it if you could DM me. I’m especially drawn to pieces with bright colors mixed with deep/dark tones, super detailed or visually busy compositions. I’ll happily credit you if you’d like (Please keep it SFW)

Just in case, I'm not a bot or scammer

[near elbow]

have you tried editing images by prompting? all the major free services support that nowadays, no need for strange workflows

[near elbow]

that's super broad - what's a strong busy visual with vibrant colors to you? what kinds of videos are you aiming to generate?
this is "anime style image, bright colors mixed with deep/dark tones, super detailed or visually busy compositions" with flux klein 9b base, just for fun (yes in one image)

probably too much text for video input purposes 🙂

generic anime cover

samurai action

[agile dew]

Ayo how

[main orbit]

Wym?

Misato's cooking...

The power of high-effort genning 😎

[agile dew]

Would you mind sharing the LoRA or dataset you trained it on?

[main orbit]

Dataset won't help.

I only used FurryToonMix IL model. That only gave it the comic-book artstyle.

The real deal comes from using Krita AI with its photoshop-like and inpainting tools.

I do everything locally btw. But Krita AI does have an option to get tokens and whatnot.

[agile dew]

The precision in controlling the character’s expressions and emotions here is honestly impressive.

[main orbit]

Thanks. I've been practicing facial expressions via inpaint. Sometimes I just use img2img + photobashing + prompting to get the results close enough.

A lot of the time I see a rawgen that looks good enough for me to work with.

And mind you: I've been doing this since Septemper 2024 with little to no drawing experience. I did learn art/perspectives along the way though via observation...plus I have a wacky/random imagination.

I did furry art a LOT (cuz they were the ones who initially taught me about AI art). Now I'm finally adding humans, starting with characters that already have datasets like Evangelion.

[agile dew]

This is crazy good. I’ve been testing things too, but no matter what dataset I use, I just can’t seem to nail the emotions or facial expressions.

[main orbit]

You can DM me so we don't clog up the original intent of this channel. I could take a gander at what's going on.

I can even link you my portfolio/DA.

[agile dew]

Take a look at this it looks okay overall, but the expression feels blunt and not genuine.

[main orbit]

Can't see it. Upload it as a png or something.

[main orbit]

I have difficulty with expressions as well. It takes me quite a few tries to find the right one. Asuka's was incredibly difficult.

Making dumb faces are a bit easier.

[main orbit]

You doing this online or locally?

[agile dew]

Honestly, I appreciate what you’re doing I’ll keep pushing to improve. I’m just running things online for now since my graphics card struggles with heavy local work.

[main orbit]

What's your specs if you don't mind me asking?

(Include VRAM)

[agile dew]

It’s pretty bad I feel like I’d get bullied out of existence if i shared 😅 😭 . But I’ll keep improving. I’m still saving up for a better PC, I’ve got $758 saved so far, and my goal is $1,200.

[main orbit]

I've tutored someone who had a laptop with 6 GB VRAM...

And very low RAM. Albeit sometimes they buy tokens from Krita AI to be able to upscale. I only recommended that to prevent forest fires.

[agile dew]

I run multiple things locally, like Qwen 3 TTS and Voicebox, and they’re pretty heavy. Until I get a better PC, I can’t risk putting too much strain on my system. I rely on them for YouTube, so they’re really important to me.

[main orbit]

Ooh AI voice. I'd like to get into that. Would love to check what voices are good on my OCs.

[main orbit]

Building a decent PC right now is stupidly expensive. Like....$1,024 for 64 x 2 RAM sticks?

[agile dew]

Oh wow, I didn’t realize you were into AI voices too! That’s awesome

[agile dew]

Yeah, it’s craz I feel like RAM prices are insane these days, like $700 for DDR5 on a good day 😅

[main orbit]

I'm not. I'd like to learn though.

[main orbit]

I built my gaming PC during end of 2022. Upgraded my GPU before the price skyrocketed around Nov 2025.

I remember when people were ridiculing me for getting 64 GB RAM a couple years ago. Guess the jokes on them now.

[main orbit]

Tbh I just want to find some good voices for my OCs.

[quasi shadow]

This would work, I just want something visually strong, not anything specific. Plus, by being broad I don't filter anything that could work but isn't exactly what I'm expecting

[near elbow]

what are you specifically looking for? e.g. "intense focused" works like a charm (anima here)

[near elbow]

that would have been perfect a year ago, but with the current price madness... hope the best

"joy happiness bliss" (and my setup 😉 )

[main orbit]

Really depends on the scenario tbh.

[near elbow]

and to mix moods you need a modern model of course, or they will just blend into each other...

[agile dew]

How are you guys getting this insanely good at character consistency? 😭

[near elbow]

not at all 🙂 - it's all in one image so they can be very similar easily. getting exactly the same folks in a new image is much harder

[agile dew]

I literally have to Frankenstein it together 😭 I’m stacking every model imaginable just to get the face right, then another for the fingers, another for the feet, another for the background… it’s like building a character out of mini LoRAs. And even after all that, it’s still a 50/50 gamble whether it comes out perfect or cursed.

Let’s run an experiment 👀 We all use the exact same character prompt and see who can create the cleanest, most consistent image. Let’s find out how perfect we can actually get it. Because some of you are way too crazy with this 🙂😅

[agile dew]

The background composition feels cluttered, and I’m struggling to find a clean way to fix it, and the faces

[gusty palm]

yeah ive seen it a bit or tried it in claude but i ditn know if it was free

havent really seen this tho

its just seed walk though so maybe its just too old to be popular anymore

[gusty palm]

https://cdn.discordapp.com/attachments/548263195135967234/1475635688308346982/biva_0223_1824.gif?ex=699e3454&is=699ce2d4&hm=b6c10872b651a8fb119cc7587f19e35acbfa52ef3499110076723bfb29204924& using lcm lora i can dramatically speed up generation

Finished in 2.7min · avg 5.1s/img
18:24:18
Batch complete!
18:24:18
Auto-building GIF from 32 frames…

[gusty palm]

https://media.discordapp.net/attachments/548263195135967234/1475647751416058028/biva_0223_1917.gif?ex=699e3f90&is=699cee10&hm=0425549b72d553b5db3ee996880e1f09f8c53d87f36d6405ee88d48832f02704&=&width=336&height=576

[gusty palm]

https://media.discordapp.net/attachments/548263195135967234/1475656413945790664/biva_0223_1951.gif?ex=699e47a1&is=699cf621&hm=11ceaea81f58c90c2f05d2b17beb40aba3cdbb4afa47df05261c5b5a647074af&=&width=288&height=576

it does a thing where it feeds the last frame into the input so it wanders less

rather than using the selected input frame over and over again

[gusty palm]

realized ti ui sends values per frame so if u cnange something while its running it gets incorporated

[near elbow]

interesting results but not the way to do AI videos nowadays 😉

[near elbow]

morning coffee

[gusty palm]

https://cdn.discordapp.com/attachments/1197516228743401582/1475756535157096489/biva_0224_0229.gif?ex=699ea4e0&is=699d5360&hm=38a58a953150e05ed4058784e8d1e2d423eca5beadbc6a269f94d8013f3d7a83&

yeah its weird

artsy tho

its retro LOL

[gusty palm]

https://cdn.discordapp.com/attachments/548263195135967234/1475776040923172968/biva_0224_0346.gif?ex=699eb70a&is=699d658a&hm=bcc806c7a6dbae587ffc3a37a00897d4ebc3c68063e2253ebfddadbdb4496c73&

i got this one to do the thing

then went too high on the noise

[gusty palm]

https://cdn.discordapp.com/attachments/548263195135967234/1475782326179860572/biva_0224_0411.gif?ex=699ebce5&is=699d6b65&hm=57f6bb905551926eb3e992ac5135b6d9807dcfd1b24000672fcfdb639c0753b0&

i can almost keep it going forever

https://cdn.discordapp.com/attachments/548263195135967234/1475785734458703882/biva_0224_0425.gif?ex=699ec011&is=699d6e91&hm=a79453f5f7cd1ec7860d43007045dfde5cc942df4234bd95d878805bcb4bde9d&

is there still public sd apis aviabile?

like the horde thing or that a11 api

[pale mural]

Super trippy

[near elbow]

there probably still is the official stable diffusion website and API and all that, but paid. horde is a community effort. i just use my local stuff (or advanced models via replicate on a pay-per-use basis)

[wispy canopy]

anima being anima

[near elbow]

"recursive neon eyes fractal stretching to infinity"

[near elbow]

and a nice kitsune

[gusty palm]

how in the did the gun into a duck

[gusty palm]

https://cdn.discordapp.com/attachments/1476029660650274828/1476503569443197072/biva_0226_0357.gif?ex=69a15c9b&is=69a00b1b&hm=c60cd1376a25f29255ace5f6046786035db5edb6cd797dbc9d5460113247e78c&

[gusty palm]

i love these

[gusty palm]

https://cdn.discordapp.com/attachments/1476029660650274828/1476519878163828787/biva_0226_0502.gif?ex=69a16bcb&is=69a01a4b&hm=c8a9c35f6a58b6c62860874a58b9ff504e35fe5167d4c2c2ee4af18ff91cd2ed&

[gusty palm]

I added a thing that gives me across hair to see where it's zooming

and u can switch CNet models and stuff on the fly between frames

[near elbow]

"a cool camera dive into a recursive fractal vibrant psychedelic construct, self-similar, enticing" with grok

[keen hollow]

Haven't tried anything with grok.

[gusty palm]

im working on adding RIFE or first to smooth motion maybe

[near elbow]

just do it, free tier, 30 videos per day, can't do anything wrong there (and i'm not sponsored by elon 😉 )

[gusty palm]

google gave me an offer like 3.99 for 2 months with access to their video thing

im so close to fully local interesting video though i dont wanna quit yet

found out gemini has free api

[gusty palm]

woa

[near elbow]

sniper kitty

yeah strange rifle and a bit of tail below, but why not...

[tawny wasp]

Mrs Snake (Fate Grand Order) illustrious - OtakuStorm_Ai

[keen hollow]

vast has become more onerous to use... I need a new service 😛

just takes much longer to set up the instance these days

[near elbow]

if you just want quality output and SFW, replicate is a good option - paid per image, and has all the big really good models e.g. seedream 4.5
"a sequence of manga panels describing the advantages of using replicate as a service for AI image generation" ... well, basically most of this is nonsense technically, but a nice sequence and semi-convincing 😎

[tawny wasp]

raiko_(lieumare)

[keen hollow]

I haven't spent a lot of time with the full services like that. Other than some of the free demo stuff. I've been using the gpu-rental to set up my own environment (typically swarm or invoke).
But they build from scratch each time and it's becoming a lengthy process that's kind of a pain. Set up the environment (with slow downloads), then transfer in big models, just takes a while.

[near elbow]

depends on the use case - personally, 4ct for a 4k image at replicate with seedream 2.5 (or 3ct with 2.0) is totally worth it since it's spot on nearly all of the time at first try, 4k native avoids post-upscaling etc, and i love the style and understanding it has
and, well, see for yourself (1:1 as prompted, no postprocessing)

still feels like magic, coming from the SD public beta in 2022 and everything that was super hard back then 🙂

[keen hollow]

At $0.035 per image, yikes.

I feel like Nano Banana gets me there for free. 😄

[near elbow]

well, "depends on the use case" 😉

[tawny wasp]

Haneda Tobari Real Form (Jingai Kyoushitsu no Ningen-girai Kyoushi) (A Misanthrope Teaches a Class for Demi-Humans) llustrious - OtakuStorm_Ai

https://civitai.com/models/2432521?modelVersionId=2735102

[agile dew]

Pretty good 😊💯

[gusty palm]

https://cdn.discordapp.com/attachments/1197516228743401582/1477916242886131814/rife_4x_biva_0224_0136.mp4?ex=69a68042&is=69a52ec2&hm=04fd97c803b7e95e8b9af749e032636aaff0df97aba2940a0eee38273163f040&

i got rife workin

no more choppy zooms

well its not perfect

[near elbow]

intersting though

"exquisite details, a fractal camera zoom into a 3d fractal landscape full of intricate shapes and textures" with grok - hm, not 100% anime related, but for fun

anime!

[gusty palm]

crazy how good that loks

[wispy canopy]

cyber-elves are smth else xD

[wispy canopy]

just anima doing anima things

[tawny wasp]

Flare Corona (Fairy Tail) llustrious - OtakuStorm_Ai
https://civitai.com/models/2438337?modelVersionId=2741594

[tawny wasp]

Female Protagonist (Pokémon Winds and Waves) llustrious - OtakuStorm_Ai https://civitai.com/models/2443490?modelVersionId=2747360

[near elbow]

crossover

[tawny wasp]

Yoruichi Shihouin (Bleach) illustrious - V3 OtakuStorm_Ai https://civitai.com/models/2446039?modelVersionId=2750303

[floral flame]

https://youtube.com/watch?v=uXgW0N-kYmI&si=UukTrHZnfFUFncQ3. Guys check out the Ai tool

[keen hollow]

veo is wild

[tawny wasp]

Thunder Raiko (Tojima Tanzaburo Wa Kamen Rider Ni Naritai) illustrious - OtakuStorm_Ai
https://civitai.com/models/2454182?modelVersionId=2759512

[keen hollow]

hmm

[tawny wasp]

My Vampire Girlfriend
https://civitai.com/posts/27181220

[near elbow]

be careful with vampires in general 🙂

[near elbow]

baroque

[tawny wasp]

New Lora Mirko

[near elbow]

allnighter

[haughty stone]

no9ggiger

[wooden sail]

With Anima you just prompt for 2 by 2 grid and get 2x2 grid ☕

[tawny wasp]

Mirko (Rumi Usagiyama) My Hero Academia (IllustriousXL) OtakuStorm_Ai
https://civitai.com/models/2463873/mirko-rumi-usagiyama-my-hero-academia-illustriousxl-otakustormai

[keen hollow]

I'll tell ya.. I had a heck of a time last night finding a model that inherently knew all the characters I was throwing at it... even very basic ones. I feel like my prompting and model selection abilities get worse over time. 😄

I figured illustrious models would know more Marvel / DC characters.

Had a little project I was working on but if I have to get a lora for every character it isn't going to work. 😛

[near elbow]

another style

[near elbow]

have you tried anima for that? just curious, i don't know most anime series and characters myself... marvel and DC of course, but also not the 1001 spin-offs

[keen hollow]

I did give it a whirl but didn't have much luck. I think the Anima preview data set is quite small, so not surprised.

[wind pelican]

please dont consult the warhammer animus for vampire girls

theyre engineered and can take stuff in their blood and afford to heal up and produce more blood or something. normal humans cant

[tawny wasp]

My Kuusetsu Enhanced Form

[wind pelican]

her eyes is like sans!

[twilit portal]

[tawny wasp]

Karin Ryuuzaki (A Misanthrope Teaches a Class for Demi-Humans) illustrious - OtakuStorm_Ai
https://civitai.com/models/2471384?modelVersionId=2778659

[tawny wasp]

do you like my gothic frieren?

[near elbow]

bored - how can we cheer her up?

[tawny wasp]

Kuusetsu Enhanced Form (Mato Seihei no Slave) illustrious - OtakuStorm_Ai
https://civitai.com/models/2475463?modelVersionId=2783210

[tawny wasp]

thx a lot 🙂

[keen hollow]

The cheese pull panel.. lol.. why?! 😄

I can't.... 😄 😄

[tawny wasp]

Frieren Gothic (Sousou no Frieren) illustrious - OtakuStorm_Ai
https://civitai.com/models/2483477?modelVersionId=2792147

[inland marlin]

Benchmarking a 5090

[pulsar drift]

damn, how do you animate it?

[tawny wasp]

Hi, you can use grok or comyui

New lora is coming :whitefox marvel rivals 🙂

[twilit portal]

thicc

[versed sapphire]

A complete musik video. About Waifu TCG Anime to real transformations. Enjoy the ride and give some feedback. This took some effort to make. https://youtu.be/kwWbIpy1cuU?is=rkCqCBxnGwSSZzM9

[near elbow]

red panda power

[twilit portal]

Rias x Demon Slayers ❤️ 💚

[keen hollow]

evenin'

[keen hollow]

We need a lot more images here... getting so empty

[near elbow]

reading "chat" i thought of other things (black chat? black rooms? 🙂 ), but right, french for cat...

[agile dew]

Nice control

[agile dew]

Hey can you elaborate on this

How?

[near elbow]

all the stereotypes

[keen hollow]

Lol the baguette. Needs a striped shirt 😝

[keen hollow]

Not local for stuff like that. Nano Banana 2.

[tawny wasp]

White Fox Marvel Rivals (illustrious) OtakuStorm_Ai
https://civitai.com/models/2506585?modelVersionId=2817577

[near elbow]

"female french red beret carrying baguette smoking cigarette striped black and white blouse, walking down the streets of paris, coy" with anima this time

strange baguette grip but it seems to work for her

[keen hollow]

horizontal stripes lol

In watercolor.

Outfit change.

Style change.

[bold ether]

why môts with an accent tho

so close

(french is stupid hard)

[keen hollow]

no idea... don't speak french.. 😄

[twilit portal]

Akeno x Demon Slayer 🖤

[near elbow]

c'est dommage 🙂

(just a few words here from highschool waaaay back then)

[wooden sail]

🥺

[muted mountain]

It's just an error from the ia, normaly There aren't any.

[bold ether]

effectivement

[keen hollow]

Who was it that was posting those fun magic outfit-change videos? We need more of those.

too quiet in here 😛

[wooden sail]

I would've posted more if my images weren't ignored without a single reaction

[near elbow]

grok free video is paid now, unfortunately, so i'm only left with bing (super censored) or local generation (too impatient 🙂 )

have a happy family instead

[wooden sail]

Pose-cloning

[keen hollow]

Try the google options.. I like those a lot. 🙂

[keen hollow]

yay!

This is wan 2.2 right? Have you tried that newer LTX 2.3?

[wooden sail]

Still haven't downloaded it. It seems to be more tuned for realistic movie stuff. Sound is not a selling point right now - for anything serious you need to assemble audio track from scratch anyway

[keen hollow]

Got it.

I did a little bit with these but they were fairly complex to use in SwarmUI.

Maybe I should try again.

there were some easier-to-use models that people released that helped combine some of the various steps in the process

[keen hollow]

Of course she'd have a third arm...

[bold ether]

that's magic for you

[near elbow]

free tier video? have to check it out...

[keen hollow]

desert wasteland coffee

[near elbow]

hopefully it's hot!

[near elbow]

in for a bite?

[near elbow]

robo bunny and friends

[twilit portal]

Yae Miko x Demon Slayer 🩷

[wispy canopy]

you might be cool, but are you skeleton standing in front of an explosion cool?

[wispy canopy]

🍄

[wooden sail]

🍄
Are you challenging me

[near elbow]

waiting...

[near elbow]

🙂

[near elbow]

fishy

[quasi shadow]

Five fingers and a half

Almost got it

[near elbow]

gals with guns

[near elbow]

beach bummed

[junior hound]

Jacob transfers to a new school, thinking he will finally find his place. But between the masks people wear and his own spiraling thoughts, nothing goes as planned

Anime name = Hight School Life

[near elbow]

"a generic anime panel sequence mocking the use of AI image generation" just for fun (yeah that was the prompt)

[wispy canopy]

bringing gun to compensate, your shroom must be tiny xD

"why is your shroom so tiny?"

[wispy canopy]

what model u using?

[near elbow]

seedream 4.5 via replicate API - just another level altogether 🙂

usually i go with anima for the, well, anime looks, and z-image turbo or flux klein for photoreal, but seedream is a ton of fun

cloud only and paid, unfortunately

[wispy canopy]

yeah, for me it's local or bust

[near elbow]

same here, 99% of the time... but this level is just cool

[near elbow]

https://huggingface.co/baidu/ERNIE-Image looks promising as well, comfy support incoming

[twilit portal]

Rias' Day Out
She rockin' that casual fit 😍

[near elbow]

slightly unhinged

[near elbow]

neko power!

[bitter birch]

What could be the prompt for this kind of hairstyle?

[keen hollow]

I asked Gemini 3.1 pro about it.

Got this:

To generate this specific hairstyle in an AI image generator, you need to focus on the color, volume, texture, and the specific retro aesthetic it evokes. Here is a breakdown of how to prompt for it, depending on the type of AI generator you are using.
Core Keywords & Tags:
If you are using an anime-specific model (like NovelAI or models based on Danbooru tags), use a combination of these tags:

Color & Length: blonde hair, medium hair (or shoulder-length hair)
Volume & Texture: big hair, voluminous hair, fluffy hair, messy hair, curly hair, loose waves
Bangs (Crucial for this look): swept bangs, feathered bangs, parted bangs, hair flowing back
Aesthetic/Era: retro artstyle, 80s style, 90s style, retro anime
Accessories: hair clip, hair ornament (for the small clip on the side)

Ready-to-Use Prompt Examples
Option 1: For Anime Models (NovelAI, Stable Diffusion with Anime Mixes)

(masterpiece, best quality), 1girl, medium blonde hair, big hair, (voluminous fluffy hair:1.2), loose bouncy curls, messy hair, thickly feathered swept bangs, retro 80s anime hairstyle, small pink hair clip on side

Option 2: For Natural Language Models (Midjourney, DALL-E 3)

Medium-length light blonde hair with massive volume, styled in a retro 1980s blowout. The hair is extremely fluffy and bouncy with loose, messy curls and thick, feathered bangs swept dramatically to the side and slightly back. She is wearing a small hair clip on one side.

Tips for Getting it Right:
The "Era" is key: This hairstyle relies heavily on a 1980s/early 1990s "bubble economy" anime aesthetic (think City Hunter or older Sailor Moon characters). Using words like "80s blowout," "retro," or "feathered" will do a lot of the heavy lifting.
Focus on Volume: Without words like "voluminous," "big hair," or "fluffy," AI tends to make hair lay flat against the head. You may need to put emphasis/weights on these words if the hair isn't big enough.

Give those a try

Should get you close 🙂

[near elbow]

"1980s big high curly hair blonde head portrait"

[bitter birch]

Alright, I appreciate it, Man

[bitter birch]

aight, thanks. Will try it too

I almost made it

NoobAI Based model

[flat lynx]

react on design

[near elbow]

a bit of overkill, holding a vote, but it's a nice overall anime design with bokeh background and good details, and the AI managed to generate a workin gbow, which is not always a given 😉

just my personal opinion of course

[near elbow]

totally wholesome

it's healthy for you

[flat lynx]

#vote for the Dawn of the Divine Archer

[near elbow]

that's nice but not anime 😉
which i would see as a prerequesite for posting things here in #🍥｜anime

[flat lynx]

@near elbow I did not get u

@near elbow check my previous designs and guide me with that too

Fast please

[near elbow]

the previous "guy in a jungle with a bow" was anime style and i liked it
the dawn image is nice but not anime, so it's not the best content for this channels. i don't care, there is no anime-only police here, just saying. that's why i didn't rate it.
also, a simple "good/bad" rating doesn't make too much sense since it's super subjective and you neither learn what you can improve by people voting "good" (there's always some helpful feeback possible if i like an image) nor "bad" (you'll never know what they didn't like)

and we're on a semi-dead server, so you'll be lucky if you get more than 2 votes if at all, which reduces the usefulness even more 😉

[flat lynx]

Bro will you guide me

[flat lynx]

Ho

Hi

@near elbow hi

@bitter birch hi

[near elbow]

well that looks more like anime. the scenery is nice, golden orange tone for all.
it's not super perfect though, e.g. the sun is shaded (hey it's the SUN there is no shading on it)

unless you say what specifically you want to improve it's hard to tell which observations are helpful for you

also, this is the last time i'll spend several minutes replying - this server is semi-dead, so there are not many folks who benefit from it, and i'm not dying of boredom here 😉

have fun generating images, and enjoy what you find beautiful yourself - don't make yourself dependent on others too much

[near elbow]

well, as expected 😉

sunset meditation

"beautiful nature scene, sunset over a lagoon, female archer bow arrows sitting on the beach looking into the sunset, solemn meditation mood, orange color scheme papercut style" (sharing is caring)

same with ernie, more prompt adherence and less anime style
(the arrows with feathers on both ends are probably not a good choice for hunting 🤪 )

adding "anime manga style"

enough for now 🙂

[dense raptor]

Hey guys, quick question — do you also get a lot of bad outputs like extra fingers or weird hands in anime images?

I’ve been working on a small tool that automatically detects these issues and filters out bad images so you don’t have to check manually.

If anyone wants, I can test it on your images and show results 🙂

[bitter birch]

ts is funny

[bitter birch]

is it ADetailer?

I've been using ADetailer

my new damn lockscreen

[dense raptor]

No I have design my own

[bitter birch]

Oooh, I get it