#🍥|anime
1 messages · Page 261 of 1
No idea, I seldom use nai, and even if I did, I use thru api not on their site
u have to like manaully add these lines (this is from that deku pic)
so like
the cfg_sched thing, set sampler to karras
the cfg manual set in it, etc
i'll probly try to binge nai all day today and see if i hit rate limit just manually using it
That seems very annoying 😭
You can get all those on the api docs for nai
but it's pretty hidden
yeah
so for me i'd use like
sec posting the mermaid rin with data so i can get the specific thing XD
That’s how I setup my nai api for usage as well
sushi 
because it made way more variety within generations
nyoooooo xD
Honestly the most I've used it for is like... 10 hours or so in 1 day, but I wasn't constantly spamming the generate button
i go like 12-14 hours and SOMETIMES i spam the gen button on rerolls
but lately i've been trying to analyze the result more, see if i can inpaint something
b4 hitting
right now gens r coming CRAZY fast tho
like 2 seconds xD
I have no clue honestly, like I said usually I just refresh if it doesn't let me click on generate
so that's when i might get a little spammy and not realize it
Yeah they fixed that not too long ago, the generation time is SUPER FAST now
Love this style! 
that's what made me tink i hit limit on day 1
cuz it was this fast when i hit rate limit too
thx 
Honestly I was surprised when I saw it generating that fast, I think they fixed something with their cluster or whatever
Either way it's nice cause I can generate even more stuff now
ye
from what i heard they updated from cu118 to cu124
its from nano machine
i thought that was only like, idk, a 30% speed boost?
Anyone got any good artist mixes/artist tags for illustrious? Kinda not sure which ones work well with it
its not as accurate
NAI gives it a bit of the default "high quality" style
the style should be worse lmao
how chinese manga stuff is
I have no idea what that means but I'm gonna assume it's a good thing! 
newer version of cuda. moar faster
That's still a huge boost tbh
iunno if it adds up tho. gotta be somethin else too i think
@rough oxide example 
going from 8-11+(sometimes stuck) to 2-5 sec never stuck?
that sounds about inline with what i got on my machine when i updated. i haven't used nai enough over the last few months to know how much of an improvement it got.
this is huke, cura, ssambatea, but on illustrious it'll sometimes be like black & white gens
It was a big improvement honestly
I love this one!
I'll give it a try later
what is that
some cuda version stuff ? 
o wait i also added monogatari (series) xD
so it's a 4-stack
cuz monogatari (series) stabilized it more (but this is nai)
i tried the 3 on their own combined on illustrious b4 tho and results r good
Apparently cfg++ samplers are good with illustrious but I have no clue if those are even a thing
Doesn't sound like a real thing
i can't get them to work, u'd have better luck asking in proj ai about those probably(?)
xD
Nor do I even see them on Forge, hmm
ur supposed to use like half cfg amount but they come out completely diff
my personal recommendation if u still want ui
is swarmui
it has comfy backend so u can do everything u can in comfy with it
I'm too lazy to install that who am I kidding
and full-feature ui
I've heard about that one, will check it out
@ivory pebble do you have any idea if theres a tag for more drawn-like stuff
Isn't that like... really bad cause you'd realistically want a higher cfg to follow your prompt?
i think i can be sorta satisfied with this eye inpaint xD
nah cuz it's.... voodoo black magic
u kno how it is with these things
it only works with half-cfg
xD
u definitely get a diff style of image though. so it's better for some compositions, worse for others
Not sure honestly, try traditional media, faux traditional media, sketch, jaggy lines..
One of those should give something
Not sure about artist tags tho
I see, makes sense
I only know traditional media, I want these shadows like in this image
lmao
A bit rougher style
idk how to describe 
literally "worse" quality
I was mainly asking if it was worth using those samplers cause apparently they give sharper/better quality images
will try those
objection!
to the better
i did not like the outputs i was getting in illustrious with the cfg++ ones
hatching (texture) is the tag
what are we hatching
I'm not exactly sure what those are called lmao, I think you can only get that sort of stuff if you use certain artist tags, I'm not sure if there are actual tags for it tho
Why not? Any particular reason?
first is what alex said, second is nubby
I was gonna say motion lines should do the trick but that doesn't really work if you just want a character that's not moving 😭
it's a shading technique
closes the door on her face
Obviously cranking the weight up does wonders
for me they were more desaturated/less colorful and looked worse. i probly didn't give it enuff testing but for illustrious i settled on euler ancestral with beta sampler
Ah yes the old reliable, Euler A
You love to see it
So how long till someone does a good illustrious mix 
you could also try adding artists that use hatching in their art such as inuno rakugaki but that one might not be very effective on NAI since their image count is pretty low.
sanae's working on one. could be promising.
advokat has one that kinda improves details a bit
artist tags makes things ugly
well not mixes, but finetunes
Hmm interesting
just essentially reinforces what it already knows and makes outputs more stable
i'm interested in what sanae'll do cuz their original finetune was like 7m+ images
so if advokat got this qual with 390k
then reinforcing the entire danbooru set could yield interesting results in short term
....before the inevitable v1.0 release (hopefully)
Yeah that's mainly what I'm looking for, but most mixes stray away from the base style of the model so lets hope for the best
oh... this ball , sure
@supple raptor 
she supposed to be "sleepy" but that works too lol
still evil 
perfectly evil, just the way she is supposed to be 

half evil now 
w-what about now?
(((evil : 4 )))
Ahhhhh she's like a weeping statue from doctor who 😱 !
this looks similar to the image I got when I asked flux to create a disgusting image!.... wait.... I made myself sad... 😢
Flux bad 
Big

Smol

@desert island 
What does that react even mean
- Pear shaped
- Skinny like a bean pole
The beanpole one was a stretch but I wanted to keep it in the vegetation category.
I was thinking of other things on her shaped like beans
Kinda

@desert island i sent you a better look at them in your dm

cute 😄
Grabby 
My niece asked me to help her draw a fantasy animal for school. She shouldn't have asked. We started with her sketch. 🤣
Cheating! 
Why?
it went random
Ragyomaxxing
🌚
when you want something done, sometimes you gotta do it yourself.. finally got around to digging through the SwarmUI code and added a feature I requested over 2 months ago. Now hopefully it gets approved 😄
while we're at it
Cyberpunk Kanna
Catto :o

motion effect came out well
Guys, tell me who uses Forge. For ControlNet, do you now need to download models manually?! I remember in previous versions missing models for preprocessors were downloaded automatically when the Preprocessor was selected.
i do, and i've always have to download my own models for controlnet. The preprcessors are there by default tho
Remind me (with links) where to get them and which ones are “in fashion” now? 😁
and more! what does this error mean RuntimeError: Sizes of tensors must match except in dimension 2. Expected size 462 but got size 539 for tensor number 1 in the list. at the time the generation starts
depends on what you wanna do with controlnet. There's like the stuff with pose, then there's the ip-adapter which is my most commonly used controlnet (i don't really used controlnet tho)
errors like that are common when using a vae that doesn't match the model
what did you change before that?
like, sdxl model with a sd1.5 vae or something
could happen with lora, controlnet, etc also
this model with built-in VAE
Well, all that I mainly used in ControlNet before was Canny, LineArt and OpenPose.
then I guess I'll just restart the SD
then get those if you used those 
um....this will sound funny, but remind me where?
I remember that on Huggingface. and here's where exactly
https://civitai.com/models/136070?modelVersionId=152216 here maybe? I'm never really used these before, i've really only used controlnet for ip-adapter for sdxl, and openpose during sd1.5 time so yeah
morning
the infamous miku
Nope lol
Its made with Flux, Depthflow for the effect and Topaz Video for interpolation
Im self taught, no course or anything and I also dont teach people, sorry
If you have questions im always there to answer them tho
Or they just delete everything and disappear 
trying to get under the radar ?
Wallpaper 1920x1080
Angy
Grabby

She lost her teddy but now she has neko in her eyes
my tiger looks a bit bigger 
tiger, 896x2048 native 🙂
not barefoot 
@nubby Well I couldn't quite get there with just prompting but I tried... I might go look for a lora or setup this workflow with an IP adapter or both but I wanted to try this out for fun 😄
Well I found an SDXL lora but it still leaves much to be desired even after messing with the prompt for like 20 minutes... 😢 . I give up for now...
cheer
what
yay! it marisa
Ooooh spooky and sweet I love it!
thx , this is complete luck lol
👍
smol marisa @supple raptor
big 🥲
Soaking feet wallpapers
new model 
yesh, it's a large scale illustrious finetune https://civitai.com/models/833294?modelVersionId=932238
new model to play with 
ai is moving way too fast lol
it seems to be a little more stable than illustrious, especially with some of the smooth art styles that i've used. anatomy is improved. i haven't noticed any significant character knowledge loss, but artstyle knowledge might be slightly worse (though still probably far better than most other models)
Here's hoping it also improves on the white background
yeah, it's a really unique style. and somehow it makes anatomy better when you mix it with other artists on illust/noob
this one was really cool other than the extra half a broom

i have the perfect prompt for comparison

massive
.
love silhouettes
oh i remembered something i did before
feet 
m?
m?
rabbit
yup bigger ears 
someone said "bigger ears"? 🙂
these are sails! 😂
yeah, the goal is for her to be able to cover herself with her ears. clothing expenses zeroed, total win 🙂
but i guess i better stop there
❤️ 🥹
cu tomorrow!
Ophrah bunny on her new gameshow!
and her co-host
O.o
she looks confused
ok .. it's an egg.. now what?
Throw it up to the egg loving magic genie who lives in your ceiling and maybe he'll grant you a wish!
ok....
It seems like maybe you have questions... so here 😄
Oooh this one is better... it's no flux, but this model merge really tries 😁
Bearable eggs
Saitama?
now thats 3D
What model is your lora based on?
this is base noobai
I'll play around with it some more after my wife goes to bed. I don't want her walking by and getting mad at me... She HATES spiders.
hehe
it's worth it
I really like this model
it's got some fun poses/expressions too
just want to verify that this is the right model https://civitai.com/models/833294/noobai-xl-nai-xl
yes
finally i have a new toy to play with
just completed a promising merging workflow and testing the first merge i'll use as a base to merge with anime models
the gens looks all pretty stable, without crazy mutations
it's so funny🤓
i wonder what will happen when i'll merge it with a cute anime model
you get better hand and anatomy but have to throw more weight at the style you want 😄
yep, the intention is that
use this model as a base and just merge the style on it
i'm using weighted blocks merge just to transfer some aspects of a model in another
if i have done well my work, mutations shouls be minimal during the transfer
i'll try with this for the next test
it looks pruned, i hope it's not a problem
good luck it seems like we had similar thoughts 😁
such experiments are only possible with comfy
it's so funny
mutated hands 😦
everything else is perfect, but the hands are always mutated, i suppose every merge need different weights
i'll try with a different not pruned model to exclude that possibility, than i restart playing with weights and find a combination with less mutations
or maybe the new model used in the mix could have been bad in genning hands from the start
again, mangled hands
but the results are very cute
tomorrow, now i'm very tired
see ya guys
She really had to take a leek!
she don't want you to take away her leek 
depressed arc, knowing she can't protect her leek
lick
halloMeeeow! 😂
^-(starts moving nodes and noodles around)
"Ooooh I just had a fun idea for a new workflow.... gimme a bit" 😁
^-(Throws stuff around the room)
"BAH! 💢 STUPID HANDS!"
Here's hand for you
looks like a children's book illustration
just had a terrible realization, most of the time, terrible hands depend from terrible clips...
Merging clip is very often cursed. If you're doing that then yeah, that could be part of your problem. Also, merging XL models is just a lot more difficult in general than 1.5 models because most of them are divergent enough that they're barely compatible at best.
yep, it's hard, but i'm forced to do it, because without a merge, the merged models, gens weird mutations
this test models shown promising results with the right weight
If the merge gens weird mutations that weren't present in the donor models, then the merge is likely a lost cause already. I'd start by trying to replace the clips with the ones from each of the donor models first to determine if that's the problem, and if not then it's time to go back a step and keep trying different things in that merge to find something that works better. More merging on top of a broken merge is unlikely to fix it in a way that makes it worthwhile to try.
i suppose that's it
i can't expect correct hands by using a model with cursed hands
anyway iìll know when it finish to test all the weights
Yep, I battled this for days back when I was making kakigori. Sometimes the best solution is to find an intermediate merge that can be used to bridge the gap between two other models before merging them. IE: instead of merging A with B, merge A with C to get D and merge D with B.
yep, i only use sd1.5 models
oh ok, merging should be a bit easier then. at least you have a ton of models to use to try to find some compatible combinations.
yep, working on xl and pony would be too exhausting
merging pony with anything other than pony finetunes/loras isn't worth it. its far too cooked to salvage.
for 1.5 merging, i've had good luck using any of Mods13 or AnonymousM's models as a bridge when trying to merge 2 models that aren't compatible with each other
oh, ty for the suggestion
from the tests, looks like i get less crocked details with a very low merge of the clips
1.5 models have some completely fucked clips from all the incestuous merges most of them have gone through. i'd try just transplanting the clip from NAIDv1 onto the merge output. that often improves 1.5 merges.
nah, the witch has some idea what she's doing
mutations were the norm
1.5 merges were often "i merged these 15 models (all of which are just merges of the same 5 other models) in random proportions i think, i don't really remember for sure even though i did this all 2 minutes before uploading and never tested the model but it's definitely the best merge i've ever used!"
it must be less cursed than what we had before
that's must be how "i have no mouth but i must scream" was written
naidv1 isn't a 1.4 model? the clip is compatible with 1.5 models?
1.4 and 1.5 are the exact same architecture, so yes they're compatible. naidv1 just happens to have one of the least cursed clips of any of the 1.x based models.
both just use CLIP ViT-L, which is also the one used for the smaller CLIP in XL models so you can even transplant the CLIP from NAIDv1 onto an XL model if you want, though idk how well it will work in that case since the CLIP ViT-G would still be the one from whichever SDXL model you're adding it too.
oh interesting, ty
i was thinking about deepen my knowledge in clips, to work better on themù
i needed this tip
xl and animagine works btw. it's what I use
someone know where i can find the publicly released model of nai?
links to all 3 here. (Full) is the one you most likely want. https://blog.novelai.net/novelai-diffusion-v1-weights-release-en-e40d11e16bd5
the file in unet directory?
yep
i'm suspetting comfy doesn't suppor nai model very ell
the only node i see is load diffusion model and it's not provvided with a clip pin
everything else is associated with the api and not the model 😦
loading it as any other model doesn't work on comfy :/
i spy with my little eye a node that maybe i can use lol
that model should load just like any other model with the normal model loader. bedovyy's ComfyUI_NAIDGenerator custom node that you probably found is just for api usage
when i try with normal loader, the gui give me error model not found
hold on, i'll just convert my old copy to safetensors and upload it.
i had a few more .ckpts than i thought so it's taking a little bit to convert them all (swarm doesn't give an option to pick which ones to convert sadly) it's almost done now though
i'll throw up sd1.4, 1.5, 1.5-inpainting, and wd1.2 as well so they're all in one place.
ty ❤️
i'll have fun with them
some of the gens are very cute, too bad for the crippled clip
so hows that work with forge?
It seems to me either not at all or with “dances” that I don’t understand. but it's damn tempting.🤣
As far as I know, nubby and Shaikorth use ComfyUI
i love how fast uploads to hf are
hf?
huggingface
ah...
shrugs. like I need more models. got to many now
let me know if any of the ones you try seem to not work correctly, i didn't test after converting to .safetensors https://huggingface.co/nubby/SD1.x_base_model_archive/tree/main
hmm what that?
i just converted all of the main sd1.x base models to .safetensors since most of them were in .ckpt originally and re-uploaded them so they're in one place.
ok but what models? the nai stuff?
NAI v1, sd1.4, sd1.5, sd1.5-inpainting, waifu-diffusion1.2, and the original offset_noise model.
it's all old stuff that not many people are probably using today, but could be useful to some people and i'd rather them have access to .safetensors versions than potentially sketchy .ckpt versions
totally agree on this part
k
This means that the NAI model is not the model that is currently in use, but rather an old one. and the results won't be as good?
Nai is currently on v3
so good for merging?
yeah, the one i uploaded is the same one that leaked a couple years ago and was recently officially released
yep, could be good for merging or for training purposes
kewl
I would say almost all sd1.5 anime model would already contains a merged of naiv1
even most of the realistic sd1.5 models at this point contain some naiv1
True true, would be huge if naiv3 leaks 
the biggest use for it in merging with more modern models is transplanting it's clip since that's a lot better than the ones in a lot of models that have been trained and merged a million times
as soon as the profit from its commercial use falls, I think
i'm hopeful that they may release naid3 publicly once they add naid4 to their site. it'll be very sad if it takes them as long to release that as it did naid1
upscaling the image i'm using as comparison
for now i'm using the clip of the base model i merged yesterday
far from ideal
another test
with a little higher res
more higher res
even if the clip is a little messed up, the result is good
i got an error >.<

the results are even more promising than the one i got during the tests
eheheh, to poor to spend on my pc for now lol
i like this more 👍
same prompt.
i really love the backgrounds of this model
i'll test it an maybe if it's worth, i'll upload it
looking at the deviances the people gen, i'm a little discouraged from posting models lol
i'm a little scared to be blamed for the bad conduct of an user 😦
not exactly what i expected, but god, it's gorgeous
reminds me of lunar radiance
indeed !!
nice
it was an experiment and i must say i like the result
me too
i see
oh
yeah
you can disable nsfw posts from your model page on civitai
oh, that's positive
on a positive note, even without nsfw in negative, nsfw gens looks terrible, by default
you can even hide specific tags or users if someone is being problematic
ops, i wrote bathing instead of basking lol
a cute (cat:1.3) basking under the sunlight
i can read in his eyes "why are you doing this to me?"
really beautiful backgrounds
indeed but not @nova remnant 
uahahah
i'll post it tomorrow, going to sleep
after the last gen
another one
good bye
wat did I do
im sad now

jk 🥲
she can hear miku's guitar from outside
@finite herald miku posting alert
And now we wait for like 30 minutes for 12 seconds of video that is actually just a basic test without frame interpolation... 😭
It looks pretty cool though from what I'm seeing in the Sampler as it chugs along... Kind of has my hopes up 😄 🤞
my favorite model couldn't draw the crystal wings of flandre scarlet.... 😭
this is the best option
still needs work but it's getting there... and my math was wrong only 7 seconds of video... anywho back to the drawing board. 😏
try flandre scarlet, wings, crystal and if that doesn't work try flandre scarlet, crystal wings. if it's an animagine based model then you've already gotten very lucky with those ones though. animagine hates flandre 
I’m not sure what model the author had as the basis for his merger. because the history of the merger is not fully traceable from the models. 🥴
What is this?
the last version of tokens in promt was (flandre scarlet:1.6),(flandre scarle wings:1.58) partly this is a rather strong weight and even with it the result is not very good.
it's common on civitai to either randomly merge a whole bunch of models and not be able to tell the recipe, or to just pretend you don't remember the recipe to keep it a secret. just a strange behavior from that community.
based on this information and the style i'd guess the model likely contains one of the early versions of animagine which had almost all of the touhou data removed from it's dataset. it's hard to say what else it might contain though.
The author in the versions indicates a list of models. but alas, until a certain point. however, there is always the opportunity to directly ask the author of the merger model. that's a good idea and I'll try to ask him.
the touhou data removed - they should be punished for this!🤬
ask for piggyback ride, get flan carrying and tackling remi. this checks out.
huh cute
yes nai knows well who it is
I used illust for that lol
I meant my result. 😁
and it's scary 🤣
Amazingly, Illustrious knows concepts that even NAI didn't learn. NAI cutoff is usually around 500 appearances on danbooru before a tag works, but Illustrious often works with tags that only have about 100 appearances.
there's something extremely satisfying about having your code officially merged into software you use on a daily basis
Illustrous can do finger heart while nai can't
Local is finally winning frfr
I didn't understand correctly?!
that's heart hands, finger heart is just the index finger and thumb
what is true?
yeah, like that
pony models don't seem to be able to do it (might be an artist thing) but it does heart hands really well
Yes, you already talked about this earlier. I noticed.
and yet - HE CAN!
But I’m wondering, is this really an anime series based Touhou?
could be Touhou Musou Kakyou. Haven't watched it yet.
Possibly memories of phantasm. I've only seen clips https://myanimelist.net/anime/55315/Gensou_Mangekyou__The_Memories_of_Phantasm
Marisa, an ordinary magician, suspects a youkai is behind the winterly weather in the middle of spring. She goes off to investigate and soon arrives in the netherworld, facing the youkai behind the strange incident. Will her friend, the shrine maiden Reimu, arrive in time to save the day or will all be lost? (Source: AniDB)
it's on youtube 😄 https://youtu.be/gjoH4PY6oSs?si=07icPX1IIQY7gdMu
There are also a ton of studio quality fan made animations for touhou
the animation quality is impressive for a fan project
Yeah touhou fans have some serious dedication. If you need more proof see how quickly Bad Apple is made on anything that can possibly play an animation
yeah, bad apple is how i learned about touhou originally 😄
It's a bop. I have it in pretty much all of my go-to playlists.
me too
I just did the math for myself... just based off of my own renders it would take me 11 straight days of compute just to render 20 minutes of video at 16fps and 720p... this is not including writing, voicing, sfx, music, etc.
this is gorgeous thanks for sharing. I've never seen this before.
I think this clip is so famous as the AMV of one German person on Evangelion
i just realized i saw this when it came out... 14 years ago
I knew nothing about Touhou and apparently it's avid fanbase until just a couple of days ago.
the funny part is all the remixes and remakes of the song/video, but then there's entire documentaries about it
it's insane
Neon Genesis Evangelion AMV (not made by me)
Music: Rammstein - Engel
This fan base is crazy. There are so many different variations. There's even a comfy node that plays bad apple in ASCII art https://youtu.be/240Vq6tIxio?si=nMjSpzsYxQQBC681
sm9519847 - コマ撮り実験アニメ「6566/6566」【Bad Apple!! PV【影絵】アレンジ】.mp4
http://www.nicovideo.jp/watch/sm9519847
**EDIT2: Ten months later and this video reached 1 million views. Thanks a lot everyone who spent time watching "Bad Apple!!" Extra special shoutout to those who felt inspired after watching this video. I can't take any credit for its creatio...
Look out rabbit hole here I come.
until i got into AI stuff, the only other place I ever saw touhou stuff was the occasional random doujin
that gives me an idea...
I have a custom made LED panel made from rbg light strips and it can play pixel art, just never had an actual use for it
i should put it on the wall and just play bad apple on loop 😄
that's cool! you should post a video of it once you have it set up 😆
and be sure to make a video for us how cool it will look!
not sure if it has enough pixels to look good, but if i get it setup i'll definitely share
@supple raptor tell me, is it really “Cookies”??
sigh animation is tough 😡 does anyone have experience doing anime style animation? Tooncrafter is a pain and I've never been able to get a good anime result with SVD
anyone have experience doing anime style animation I think such people work in studios
well I'm gonna keep beating my head against this wall until I get at least a little closer.
Yep. Marisa was following Yue around
So this is Yue's cunning and insidious plan!?!!? how expected! 🤣
I've got a couple more ideas that might help some... I'm using an IPAdapter already but I think I have a couple more techniques I can try from my SVD stuff and then I'm gonna see if I can use florence to better alter some prompts for specific keyframes.
oh this YUE
Yesh
Perfect time to share this: https://youtu.be/Grp6m57IRdg?si=8V1djiPOZaqWFwUU
I got #2 on the canvas btw 😎
There's an upload already at https://www.youtube.com/watch?v=51FKNhfKXts , but this one has fixed colors and audio offset, and the video is lossless (it's still reencoded by Youtube, but this means the quality will always keep up with server-side encoding improvements)
now that I think of it, I should've rendered t...
Done by neuro community called the swarm, me included 
but in appearance, today she decided to imitate Chino Cafu from “Rabbit House”?!
gn chat
if a model is not trained for that concept, it's very hard, no matter the effort
images get interpretated by the model in something completell different
and you need the right prompt to help the model understand the forms and simply regen them in anime style
That image is also going to be particularly difficult because of the very low contrast of the man's skin and the dark backdrop.
yup
the ideal is to cut the single subjects from the image, regen them singularly using diffuse
and regen separatelly the background
and than fuse all together and regen it at very low denois to fuse everything
this is what i pushed for!
my workflow works!
now i'm the chibi master!
cute things are at my graspt!
the mangled hands fight back!
"(((miku))) as a cute snowman, winter wonderland"
"(((miku))) chibi rock star stage lighting electric guitar cool" - no special models nor LoRAs were harmed in the making 🙂
lol
"(((miku))) chibi rock star stage lighting electric guitar cool, semireal style"
cinematic makes it... creepy
whoa!
"cinematic still (((miku))) big head cute chibi rock star stage lighting electric guitar cool . emotional, harmonious, vignette, highly detailed, high budget, bokeh, cinemascope, moody, epic, gorgeous, film grain, grainy"
with flux schnell unchained, no negatives
the core is the "miku ... cool" part, rest is just for the cinematic look
native 2048x768 - no duplication issues with flux 🙂
insane
adding cyberpunk... now all we need is a 3d printer!
that's why she's crying
You sound horrible 
I missed a whole Miku adventure today 😆
There used to be a miku thread in here that people often post miku in it. Since that’s gone, people with stable society role been posting them in #💣|stable-society channel
ugly
I don’t have access to the channel so that means it doesn’t exist
i demand miku-mash to be re-enabled 
do u now?

he speaks the voice of the people!

i second this motion. miku deserves her own thread. also, miku role please
🙃
interesting results with trading card template in the prompt
sad miku pokemon?
i'm starting to think that it's better to simply use the nai clip and thrash everything else
the quality is totally on another level
powermoth 🙂
using nai clip also often helps make most loras work better on your model
it's like a cheat
what's the point of beating the head mixing clips?
much better pooling all together a clip well done
there isn't one. merging clip is almost always a bad idea.
just pick a good clip and slap it on the model and call it a day
you don't even have to beat the head understanding new models
totally agree
by the tests looks like that the cured and full model use the same clip
with one or the other, the results were the same
unless i just found the same shared tokens lol
anyway it was a funny adventure
i try to mix more models after some tests on this one
Grabby 
Not grabby 
I don’t think Marisa would part with the coin so easily.
it's ok, she borrowed it from reimu when she wasn't looking anyways



