🍥｜anime

[river shale]

Got help setting them up and have had help with choosing models and such stuff. 🙂
I believe the wildcards I used for those I made myself 🤔
With inspiration from others ofc.

[tawny wasp]

Minami Shiunji From (The Shiunji Family Children) illustrious Apex_Ai
https://civitai.com/models/1678679?modelVersionId=1899979

[sharp imp]

sussssyyyyy, and blocked

[gleaming flame]

playing with nai-illu models today

[river shale]

[undone prism]

Made By Nimbus AI

[river shale]

Have a Leshy

[river shale]

And my rendition of will-o-wisps

[finite nova]

stolen prompt

[gleaming flame]

last one for now....

[tawny wasp]

Haumea From (Fire Force) illustrious Apex_Ai
https://civitai.com/models/1684092?modelVersionId=1906068

[umbral cave]

请给我生成一张labubu的手机动态壁纸

[indigo oyster]

the Rudeus RGB team😁

[acoustic agate]

Love the chromatic aberration on this

[steady pumice]

Thanks friend

[misty violet]

i think this is a sign to rewatch this anime series, btw love this one!

[tawny wasp]

Arrow From (Fire Force) illustrious Apex_Ai
https://civitai.com/models/1690145?modelVersionId=1912805

[finite nova]

colab works better than kaggle for no god damn reason its the same code 🤑

[river shale]

Makin' scene girls

[woven gust]

hi, which model for the first picture? Its very detailed and night style

[sharp imp]

Personal merge I'm tweaking and testing. Closest thing available on civit right now is Ramthrusts Red Shift. Once I'm satisfied with my merge I'll upload it to civit.

[woven gust]

thank you very much

[sharp imp]

Yep. He's made a new Black Magic that's really clean too. More 2D clean art style, less detailed. And he released a new version of Pink Alchemy that has less glossy skin. All top tier merges.

[misty violet]

when three powerful characters from frieren anime are in one framee is freaking awesomee...btw i just recently finished watching it👌

[river shale]

I've created a space for sharing anime waifu's it's 21+ - If you're interested LMK

[tawny wasp]

Otaku MIX illustrious XL V3 Apex_Ai
https://civitai.com/models/1693483?modelVersionId=1916558

[indigo oyster]

walnut + PS

[wooden sail]

Requested. Or or?

[river shale]

🧀 🪿

[misty violet]

love the diverse set of characters bro... awesomeee

[sharp imp]

Thanks, I was testing the models knowledge of em lol I've got a pretty long list I've still gotta test.

[fickle oyster]

You've been holding out on us

lol

[sharp imp]

i have not lol i thought i shared all of those with you 😛

i may have forgotten

[fickle oyster]

Me too my brain is mush today

[sharp imp]

i shared some, not all heheh you just have to scroll QUITE far to find em

[fickle oyster]

lololol

Sd fam is fam

[river shale]

now THIS is showing off 💪

[fickle oyster]

Yesh I helped him make the model :3

He just does wild things with it and surprises me er'day

[river shale]

Always a big fan of Rets work 🙂

He's helped me learn alot n_n

[wooden sail]

This is from Chroma v34

[fickle oyster]

I have so many images but

https://tenor.com/view/what-are-the-rules-iasip-gif-27514763

[sharp imp]

who even knows anymore

uhhh no nsfw stuff iirc

[fickle oyster]

no nsfw stuff?!

[sharp imp]

correctomundo

at least thats what i was told

[sharp imp]

borgar

[sharp imp]

she went to LA recently

[tawny wasp]

SCP-1471 "Mal0" (SCP Foundation) Illustrious Apex_Ai
https://civitai.com/models/1706446?modelVersionId=1931116

[bold ether]

yup this is a christian minecraft server, keep it sfw please.

[fickle oyster]

How much SFW are we talking?

[river shale]

I hear from Ret that you do lots of cool stuff and things- (I also love your work you post here) I was wondering if by chance you had a twitter? Im making one to go with the discord server Im starting and would totally love to support your work on there if you post~

[fickle oyster]

@sharp imp can you send her an invite?

Fuck it, right?

[sharp imp]

sure

she's good people, and eager to learn lol

[river shale]

Awe shucks, you're good people too ret.

Thank you both 🙂 I hope I don't disappoint!

[sharp imp]

you bout to see some cool stuff lol we got some people that put me to shame in that lil hidey hole

[fickle oyster]

Yeah we're a friendly and accepting group excluding me

I'm the asshole :3

[river shale]

I cant wait to see it 🙂

[river shale]

BAH

[hearty heart]

me!

[glacial coral]

Anyone know some models for 3D style u can recommend for me if it’s blender, SFM or unreal engine idm I just forgot the ones I was recommended before and im struggling to find some on civit

[sharp imp]

uncanny valley and HS UltraCG HD

are two i recommend

[glacial coral]

Ah ty coming in clutch ty

[sharp imp]

i gotchu fam

[glacial coral]

any specific version of Uncanny that does it best

[mossy depot]

https://civitai.com/posts/18639430

[desert island]

ccg card maker! the workflow is kinda ugly but if you all want a copy of it I'll include it and make quality of life improvements. You add your own text in where the sample text is and right now it just creates a random icon up in the top left. It's kind of fun.... The last image is the template for the cards.

[desert island]

Then you can do way better holographic cards like this 😄

[stray vapor]

cooool!

[desert island]

so for the video it's just indexTTS and Sonic for the lipsync. This stack makes it so that I can do voice cloning and lipsyncing without having to use any of the paid services. I like doing local generations...

[tawny wasp]

My Wendy Corduroy From Gravity Falls

[dapper bane]

She is cool)
__
Where on this server do I enter prompts to generate images?

[desert island]

This isn't that kind of discord as far as I'm aware. This is a place where people share the things they've created using stable diffusion, whether it be a locally generated image/video/music/audio via comfyui, swarm, forge, etc. or rented server hardware to generate the stuff they are making.

Wait... I stand corrected... I don't know how much it costs but you can do it in the artisan section... here's the info -> #artisan-faq message

[fickle oyster]

Testing a brand-new k sampler with strategic noise injection. It's absolutely wild.

[stray vapor]

omg this is coool. where did u make this???

[sharp imp]

comfyui

local gen

[tiny flower]

devcdxa cfwe2

[velvet drum]

whats the lora for that style, please?

[gleaming flame]

96yottea and Ohuton loras, both from civitai

with some creative prompting

[velvet drum]

@gleaming flamethanks

[scarlet oriole]

PART 3.5 GIOGIO

[devout swan]

A lot of my illustrious generated images keep having double upper eyelids. Does anybody have a solution for this? It almost always happens with something like half-closed eyes, jitome or narrowed eyes, but also quite frequently with normal, wide open eyes.

[sharp imp]

Sounds like a model issue. Buuuuttt i dunno what model your using so I couldn't say. Could try an eye detail lora and see if that fixes it.

[solar herald]

hello there, i need some help about creating consistent characters without a character Lora, im using ComfyUI and the WaiNSFWIllustrious model. i already have a image of my character and now i want to know how i can create a workflow or use a working workflow that gives me consistent characters in anime style, until now i used the workflows from YT but they all use them on realistic characters not anime style and they dont seem to work. so if anyone has such a workflow or knows where i have to search for it please tell me. Thanks ind advance

[steady grail]

ip adapter or img2img are the best picks

using the same seed, usually helps in that

[solar herald]

ok i can imagine that it might work with ipAdapter but with Img2Img? what i want is the same character (as in the image) in mutliple poses (with controlnet, prompting or both) and maybe some multy character images would that work with Img2Img too?

[steady grail]

the image is used as noise, so, starting by a simple sketch and using the same seed, can limit the freedom from the model and achieve a predictable result

with a better sketch and less steps, the result may be closer to the envisioned result

another pick is ip adapter and open pose

ipadapter picks the details, and open pose picks the pose

using styles and other characters, is another way to limit the freedom of the model

like, misty_pokemon, blue_hairs, big_eyes

[coral blaze]

Hy anyone's have hey gen ai membership contact ya who know how to do ai voice dubbing contact me

[tawny wasp]

Wendy Corduroy (Gravity Falls) illustrious Apex_Ai
https://civitai.com/models/1716094?modelVersionId=1941987

[nova remnant]

Hey, may you send me a friend request ?

[solar herald]

so after some reasearch and try and error it seams i found/created a comfyUi workflow for consisten anime characters while ussing an illustrius model but now there is a new probelm...clothes it seams that the Ip adapter cant put them on i have images of the clotings (top, skirt etc.) but it seams not to woirk has anyone a idea?

[solar herald]

here is a part screenshot of my workflow dunno if this helps

[wooden sail]

You want to put clothes on a stylized character? Only one model can do it reliably - Flux Kontext with multiimage.

[solar herald]

ok thanks for this info but no matter what i make i dont get any flux version to work on my PC (GPU RTX4060 8GB) so i gave up on flux. im currently tyring it and i think i get it lowly to work

[wooden sail]

It's not released yet. A new model after all. And IP-Adapter is very very limited. Think of it like an additional prompt text - it can't combine things unfortunately.

[solar herald]

...does anyone knows a openpose model that works with Illustrious models? my seems to work only to 75%

[elfin moat]

been having fun with the 0huton lora, thanks for sharing

[gleaming flame]

boring as hell

[misty violet]

maybe there's more fun in hell tho , just kidding :>

[wooden sail]

Do we even need character loras if with Flux Kontext..?

[sharp imp]

i just send characters by name with my ckpt lol dont need loras for most of them

[wooden sail]

Do you have a character that you want to generate but the model don't know about nor a lora exists?

[sharp imp]

havent found one yet in this 30k character list lol but when i do ill let you know

of course when i do, ill just make a lora for it, then bake it into my model, and boom problem solved XD

[wooden sail]

Just give me an image of a single one

[mossy depot]

https://civitai.com/posts/18819261

[random oracle]

aslamuku alykum

[nova remnant]

@autumn roost I dunno who to tag but bro is posting scams

[autumn roost]

torched

[nova remnant]

Thank you good sir

[mossy depot]

https://civitai.com/images/85004704

[merry shale]

👍

[keen hollow]

Been traveling for a few weeks... what's cookin'?

[sharp imp]

New merges, new loras, new comfy custom nodes. The usual

[keen hollow]

Eh, more of the same. 😛

[sharp imp]

The more things change, the more they stay the same.

[wooden sail]

Flux.1 Kontext [Dev] has been released

[keen hollow]

Oh? What is this?

[keen hollow]

Interesting. What's the workflow?

[wooden sail]

https://docs.comfy.org/tutorials/flux/flux-1-kontext-dev#1-basic-modifications

[keen hollow]

Comfy only so far?

I'll have to check on swarm support

[wooden sail]

I believe swarm has comfy under the hood.
Btw another example:
Draw this character serving drinks in a tavarn

[keen hollow]

Yes it uses the comfy engine by default, but native integration still has to be built for new stuff. You can go manipulate the comfy canvas manually but, at that point, might as well just use comfy.

[tawny wasp]

Yoruichi Shihouin is coming

[sharp imp]

interesting interpretation of off angle camera lol

[indigo oyster]

Because of a long time spent in blender, I once wrote "looking at camera" in my prompt instead of "looking at viewer", and got 10 images of my character working on a movie set instead💀

[sharp imp]

lol yeah, i should have put off center composition or dutch angle. oh well, wildcards be wildcardin

[nova remnant]

I see you liked my character lol. She was made for a chatbot OC :P a wizard

[stable crane]

VLC-chan

[tawny wasp]

Yoruichi Shihouin (Bleach) illustrious Apex_Ai
https://civitai.com/models/1729133?modelVersionId=1956921

[tawny wasp]

Wattson (Natalie) From Apex Legends illustrious Apex_Ai
https://civitai.com/models/1731773?modelVersionId=1959976

[tawny wasp]

Sui-Feng is coming 🙂

[keen hollow]

We need more of those Aerith & Tifa combos.

[golden vessel]

https://www.instagram.com/reel/DLjVyooSwA8/?igsh=NGFlODdrenNld2p1

finblan_ava

Camel bells enter Chang‘an
#aiart #art #artists #creativity

Instagram

[merry shale]

I love this

[noble abyss]

need car images

[indigo oyster]

https://tenor.com/view/low-tier-god-banned-mods-smite-gif-4866200577432974147

[lyric zealot]

two different scams at the same time

[sharp imp]

go to bed bot spammers

[cold nebula]

i love the details. what is ur prompt here??

[misty violet]

love the fierce yet somewhat innocent looking aura they possess man,,

[sharp imp]

Thanks lol I'm trying a new merge method but someone in another discord shared an image with a similar angle and depth and I wanted to test it

[lyric zealot]

i will help him reply. It's a nai prompt so here goes: 0.5::nagishiro mito::, 1girl, 1::sepia, sepia toning::, old photograph, photo (object), fox girl, fox ears, 0.2::red eyes::, white hair, shrine maiden, miko, blue sky, clouds, mountains, signature, 2::ofuda::

[primal wharf]

The prompts should be in the image, and I used NovelAI. The style is sepia and nagishiro mito

[primal panther]

https://cdn.discordapp.com/attachments/1142150111032512654/1381006862350028810/image.png?ex=6866e7dd&is=6865965d&hm=3cc145313f6ce6f9531de06289dbb0b1c3bf53105057f6a0d1dba30ca808761c&

[merry shale]

1

[misty violet]

kawaiiiii,, so innocent looking gurl

[frail bough]

in:#tech-support

[merry shale]

really good.

[misty violet]

this is another top art man! simple yet elegant looking imageee 👏

[keen hollow]

Happy Independence Day 🙂

[steady grail]

https://tenor.com/view/independence-day-white-house-attack-explosion-target-gif-12077937620217039154

[merry shale]

wow

[nova remnant]

Why are you taking Rina Tennoji and turning her into this? Damn imma stop posting my crap here

[wooden sail]

Everyone, stand up!

[steady pumice]

bruh you gave her scoliosis

[tawny wasp]

Pulchra Fellini V1 (Zenless Zone Zero) Illustrious Apex_Ai
https://civitai.com/models/1750350?modelVersionId=1980942

[small sable]

siguoii

[merry shale]

lol

[glacial coral]

Simplistic I like it

[soft parcel]

beautiful！

[tawny wasp]

Anemone (Kizetsu Yuusha) illustrious Apex_Ai
https://civitai.com/models/1757394?modelVersionId=1988911

[finite nova]

Ok

[woven gust]

mana

[nova remnant]

I just love making AI shit posts

Ever since I could finally have 2 different characters in the same photo without excessive inpainting/control net life has been good

[woven gust]

How did you do that

[nova remnant]

I'm using waiNSFWIllustrious_V140 it has training data for a hell of a lot of characters and most don't require you to do more than write their names.

I believe that this causes less issues with characters mixing up eye and hair color and stuff cuz you're only using their names as a trigger word

while the model says NSFW I've been using it to generate SFW content and it's fine. You can add (General) to the positive prompt to enforce it as well

[quasi shadow]

Tysm

Really enjoy the double mountain

😶

[fickle oyster]

Mind if I give this a quick upscale for you?

here you go!

uh

[misty violet]

hey man have u tried mixing ai gens to generate photos like for enhancement like using comfyui with other tool? like pykaso, savro,,etc

[fickle oyster]

He uses comfy yah

[indigo oyster]

https://tenor.com/view/low-tier-god-banned-mods-smite-gif-4866200577432974147

[steady grail]

https://tenor.com/view/pussy-cat-uncensored-funny-gif-22745274

🤣

[desert island]

What are you doing for your hands?

[topaz juniper]

"casting spell towards viewer", not much just a illustrious model

[fickle panther]

Which one??

[topaz juniper]

oneObsession atm

[desert island]

thanks I'll look into it. I think I'll try starting with that model and then after my second pass refining specifically the hands segments with that model to see if it can fix my hands. Thanks!

[desert island]

wait so isn't one obsession a Lora for illustrious xl? Sorry I'm on civitai trying to find the models to make this work and want to know if there is an actual model or if I'm understanding this correctly.

[topaz juniper]

checkpoint

not lora

[desert island]

interesting, okay, I didn't see it on civit let me look again. thanks again.

[desert island]

nevermind found it... weird I don't know why the search was pushing me towards the Loras 😄 thanks for the assist!

[topaz juniper]

https://civitai.com/models/1318945

[stable crane]

it knows Falin's special form

[desert island]

Thanks I'm gonna have to work on it more to make it work for me but yeah better hands! Thanks!

[wooden sail]

"On the right track"

[desert island]

Ah yes, the Darko Void piano. Lol. 😆

BACK TO WORK! MAKE ME MORE MUSIC! 🎵

[near elbow]

visiting very infrequently lately, enjoy

[desert island]

Wanted to try some more iterations with this character. It's not a perfect recreation of her but it's okay I guess.

[desert island]

MIku! 🫶

[desert island]

MadMiku: Kawaii Road

[near elbow]

getting into the mood

[desert island]

Miku would totally slay in any environment!

[tawny wasp]

Jean Grey (Marvel Rivals) illustrious Apex_Ai
https://civitai.com/models/1784427?modelVersionId=2019487

[acoustic night]

/animate prompt:"Michael Myers slowly walking forward on Halloween night, dramatic cinematic lighting, glowing jack-o'-lanterns in the background, subtle fog, horror atmosphere" motion:zoom-in style:cinematic

[desert island]

https://tenor.com/view/halloween-pumpkin-simpsons-gif-26847549

[acoustic night]

This dog sit on plan

[drowsy quail]

#🍥｜anime chair memetoken

[tawny wasp]

Finished 🙂 My Symbiote Squirrel Girl

[desert island]

Knull wouldn't stand a chance!

[tawny wasp]

SSJ3 🙂

[tawny wasp]

Symbiote Squirrel Girl (Marvel Rivals) illustrious Apex_Ai
https://civitai.com/models/1791176?modelVersionId=2026988

[near elbow]

the 3rd image is super explicit and therefore against server rules, just saying... i'm not a mod

[tawny wasp]

ok,thx a lot

[near elbow]

some evil cat mechas for your enjoyment

and back to cute

[desert island]

This was a fun character but I think I'm gonna make someone new to play with now.

[keen hollow]

Took a week off, haven't been on discord in a while. How's it going?

[nova remnant]

It'd be tricky if someone were to create an AI video of an anime girl dancing under the moon at the park.

[desert island]

why would it be tricky?

[wooden sail]

More details on how it's done on StableDiffusion subreddit:
https://www.reddit.com/r/StableDiffusion/comments/1m6glqy/hidden_power_of_sdxl_image_editing_beyond_flux1/

[woven gust]

kino no tabi

[near elbow]

isn't the new hidream out, also with editing?

[desert island]

Okay last one of this style 😄

[desert island]

I noticed I was getting this weird haze and artifacting at the bottom of every image. I've since corrected it but I'm not sure what's causing it. (Solved by upscaling and cropping before it gets to the actual upscale model and nodes). Anyways, anyone know why this was happening? I created a work around (as I mentioned) but I wanted to know if anyone knew what the source of the issue would be.

[near elbow]

or cut the "heavy fog below" part from your prompt 🤪

[desert island]

I didn't have that in my prompt. LOL.

[desert island]

I love the text you throw on your images... they always make me laugh.

[near elbow]

Steps: 7, Sampler: Euler, Schedule type: Simple, CFG scale: 1, Seed: 3209810855, Size: 1152x896, Model hash: a40a0a48fb, Model: fluxFusionV24StepsGGUFNF4_V2NF4AIO, Version: f2.0.1v1.10.1-previous-669-gdfdcbab6
Time taken: 4.8 sec.
A: 13.67 GB, R: 14.25 GB, Sys: 15.9/23.9883 GB (66.4%)```

[desert island]

I like flux but I always gravitate back to sdxl based models for image generation. 🤷‍♂️

[desert island]

fair, and how can you deny such a salient point delivered by something so cute? Impossible.

one last vampire girl

[near elbow]

but i do see that sdxl has its uses, especially for more specific anime characters and tropes - flux is universal but that also means it's not that well trained on manga/anime details

(that was too much text even for flux - chatGPT could probably do it but is censored up to 11)

[desert island]

It (Flux) is definitely monolithic though. It has so many great applications. I use it for product shot workflows

[near elbow]

cool - and supposedly it also trains very well, have to try that myself one day (i did an embedding in SD1.5 back then but never since)

[desert island]

I use flux gym to train lora's on products and then put them in shots. My wife does graphics design and occasionally she asks me to do this for her clients. By occasionally I mean she asked me to do it twice.

[near elbow]

neat!

probably with flux kontext you could just modify the original product photo instead, tried that? really cool, rather new

it's like having a photoshop professional at your service

[desert island]

it's been eons in ai time since I've done flux stuff so not up to date with all the new stuff. I'll look in to it though thanks for the tip. 😄

[near elbow]

if you have comfy up and running, it's really easy - https://comfyui-wiki.com/en/tutorial/advanced/image/flux/flux-1-kontext

[desert island]

I'd have to move flux back over to my nvme and kind of just wanna play and create but yeah I'll go through this probably tomorrow when I get up. Always down to learn new stuff and tools that will help me. 😄 Thanks for the handy link! Opening up the tab so I don't forget.

[desert island]

I couldn't help myself! I wanted to play around with the concept of this character a little more. 🤦‍♂️ couldn't land on something I really loved but... yeah now I think I'm done. On to something new and different!

In case you're wondering I'm really trying to land character consistency between prompts without a lora and it's not perfect as you can see, but that's what I'm working on (I was changing art styles slightly between gens too just to see how it would handle it). 😮‍💨

[desert island]

SOOOO CUTE!

[steady grail]

yup ❤️

[desert island]

he looks so snug... it made me sleepy too. Night little kirby friend sleep well and dream of something delicious.

[steady grail]

the last one is my new desktop

[steady grail]

nighty night

[desert island]

it's a good one

[empty pelican]

Padding or adjusting margins might help maintain the integrity of the image borders.

[desert island]

basically what I did with the upscale then cropping it just shaves the bottom off to solve, but yeah I was trying to find the root cause... like is something creating empty latent space under the image? Thanks for the reply though 😄 cheers 🍻

[compact sage]

Just joined this server and dont know of this is the right place to post...sorry if its not!

I have a bunch of loras for a game that I play (anime visuals), but i wanted to make them look like real life. is there any illustrious based checkpoints that would help me accomplish this?

[near elbow]

no idea about illustrious... you can run images through e.g. flux kontext to transform the look, though

random image i just generated, anime style

[compact sage]

so the loras i was going to use are all illustrious/pony based. is there a way to transform them to work with other checkpoints? i just know the basics of SD

[near elbow]

i doubt that, but just try it out, what could happen...

here's the result for flux kontext dev with "transform the style to a realistic photograph"

"transform the style to a baroque oil painting"

"put a grinning cat on her head"

well you get the idea 😉

[compact sage]

Is flux Kontext only on comfy? I use forge...

[wooden sail]

wtf

[near elbow]

yeah, unfortunately... i have comfy running for e.g. flux kontext and hidream even though i prefer forge by far

the gain with flux kontext outweighs the pain with comfy, at least for me 🙂

[desert island]

So what is the end goal? You can basically use the anime produced image and use it as noise for the next run but increase the denoise value or decrease for more realistic or more anime as desired... does that help?

In my particular workflow I use 2 models (illustrious one obsession) & (juggernautxl). typically the workflow goes like this:
create a brand new image using illustrious
2nd pass denoise at around .35-.45 on the illustrious image using juggernaut.
Then onto the upscaled image with illustrious doing the face detailing on that image (this leads to a very realistic body and background with a pasted on face that looks anime)
3rd pass brings it back to more anime to smooth out the final product but you could switch out any of these parts and renoodle with the concept so that it denoises in the way that works for the realism you are looking for.

Hope this makes sense (again comfyui but I'll share my workflow if you want it) additionally it makes character consistency pretty easy with zone based prompting and setting itself up for modularity should you choose to go that route (my intention is to use griptape so I can have an LLM work on the prompt in a repeatable way).

[near elbow]

cool results, quite an effort - this is "transform the image to a professional fashion photograph" with flux kontext based on the original image

and this is "transform the image to anime style flat 2d cel shaded 1980s" for a more vintage look

and this is "replace the background with a neon tokyo side street at night" 🙂

almost feels like cheating 😉

[desert island]

I'm sure it's way faster then doing it with comfyui just using kontext to do all the work... 😄

[desert island]

I love the joy of noodling something together with comfyui. To me it's like playing with legos. Half the joy is putting the blocks together and then tinkering with the settings to get what I want. time consuming yes but I can tool it out the exact way I want.

[desert island]

kontext definitely keeps better character consistency in the gens! this really illustrated that for me. Thanks.

[compact sage]

sorry, im really a beginner so what you said is beyond my understanding. im an ex-1111 user, current forge user who just knows how to enter prompts and download models from civai.

I do wanna learn though.

how do you "2nd pass denoise"?

[steady grail]

when they pass with the second serving of noise, just answer a "no, thank you seer!"

[compact sage]

is there a video guide that you recommend that teaches this concept?

[steady grail]

it's not so complex. what gui are you using?

[desert island]

let me just make a quick video that shows you what I'm talking about in comfyui. I also would recommend comfyui because it let's you understand what is going on behind the scenes more and understanding concepts. I'll get back to you with a link in a second.

[steady grail]

oh, that's so charming

[steady grail]

don't forget adding cats to the video lol

[desert island]

will do

[steady grail]

cute things are justice lol

[desert island]

OMG WTF! I had my mic off AGHHH!!!!

[desert island]

screw it. It'd be faster just to do it in the lounge if and when you actually want me to walk through it with you. plus you can ask questions live and we can actually have a discussion rather then me recording something and walking you through it that way.

[desert island]

Also here are some cats....

[compact sage]

is it possible if I cant talk? my mic has been broken for a while

[desert island]

sure should be fine 😄

[compact sage]

im on

[desert island]

one sec

[desert island]

Catgirl fight!

[compact sage]

so i actually managed to use kontext dev. gave amazing results, but any way to make the skin look less plasticky and more life-like?

[desert island]

prompting? skin details, skin textures maybe?

[compact sage]

For reference, here is the original image

This is use flux kontext dev Q8

Its great of course for what it does

Im just looking for ways to improve on it

my prompt was "Transform into realistic photo"

I read online that reduce the CFG scale to 1.8-2 improves skin quality, i did so and the above was the result

I am using this on forge by the way

[desert island]

try adding skin details into the prompt... also the image might have too much cleavage for the channel might want to remove them or hide them. I don't know the exact rules or how the mods rule on this stuff.

[compact sage]

woops sorry

[compact sage]

there was another image using the same realism prompt, but i added to make the character look more aged/mature looking

it produced a completely unrelated image

[desert island]

yeah I mean it can be hard to get that real photo-quality I think you're looking for. It's a rabbit hole I've gone down a few times, I've survived the journey but I've been forever scarred by chasing something that's just always a little out of reach.

[desert island]

pm me the image and when I get back from the store I'll see what I can do with it on my end.

[compact sage]

sent, thank you! I tried to do a bit of research but almost everything is comfy related

saw a video about using some comfy layer extension to mask the skin

and go from there

but that extension isn't on forge

[compact sage]

Question btw...can I add LORAs straight into the flux kontext prompt? for example

"transform into realistic photo, [insert lora]"

[desert island]

with a flux specific lora sure... don't know how forge does it but it should be simple enough.

[steady grail]

nowadays forge is very outdated, or unstable

[near elbow]

... but a bit more support of recent models and technologies would be nice.
all fine with flux though, and even chroma

[steady grail]

it's more like a developer's testing ground

[near elbow]

well, i don't see too many alternatives... comfy is inconsistent and nasty, and point solutions (e.g. framepack studio) bring their own infrastructure, which uses more space and makes common resources harder. it's 80% forge with some flux variant for me, and 20% comfy with flux kontext or hidream or the like

should be "works for you", flux and longer text requires a bit of luck

[steady grail]

try swarm, gui support for comfy, it's pretty much like auto1111

check the boxed and set the parameters

support the most popular custom nodes, so it's a good set up for who don't like the noodle puzzle

many people don't like the noodle challenge, but i'm all into it lol

[desert island]

I ❤️ noodles!!!

[desert island]

I LOVE THIS SO MUCH! AMAZING!!!! really cool character design!

[wooden sail]

||```
1girl, <lora:Smooth Booster:0.7> turnaround, upper body, side view, fantasy, (redum4, hata4564:0.8), masterpiece, best quality, [:aged up, multicolored hair, breasts, chiaroscuro:0.2], red eyes, contrapposto, rating:general,
magical girl,
Negative prompt: cleavage, armor, full body, [:sepia, text, skinny:0.2], full body,
Steps: 28, Sampler: Euler a, Schedule type: Automatic, CFG scale: 5, Seed: 1990436563, Size: 1216x1024, Model hash: 789461ab55, Model: waiSHUFFLENOOB_ePred20, ADetailer model: face_yolov8n.pt, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer mask blur: 4, ADetailer denoising strength: 0.3, ADetailer inpaint only masked: True, ADetailer inpaint padding: 32, ADetailer use separate steps: True, ADetailer steps: 18, ADetailer version: 24.6.0, Lora hashes: "Smooth Booster: 2e1d7277a565", Version: f2.0.1v1.10.1-previous-636-gb835f24a

[desert island]

I tried to recreate her, in my own way... I couldn't quite get the vibe or energy your original had. 😢

[desert island]

UGHHH! 💢 I give up!

[wooden sail]

Try Flux.1 Kontext

[desert island]

fine I give... I'll try flux kontext but... I only have 16GB of vram doesn't that mean I have to use the GGUF version. I need a little more wiggle room for my LLM model.

[wooden sail]

Try Playground on the official site first. It gives you like 50 free generations.

[desert island]

I don't like genning outside of my local. Regardless I'll try it with the GGUF version.

oooh so many different quants! https://huggingface.co/bullerwins/FLUX.1-Kontext-dev-GGUF/tree/main

https://tenor.com/view/grinch-gif-7196328019405537697

[tawny wasp]

Wendy Corduroy (Gravity Falls) V1 illustrious Apex_Ai
https://civitai.com/models/1806554?modelVersionId=2044421

[desert island]

It does feel a little dirty, just how easy it is...

[desert island]

I can now tweak my images and characters and move them through various states of realism, or whatever the hell I want! thanks @wooden sail and @near elbow 🫂
(anime, realism, digital art)

[near elbow]

yeah, it's like having a masterclass photoshop expert at your service - really cool
"transform the style, replace thisandthat, add a background, make her sad, make her hold whatever" all just works

[desert island]

it opens up a lot of new capabilities for me plus I can still just use it as latent noise if I need other models to do some work for me. Or shift the image to be used as noise in kontext as you do normally. It's great!

[desert island]

Original(Juggernaut Illustrious Hybrid)->Altered concept(Flux Kontext)->noise to (illustrious) for final polish

[wispy canopy]

thotslayer!

[near elbow]

countermeasures from a distance, incoming 🙂

[wispy canopy]

who remembers the 2023 days when SD1.3-1.5 was just released? o7

[tawny wasp]

Beauty 🙂

[near elbow]

yeah, stable diffusion beta participant here - although they killed all the old channels here (or are they just invisible now?)

[wispy canopy]

looks like they archived/deleted it?

[near elbow]

i have several roles with "archive" in them but cannot see any

probably deleted, there was a lot of NSFW in the early days from folks experimenting with prompts like "succubus" 😇

[wispy canopy]

i snuck in one alphonse mucha nudity that never got noticed xD

but shadows covered the nudity

but eh.

[near elbow]

those were the days...

[wispy canopy]

they were. if they didn't go full prude, this channel/server would be so alive

the moment you ban nsfw -> that's it. your guest list is going to be empty sooner or later

[near elbow]

well, investors and other mythical beasts... and mastercard coming down hard on NSFW lately as well. interesting times

and now for an...

succubi with sweaters are a bit strange (but fully SFW)

[wispy canopy]

should be sfw, but spoilered just in case

but i agree. the moment you sanitize an online space -> any form of real interest will be just gone

the mastercard /visa powergrab is a whole different political story which i have a very specific oppinion about, but that's just a cartel / duopoly powergrab that isnt'opposed by any form of actual lawmaking

[near elbow]

well, in the current state the lawmaking (or more, the presidential orders) are a topic of their own 😕

crazy times

[wispy canopy]

i don't even go usa, fckn puppet show

if you want US politics summarized: "nothing will happen"

it's all just a scapegoat show and the majority buys it. "everybody is (un)happy"

but i digress. this is an a channel for waifus

and husbandos

[near elbow]

voilà

and cyber neko killer ninjas

[wheat falcon]

it was 2022 😛

when the models standard size was 7gb, and 512x512 was very slow xD

[fickle oyster]

Talent plz

skillz plz

I see so much not being posted

I seeeeee it

[left wind]

I'm new to this any recommendations would be appreciated

[near elbow]

😎

[near elbow]

well, the first feedback would be "don't go crazy with the upscaling, for discord posting around 1024x1024 is enough" - your >40mb PNGs don't even get displayed here since they are too big 😉

[near elbow]

cool to see others using images instead of text (or yet another meme gif animation 🙂 )

[tawny wasp]

Anemone version 1 is coming soon

[tawny wasp]

Anemone (Kizetsu Yuusha) V1 illustrious Apex_Ai
https://civitai.com/models/1817543?modelVersionId=2056794

[near elbow]

that might be considered spam since it has commercial intentions

[near elbow]

on the hunt

[keen hollow]

Tried getting some more unique designs today, but it was a struggle

Id like to get something a bit more ... Jagged and poppy... Almost spiderversey

I know those aren't words 😝

[desert island]

but they should be!

[keen hollow]

Trying to get something sorta like this. (This is not generated, but I think is by Sean Anderson)

But like.. not sonic

Just that kinda hyper emoted, bright almost holo/iridescent, perspective, style...

[desert island]

IP adapter with style transfer? Let me try something.

[near elbow]

"anime style waifu jaggy poppy spiderverse themed" 🙂

[near elbow]

"anime waifu, hyper emoted, bright almost holo/iridescent, perspective, sean anderson style, action scene"

[keen hollow]

not bad, but needs to be a bit more abstract

[desert island]

🤷‍♂️

[near elbow]

"abstract minimalist cubist flat 2d cel shaded, anime waifu, hyper emoted, bright almost holo/iridescent, perspective, sean anderson style, action scene"

(all with flux fusion)

[keen hollow]

lol norio knew to do a Tifa... 😄

[desert island]

I mean what else would you do? I'm assuming you're going for a slot machine type background too?

[near elbow]

tifa you say? 🙂

the cel shaded look becomes her

[keen hollow]

looks kinda streets of ragey

do I do a lot of slot machines? is that a thing? lol

[near elbow]

"mad max atmosphere" added 🙂

[near elbow]

no such thing as too many of them

[keen hollow]

ya'll need to hone your Tifa skills.. these are like Tifa's 2nd cousin 🤣

[near elbow]

but they can spell thanks to flux 🙂

[desert island]

I thought tifa's limit break had something to do with a slot machine... it's been ages since I played ff7.

[near elbow]

the whole current prompt is "headline text "Omnia Burger" greeting card, abstract minimalist cubist flat 2d cel shaded, anime waifu tifa female, hyper emoted, bright almost holo/iridescent, perspective, sean anderson style, action scene slot machines" so i'm amazed that there is anything of tifa visible at all 😉

[keen hollow]

Yeah Sean Anderson is pretty close to what I was trying... https://www.instagram.com/skyscraper48_art/?hl=en

That perspective shot looking down a city street with the character jumping off that page at you...

[keen hollow]

wow the scams are getting a bit more annoying eh?

[desert island]

they come in waves

[keen hollow]

ok fine geez ill spin up an instance and try some stuff again...

that isn't bad

mwhoo ha ha ha ha I'm getting very close to where I want to be....

Need to get better quality in the little enemies...

ignore the extra shoe 😄

hmm those are a bit more epic in widescreen 😛

[keen hollow]

that's kinda neat, has an arcade game feel

[desert island]

I'm gonna keep trying a couple more things cause it's fun. 😄

[keen hollow]

yeah I get hooked on it

and now I just rent 4090s

can't go back lol

[desert island]

My 4080 is fine eventually I'll want to upgrade but I like my system 😄 RGB desk beauty.

[keen hollow]

I have a 3060.

But once I started renting the 4090 I couldn't go back. 😄 And frankly, part of that is just because I don't have to manage a bunch of sketchy garbage on my own pc.

[desert island]

I know it's not the place for it but I never get to show it off. 😆

[desert island]

stunning!

[keen hollow]

Yeah I built a new 9800x3d machine... I should probably get a 5000 series video card. I just don't want to give nvidia their garbage prices. 😄

Yeah I'm getting some pretty cool ones. I liked that one.

[desert island]

5070ti in my opinion is best bang for buck but then the 5090's obviously are god tier amazing if you got an extra 3grand you can shovel over for a hobby. (Not me).

[keen hollow]

I generally post non-upscaled versions here because of the 10MB limit, fyi.

I can afford the card, but I do not think it's a wise way to spend 3 grand. 😄

Just not gonna do it.

Hoping for some 'super' variants with more vram.

But at this point, my AI workloads are going to be in the cloud. It's just too easy, cheap, and keeps my own pc clean.

[desert island]

💵 <- 🔥

[keen hollow]

cats

what's she laughing at?

[keen hollow]

ok I gotta stop... I can spend all day doing this

[tough matrix]

Faces

https://cdn.discordapp.com/attachments/720987986241781793/1399432483170160650/ComfyUI_01247_.png?ex=6888fa90&is=6887a910&hm=ba266c770e8d61c14afb59ebb45989a48bebad74f91df1895a6998e17dc83e48&

[keen hollow]

Faces?

[tough matrix]

The zombie faces look off. Have you tried running detailer on them?

[keen hollow]

Ah yes, I didn't focus too much on background detail.. was chasing a specific style

[tough matrix]

What kind of prompt for these angles and poses?

[tough matrix]

Just referring to the dynamic angle

And do you run any face detailer for the main characters face?

Since it seems better than the zombies

[keen hollow]

The word you want is "foreshortening"

I did not run any detailer/inpainting for this run today

[tough matrix]

Do you use comfyui?

[keen hollow]

I do not. I used SwarmUI for these, with the comfy backend engine.

[near elbow]

a nice thing about a 4090 is that it happens to run cp2077 with max details and raytracing at >60fps, that's dual use for me 🙂

cats you said?

[tough matrix]

nice foreshortening is really nice

[tawny wasp]

Sui-feng / Soifon (Bleach) illustrious Apex_Ai
https://civitai.com/models/1820488?modelVersionId=2060114

[keen hollow]

Yes it comes in handy. You can specify what you want foreshortened with some degree of success, and utilize the image dimensions to frame accordingly.

Also, that poor lady is going to have back problems.

[tough matrix]

its ok she goes to the gym

[desert island]

almost cool... 🤦‍♂️

[placid grove]

hi

[toxic lotus]

I've been dead

But I'm back in ai again

What are some good models that I've missed out on 🙏

[wispy canopy]

"hey, you're in the wrong game..."

[keen hollow]

Ha, yes. I'm playing ffvi for the first time currently.

[wispy canopy]

oh enjoy, that one is precious!

[keen hollow]

Yeah I've always heard it's one of the best

[wispy canopy]

absolutely

also play chrono trigger if you haven't done so already

[placid grove]

hi

[terse ore]

Wan 2.2 is the big new thing.

[near elbow]

flux fusion for all your needs, unless you want something very specific 🙂

not really new but still great

[glacial coral]

Need to try can I get a link

[wispy canopy]

flux krea is pretty cool

yeah, it's fun ^^

[left wind]

You guys have some cool styles

[tawny wasp]

Ember Lumen

[tawny wasp]

jentry chau

[near elbow]

https://civitai.com/models/630820/flux-fusion-v2-4-steps-gguf-nf4-fp8fp16

[indigo oyster]

this has strong Lloyd energy xD

[indigo oyster]

(source: greatest estate developer)

[toxic lotus]

Mmmm I downloaded a flux model but it didn't work, came up with some error I guess I did something wrong 😅

[terse ore]

How do i get the res_2s sampler and the beta57 scheduler?

[bold ether]

Dont go there, it s a scam

[near elbow]

what do you mean? the civitAI link is 100% legit... @toxic lotus have you read the instructions there? the specific model you want to use differs depending on your setup, e.g. i'm running fluxFusionV24StepsGGUFNF4_V2NF4AIO.safetensors with forge out of the box

[bold ether]

I was talking about the """""external discord supp"""""" message that I deleted.

[left wind]

Pretty sure it depends on your GPU try a smaller one like the V2 fp8 file instead of the 20+ some GPUs doesn't work on bigger safetensors files I believe it's down to VRAM 🤷‍♂️

[near elbow]

ah, sorry, all fine then 🙂

[wooden sail]

Is it?

[steady grail]

arasocial relationship robbed her of her left eye!

what a bane indeed!

[near elbow]

within a suitable definition of "fine" 😉

[near elbow]

with apologies to the meme

[fickle oyster]

Hello

[keen hollow]

hi

[terse ore]

I think I'll save up for a 3090 PC...there are hardly any on Craigslist or Facebook marketplace here which is where there would be a deal if there's one had to be found. Maybe ebay? 🤔

[left wind]

woldnt a 5070ti or 5080 be cheaper

[terse ore]

Idk but the 3090 has 24gb vram

[left wind]

yeah true

[fickle oyster]

whatcha gonna do with all that vram?

[left wind]

i use a 5080 its decent but doesn't do everything due to VRAM so 3090 would be a better shout for rendering but if you game too 5080 would be the better option id say

[terse ore]

Wan2.2

[fickle oyster]

fair...... But it will be unbearably slow

[terse ore]

Luckily I dont game but yeah. I picked up a 3060 a couple of years ago on the cheap which'll help pay for it I guess

[twilit portal]

🏯 Samurai Ayaka ⚔️

[left wind]

dont fall for it people its fake im not sure why so many of them are on here ahah

[near elbow]

different style, again

[tawny wasp]

Houshou Marine From Hololive illustrious Apex_Ai
https://civitai.com/models/1838670?modelVersionId=2080711

[rough bay]

anyone know any good lora for blurring background. i feel like if i blur the background my character will be the focus more. i tried searching but most are sd 1.5 or for realism would appreciate any help xd

[wheat falcon]

Anyone here knows how to get IP-Adapter working with Illustrious/noob on auto1111/forge ?

[fickle oyster]

AM I IN TROUBLE ALREADY?!

oh lel

[fickle oyster]

afraid not

[wooden sail]

Try this - follow the instruction

[wheat falcon]

Thanks!

[wooden sail]

IP-Adapter is image prompt. I prefer stylistic transfer:

Using controlnet Reference

[slate remnant]

What is the name of that Twin tailed blue haired girl everyone likes?

[tough matrix]

wow nice! how?

[steady grail]

yume kawaii, very awa, narue, traditional media, (watercolor (medium):1.2), chon (chon33v), soft style, pastel colors, spot color, limited palette, (dynamic_angle_shot:1.385, dynamic pose standing:1.385), Disgustingly Adorable,
BREAK
1girl, pink cirno, touhou, kirby hood pijama cosplay, double v sign, ><, :), dancing silly pose, pink dress, sunset, public of people cheering cirno, fantasy background
BREAK
official_art, very awa, masterpiece, best quality, newest, absurdres, highres, amazing quality, very aesthetic, high resolution, ultra-detailed, absurdres, newest, scenery,depth of field, volumetric lighting, (simmetrical:0.3)

[tough matrix]

what is the model?

@steady grail

[steady grail]

https://civitai.com/models/1170176?modelVersionId=1897747

[tough matrix]

yo nice model!

[steady grail]

https://cdn.discordapp.com/attachments/1091193032273043516/1397494006040105000/ComfyUI_temp_zqprs_00003_.png?ex=689310b7&is=6891bf37&hm=09ecd607f68eba549543f76f523f5f29f2f9d339616f5ee70c7bd545e2934413& https://cdn.discordapp.com/attachments/1091193032273043516/1400055948931760128/ComfyUI_temp_apeer_00004_.png?ex=68932836&is=6891d6b6&hm=1b755bbe3ff3d0bdaf459d27ac580697385b08ba5aceb6bfe4947340c36e45c9&

[steady grail]

same model

[tough matrix]

jsut trying it. what an artistic model

[steady grail]

yup

illustrus models are very good

[tough matrix]

but not all illustrious modesl are the same

for my stylistic preferences, there are some better than others

[steady grail]

yeah true, i look for finetunes and lora merges usually

[tough matrix]

ill share with you sona in a few differnet models

[steady grail]

oh good

[tough matrix]

One Obsession (Mature version)

One Obsession (Default version)

Smooth Mix

Diving

@steady grail for my preference, I like one obsession mature and diving

[steady grail]

default and diving for me

[tough matrix]

I think obsession mature and diving are quite similar

diving is more flat and matte

[steady grail]

they are all pretty similar, the difference it's only in the fine details

[tough matrix]

two are realistic, the other two are more cartoon IMo

[steady grail]

cute

[tough matrix]

diving is just so good effortlessly

[fickle oyster]

arghhhhhhhhhhhhhhhh

[bold ether]

two asukas for the price of one.

[green quest]

hatsune miku? she's more teal hair than blue hair but

[glacial coral]

Soooo what new models has u lot been trying out anyone tried SIH 1.5 out yet?

[fickle oyster]

huh?

What is

[gleaming flame]

diving semi real is good for typical anime too... from what I see it's a bit smoother with thinner lines

[indigo oyster]

https://tenor.com/view/low-tier-god-banned-mods-smite-gif-4866200577432974147

[near elbow]

back in the days the spam didn't survive for more than a minute...

[glacial coral]

https://civitai.com/models/1217645?modelVersionId=1371790

click on recent model

[fickle oyster]

...It's a... v-pred / EPS hybrid?

noob eps finetunes -> vpred + noob vpred finetunes -> various incest maneuvers

1.5 - 1.0 + various finetunes, illust 1.0, illust 1.1, animagine 4.0, unreleased noob vpred

...

I'm literally lost.

[glacial coral]

Pretty much tho it works on v-pred mode im pretty sure as on the mac UI it didn’t work with Eps only V-Pred

[fickle oyster]

I mean, I also make checkpoints etc and this concept and idea is absolutely mental.

But if people like it it is what it is and you can't go wrong with subjective matters

[glacial coral]

The individual is indeed mental with what they do with models but it seems to cook nicely

[tough matrix]

@steady grail what kind of settings do you use? sampler scheduler? cfg?

[steady grail]

euler a cfg pp + beta

hyper sdxl 4 step

[tough matrix]

thats the upscaler?

[steady grail]

8 step + cfg 1.7

nah

for upscaler i use omnisr x2 div2k

[tough matrix]

cfg 1.7? 8 steps only

wow thats small

I'm using dpm 3m sde + karras with 30 steps as a base and i love the look

[steady grail]

hyper sdxl 4 step is a distiller lora

reduce the necessary steps and cfg to gen

making stuff faster

[tough matrix]

what is the drawback from using this?

[steady grail]

some small loss of quality

in my opinion, it's wort it, on xl models

[tough matrix]

I make multiple passes in my model, as well as run multiple detailers

wont the lora infect the entire chain

[steady grail]

i use comfy, my chain is a little complex

[tough matrix]

me too

where do you load lora? after checkpoint and before ksampler right?

you have to attach the clip from lora to ksampler

[steady grail]

i just load model only for hyper

[tough matrix]

i see

so you seperate your workflolw

to be honest, image generation takes only like 15 seconds

i dont think i can use that hyper speed up for my second pass or detailers

[steady grail]

well, i needed something instant to img2img constantly while painting, at least that was the project at first

then i just found it more efficient like this

[tough matrix]

this sounds awesome

can you share more

[steady grail]

if you use comfy, link it with krita

it's very comfortable while img2img while hand painting

https://github.com/Acly/krita-ai-diffusion

you can use the workflows set by him, or load an handmade workflow

with such speed, just play with krita. it's very funny, even with default workflows

[glacial coral]

Got a link real quick?

[steady grail]

https://huggingface.co/ByteDance/Hyper-SD/tree/main

[glacial coral]

Cheers could I use HiRes with this or is more of the upscaling stuff is done in img2img with this?

[steady grail]

it's not an upscaler model, it's a lora used to lower the steps necessary to gen

but you need to lower the cfg

[glacial coral]

I said could I use HiRes with it not can I use it for HiRes

[steady grail]

yeah, sure