#🍥|anime
1 messages · Page 264 of 1
Whatcha up to?
clipseg works pretty well for more realistic styles. it struggles with anime though Anzhc Face sec 1024 v2 y8n.pt from here is very good for face detailing https://huggingface.co/Anzhc/Anzhcs_YOLOs/tree/main
it segments just the face like clipseg would, but consistently works for both anime and realism
safe version?
the ones marked suspicious should be good. for comparison here are the official ultralytics models https://huggingface.co/Ultralytics/YOLOv8/tree/main
Still would prefer a safetensors version 🙂
idk that i've ever seen a safetensors version of an ultralytics model. i'll have to see if i can convert them later
well, it claims to have converted properly. now to see if it actually works
nah, it fails to load. i'm guessing ultralytics models aren't compatible.
@green temple :<
ok getting better results since I changed compatibility back to none. oi
her legs are a donut too... or is it just leg at that point... 🤔
corrective surgery?
nah
then yes... I'll have a donut.
NO! it's mine! 💢 STOP! DON'T EAT IT! NOOOOOO!!!!
whats this model?
animagine but I'm using flux as a refiner and then running it back through animagine to bring it back to an anime esque vibe.... you want the workflow?
ah yes, if u're alright
@bitter latch still using forge?
yes
new update or the old one?
I was wondering is it okay to update now?
the newer one, tho hven't updated since the first update after the first update a month or so.. umm
I explain badly
I see, so speed is like the same?
still good
nice
I didn't clean it up yet since I'm still working on it.
now if my shoulder would quit hurting, not sure why it's hurting.. ow
that damn symbol keeps popping up . wtf?
shoot some of the settings were off I forgot to put some things back.... one sec let me put the right one up...
Okay this one... has the seed for that image too so it should just gen another one unless you change the global seed
👍
oh what flux model? Dev?
I'm using a gguf one ummm lets see... oh the unchained one 😆 https://civitai.com/models/662112/gguf-flux-unchained
stow-a-way?
I'm a little late ... but 😄
sailing away.. almost time for work.. again
oh okay okay
Should work with any gguf flux model though. One sec I'll verify.
it does although I might need to tweak some settings but I just tried with this model https://civitai.com/models/657607/gguf-fastflux-flux1-schnell-merged-with-flux1-dev
I got this result but I can see a problem that needs to be adjusted for this model. pretty sure it's the cfg scale if you care at all... like I said it still needs work 😄
Back to the aom days with that style
yeah it really breaks down at the upscale in the latent space when it's over on the flux side... the first flux pass is fine. 🤔 curiouser and curiouser... I wonder what's causing this.
Image order shows the first flux pass followed by the latent upscale. Normally with the unchained model it improves at this step but here with this model it's losing alot of fidelity. I'm vexed.
dang, accidentally deleted a bunch of old models yesterday
ah its fine I'll try with unchained one
thanks
yeah it's a bummer... a lot of early stuff that probably isn't really out there to get anymore. That said, unlikely I really needed it. 😛
more for my own edification. Just want to learn, so I'm glad I tested. Thanks, btw 😄
In computer terms I suppose it's okay to be a hoarder because HD space is relatively cheap, but at the same time accidental spring cleaning isn't too terrible either 😁
yep, it is what it is
because of the large size and transient nature, no backup of 'em
so they're gone.. whatevs

@supple raptor you guys used something to upscale nai images right? like something going with the API or smthing, got that?
doesnt cost anlas?
older models where so nightmare inducing with all the mutations
Not able to get flux dev GGUF or NF4 versions to work in Swarm. If ya'll know how, let me know. 🙂
do you have them in the right folder?
they need to be in the diffusion_models folder
funny stuff i canrun fux on my pc lol
maybe flux instead? just pulling your leg
nah, flux is to heavy, i'm running his Chinese knok off
funny thing, for how much high tech china is today, Chinese knock off could be even be better performing than west models lol
what a world, we can't use Chinese products as a joke any more 😦
i remember when it was still the "oh the japanese are taking our jobs away"... how things have changed.
did you know that "made in germany" used to be a mandatory label on imports from there to the US, to scare off buyers?
the only constant is change...
up next
what? when the wall was still up? lol
way before that
i can see why
"The British officials who coined the phrase “Made in Germany” intended it as an insult. In 1887, alarmed at an influx of low-priced German products, the British government required goods imported from Germany to be labeled as such. Back then, Germany was to Britain something like China is to Europe or the United States today. It was an aggressive emerging economy with a large store of cheap labor and ambitions to become an economic superpower. But Britain’s attempt to shield domestic companies from competition backfired. Made in Germany became a synonym for quality."
ok, not even the US back then 🙂
ehehe
too much corruption at the time
like with the urss back than
i vaguely remember something
by reading story, it's sad to see how things never change, but simply move places
yeah i used a workflow with these nodes to generate the base image over nai's api, then sent the output of that through local upscaling and img2img as well as face detailer. at the time i used kohaku-xl-delta as the local model because it was responded the most similarly to NAID3 when given the same prompt https://github.com/bedovyy/ComfyUI_NAIDGenerator
if you have an opus tier subscription you can enable the limit_opus_free setting in the main node and it will prevent you from setting parameters that would cost you anlas
oh thanks a lot

charming
I got a kick out of how realistic that pastry is... 😛

new version of noobai is out. seems even better so far.
I'm gonna wait for the final version, it's still not good enough imo
unless I'm completely missing a vae here
1girl, solo, loli, nahida \(genshin impact\), travel attendant, mini hat, white gloves, :d, white background, miniskirt, pencil skirt, (holding clipboard:1.2), (from side:1.2), (from above:0.6), black coat, black skirt, black pantyhose, white collared shirt, +++, looking at viewer, (presenting, cupping hand:1.2), clipboard, holding notebook,```
Or maybe I still haven't found the artist combination for noob
here it is with just adding quality tags to the end of the prompt
Cute, I think I'll also play around with the samplers a bit more
i was using dpm++ 2s a with DDIM uniform scheduler on previous versions, but i'm liking good old Euler A with normal scheduler on this version
i've never had luck with the CFG++ samplers. others seem to get nice images with them though
That 5th is ugly
runs away
Help me take down all those fake elves

@green temple hold the smol back! Before a knee is kicked 
She's in like a dynamic fight pose. And her... physique... is canon. : shrug :
... when you set up a set of 100 gens, walk away, come back later, and realize you locked it to 1 seed
screams
Get him zombie Miku!
my oc, idk why she is crying tho
nice fingernails

okay last ones for a bit.
oops
1024x 64 (not 640...)
uhh oh , miku horror
now thats random!
hmm someone turned it inside out?
or removed armor
sleeps
These are so cool! I love what you're doing today! ❤️ ❤️ ❤️
I'm loving the new noobai version. Can't wait to see how the finished vpred version turns out
おかえり
Fine, welcome back... shrug
I don't know what cnibo is but this marketing sings to me! TAKE MONIES! 💵 💳 💰
Rare Alice post
working with your image, just using it so I can figure out my flux refiner settings for the workflow I'm putting together but I saw this and thought it was cool so I wanted to share it with you.
Nubby is here
You might want to check out the rules
runs away
Doesn't bother me any, but those 🐫's are likely to get you timed out for a while when a mod comes looking through
Grabby 
no grabby
no cry
cute
someone grabbed her teddy...
awww, this made me so sad when I saw this... 😭 ... 😆 nice job
cute
now draw her gtting an education

grabs you

AAAAAAAAAAA
not allowed ( she will bite me ) 
she beet
good now draw her eating a hamburger
she doesnt like hamburgers


1st and 2nd try with a character who appears only 88 times on danbooru. accurate character and she's holding a bow that's nearly perfect aside from string placement in both. noobaiXL_epsilonPred075 is impressive. even NAI struggles to get a bow that's half as good.
Try i-no

i'll dm that one

My guy I'm giving up on Instagram reels addiction and you're giving me flashbacks
cringe
imagine getting addicted to reels and short form media content
Dopamine is crazy when you don't regulate it properly
because they're cute
true that
i do take pyschology as a subject
i can understand
yes but are you one of those people with an alt account who think otherwise
You can research short form content for bachelor or whatever other stage you're on
what?
nvm
I'm afraid he wants to point out illegal activities
At least that's what alt acc part leads me to believe in context of anime lolis
it's usually projection or not understanding that anime girls aren't real
It's so tiring bros...
Yeah! Shut up and gen stuff! 😆 j/k
It might be projection or it might that people like that are out there
Apparently they haven’t seen the generation yet @green temple and mine😁
hi everyone guys.
Mine interest is piqued, can you shine some light over that pretty please?
just search the channel for images posted
Can't type in from: cyrrylic letters
it's not like i'm the only one posting cute anime girls in a anime channel
or search by name using the search in the top right
How much do I understand once again the conversation about appearance on generations?
Ain't got Cyrillic keyboard to type in this guy's name
Just look up the person's nickname
funya_nyanodesu
Ctrl+C Ctrl+V ФуНя-Нянодесу
That's helpful
Phone
Come on be like funya and provide solution
Idk I just checked their profile
double tap hold your finger on their name to copy?
Some just do not deserve anime lol
Gatekeeping is good
Open profile tab, not copies the name lol
I was just screenshoting it lol 🤣
I don't agree with gatekeeping, everyone should be allowed to try new things and not be kept out.. but if they're going to be dicks it should be easier to get rid of them
I just like the easy way out lol
Not for me
Probably much simpler
But anyway a valid solution was already posted
Just tap on their username
glad it's solved 😆
eeeh... this is a slipery slope
imma just
leave now
What is this?
Illegal activities, mr varghoss!
I'm the cute girl police
Make her smug and :3

I'm sad cat now
Need all the cute images to cheer up
+1 
@modest fog i have a message for you
cat midlife crisis?
Yes
then - “Hugs”!!!
Seems interesting I guess I need to check my local at some point
why sad
@autumn violet
Coffee Cannon™
What model is that? Very pretty.
that lora is adding some interesting clothing and background details to my normal prompt results
I think you need like 2-3 strength on it
To get the art
this is 1.8 😄
the artist tags i'm using are probably pretty strong as well, but this is 2.5 weight on the lora
I have heard about the base for that. So far I've only tried out flux from some of the newer stuff. I need coffee cannon right now though, very tired.
illustrious is very good and the latest noobai version is even better
the actual style isn't coming through very strong, but i love what it's doing to the clothes, hair, and backgrounds
it have less senseless details
It's cute
I only tried on a realistic model, merge of pony, autism and realistic and pony itself
Except from the realistic model it worked well lmao
I had to try 
I’m using autism
Try some other stuff on prompt 
same prompt as my last image, but i took out the artist tags, lora weight 1.9
weight 3 😄

wait imma need the sauce for that one
the rest is in the metadata but
Prompt:
orenji \(wholesomeorenji\), 1girl, dot nose, blush, komeiji koishi, scarf, holding disposable cup, coffee, steam, cowboy shot, blurry background, outdoors, autumn leaves, falling leaves, best quality, amazing quality, very aesthetic, absurdres
Model: https://civitai.com/models/833294/noobai-xl-nai-xl
been out of the game, what's the current best anime model? flux? pony? MJ/niji? is 1.5 obsolete?
I use animagine but I'm not really the best person to ask... I'm sure other people will give you better answers as time goes on. 😄
realistically they all have their pros and cons, depends mostly on the style and content you personally want, if nsfw is important, if you have the hardware to gen locally, etc
some people are still using 1.5 stuff because it's fast and they have a model/lora combo that works for them, I like autismmix (pony model), flux is pretty good but I haven't tried any of the anime finetunes so i don't really know how it performs besides it being pretty heavy on hardware requirements
😄
Ooooh pretty!
Sooo crisp sooo clean! ❤️
thanks, I use the EPI ,DPO, More detail lora's and hypernetwork mjv4hypernetwork's in a style thats saved
then I mess up and get this
sigh
makes some adjustments
getting there 😄
interesting
would make a great poster on a wall
adds "art style of Lovecraftian mythos" for giggles
hmmm.. ok... umm weird
that's really pretty I love the colors!
thanks
hmmm fantasy borg?
ok I would watch this anime
I think it got confused
she's upto something
-turns off dtg-
😆
move move !, coming thru!
hehe changed from eldrich to walt disney mythos
cyber fantasy
hmmm I usually would be in bed but slept tomuch yesterday, but yawn... am tired
changes models
oooooh
what model did you switch to?
a merge of a RPG, sci-fi ish and Realtoon models
hmm cyber bat zelda?
hmm to .. realistic
kinda creepy
ok back to a more anime model
there.. better
Wow okay that's not bad... of course it takes 7 minutes to gen but okay....
Animagine->AnimagineUpscale->Flux+Upscale->Animagine Final Pass
Thanks for the reminder 😄 I should probably go do the same.
eating? yeah me too. I'm trying to quit but it's pretty hard 😄
no no the cyber fantasy sytle
Okay I'm done with flux in my workflows for awhile I'm not sure why but my VM crashes after x amount of gens even though it still has vram and ram to play with and I don't know why. To lazy to actually comb through logs to figure it out.
The error I got was 'too many gpu starts and restarts' but that's as far as I'm looking into it back to just playing.
still needs some refining work but I think I can get around not using flux if I modify some stuff. I think hands and distant shots are gonna be the bane of my existence though we'll see... gott gen some more to find the pain points.
i'm a player!
i didn't know i could genderize my stupidity. you opened my eyes ❤️
It's 2025

from now on i'm gender stuipid and when people call me stupid, they are simply stating my gender lol
taking pride in stupidity XD
We are both the same

I stupid cat chaos
I need a little bit of food like her
To grow more stupid

That's why the cat is depressed! 🤣
those are cards
@steady grail these are actual cards from the game

is a particular style? or is obtained by genning an image with low denoise?
its a style lora I made 
oh, it's cool
https://civitai.com/models/897923/path-of-exile-divination-cards-art-style?modelVersionId=1004752 if you want to play with it
Its for pony
the images are very polished, but a little papyrus grainy
i rarely use pony. too slow for my tastes and i'm playying with a new workflow
i'm delving in translations
i found a way to connect a translation model with comfy and i want to see if i can automatize some imputs
wow, comfy uploaded a lot of stuff
oops... 😄
she definitely did some kind of prank
I think she was trying to show off her piloting skill but then made a big mistake 😆. Now the whole bridge crew is yelling at her.
especially those who were in the WC at that moment 🤣
Isn't she Captain Kirk's daughter by any chance?
nepatism... smh...
anyways today i tried to draw without learning art theory by looking at different shit and kinda like meshing cant say went well
Try it 10000 times
you'll get gooder
How much of a first time was it
And using anything like a tablet
so i do copy stuff rite like putting the image on top and then copy
but like creation its first time
(i dont give a fat fuck about art but i want to be able to copy shit into my notebook at school so i do copy more)

shut up
how about i give u 30% but u give me 5k to prove u have 100k
working on sdxl version of my 1 year old yameroyandere lora, fun to be back making sd loras
someone wanna test my lora?
so cursed.... lol
im gonna take that as a yes
activation word is yameroyandere and its trained on NoobAI-XL 0.75 version
i don't have a pony model on pc
i said its noobai, which is not pony
Not at all worth if ya ask me
Yeah not for this but it has other applications 😄
like? (not being rude, i actually want to know)
flux has a problem with composition alot of times but is really good at details. Flux ends up fixing hands or other details that most anime models have problems with (could you use segmentation to fix those things? maybe but I've not had a lot of consistent success with them). Also if you have lora's that are specific to sdxl models you can use those in your first gen then in the same workflow use flux to refine those details before passing it to it's final pass in the original model.
real world application example: if I want to create my own character and a lora for that character I need the details to be refined when I make the character sheet to build off of. This is important if I want to have the whole process automated and I don't want to have to monitor it for inpainting or other modifications. The idea is to not interact with the application so automation can happen and I don't have to be glued to each gen.
prompts only get me so far... and inpainting requires my interaction. Segmentation and masking doesn't always find the hands so again I'd have to interact with the image at some point, my goal from an engineering perspective is 80% success without human interaction.
So basically just seed gacha but even more complicated lol
probably rube goldberged it, but yeah.
lora in training, so preview image looks ass
For sure
playing around with some post processing with https://github.com/HellerCommaA/SwarmUI-PostRender/tree/master
.... 🎵 Within the sound of silence
In restless dreams, I walked alone
Narrow streets of cobblestone
'Neath the halo of a street lamp
I turned my collar to the cold and damp
When my eyes were stabbed by the flash of a neon light
That split the night
And touched the sound of silence 🎵 ....
what did she eat? That looks like a toxic cloud..
lol 😁 those healing potions make her gassy.
a pastel poot...
meow
Have ice cream instead 
@green temple you have a rock 

She stupid 

(i forgor to say so answer now lol) phone ibispaint x finger
phone with insane ghost touch problem
Phone is cursed but phone with ghost touch problems is more cursed
Mid!
Where are they from
I dunno
First thing that came to mind when I see them

Grabby 

No 
Happy Howloween
fake elf
ai?
yep, noobai with kikuchi milo lora. full details are in the metadata
COOL
de los muertos cyberpunk glow fun
Exorcise it

Dm me

Hey! Secrets don't make friends!
.... Secrets make the best of friends....
Wow cute
Apple pie 
Lies 
update after 1 year lol https://civitai.com/models/48579
would appreciate feedback, its made for NoobAI-XL (NAI-XL) 0.75 Version, NOT pony
my reaction when people LIE like that
what model are you using????
awee
what model is this one too??
these are amazing
i cant figure out the best model for em
random merge of pony and realistic stuff
Idk its from a friend
I don't know how things work, he smart 
err
Can you do merges ?
I think he said to me the models
and numbers
to make it
its just a realistic pony one and autism
I don't have local
pc bad, I run online
hahahah
same here
what do u mean run online?
but yeah no i'd love that model
can u ask if that friend can dm me about it??
idm if not!
I just gen online
like many people lmao
using cloud based resourses rather then local systems... if you're running locally you can use the model merge to get similar results. Which does the model merge on your first iteration. If you need a sample workflow let me know.
btw most images I posted is that same model lmao
mhm
i do merge models
but which ones can result in such ART
damn
nice
:D
so by online you mean some service?
yes
like a website
yes
right got it sorry XD
I'll tag you if he responds
yeah sure! idm
you can do the model merges using the comfyroll nodes. If you didn't know.
@agile meadow dm
yeah yeah! gotcha
thanks tho ,3
<3*
miki? owo
Thick 
What is she holding
I think I'm finally trying to get the hang of the pony anime models... I think
My friend that is the best hang of it
probably smartphone
looks like a sliding door handle
shes adorable
gimme
For these new "Illustrious" models, what's the best sampler to be using?
I summon help: @magic wave
I use autismmix confetti
Euler Ancestral CFG++, 1.5 or 2 CFG
Thanks
noooo this is smol wolf 
Smoli

loli wolf
You never dm'd me


just wanted to invite you to a server
@green temple I'm innocent, you can ask sooma

invite me

why sad , no sadness allowed 

always sad cat 
i have two gpu's one with vram 12 gb on with 16 gb. is there any possibility to run stable diffusion video using these two. its would be a great help. i am new learner .
what two cards do you have specifically?
nvidia 4060 ti and nvidia 3060 only not some high end
4060ti 16 gb and 3060 12 gb
i can run 2.1 but try to check 3.0 or 3.1 or svd
I use svd to maximize your resources I'd suggest gening images on the 3060 unless you are going to need to do more specific stuff with your images then utilize the 4060 to just churn out video. It will be slow since the bus size isn't great on either but you can use the 3060 as your workstation and the 4060 can just totally dedicate itself to doing video and frame interpolation.
any reference artical how to do it. if possible can help a lot.
depends on what you want? gimme a use case example. are you doing video to video image to video? What is your end goal? Do you want to be able to pose and control the character/characters?
... 🤔 you know what let me just send you over to this guys youtube channel https://www.youtube.com/@mickmumpitz
my end goal is image to video. i want to control a character. my end goal is generating a animated frame out of it.
see the above channel. After you go through the videos that seem like they will help you on your quest, contact me again if you have questions. 😁
he includes free workflows but also has paid workflows on his patreon, but if you know your way around comfyui you can often just take the free workflows as a template to create what you want.
thanks I will have a look
you won't be able to split inference between them, but you could use both of them simultaneously to inference different requests.
It's been awhile since I messed around with video... I think I might pick it back up again thanks for the insporation 😄
