#🍥|anime
1 messages · Page 139 of 1
lmao
whacky requests
i think he reported some of your posts... cant remember but posts from yesterday maybe torqx and smile and lotta
what? lol what about my posts
a girl, brown hair, long hair, white shirt, outdoors, roller skating
horny jail for anyone doing lolis in bed 😂
lol but i dont gen lolis
can't remember but I think there are some reported, just keep it reasonable safe
oh man, i wonder... but yeah i keep all my images non nude
Buttons on the right place
👍
so is this channel going to do some great anime vampire/halloween stuff or what
I might know of something that's in the works
did this a while ago

Once my GPU is online I'll gen some Halloween stuff
:p
@nova remnant did you know you could use google collab free with your google account to run SD?
'used to' google got rid of the freeloaders a while ago
i think i'm going to go for princess miyu
Wait a minute, is that the banned girl, brown hair, long hair, white shirt and rollers?

rollerskates are tough
sagemaker has some sort of free tier that I know some people are using. paperspace seems to be the popular paid site that most are going to. it's like $10/month and gives unlimited time I think.
yes on dalle got that censored
interesting choice for wheels
open shirt was censored
I like dalle3 too, @supple raptor . Wish I had more "credits" or whatever. 😛
no clue who that is but good luck Mr mod 
vampire princess, like a japanese 90s series
yeah, dall-e is great just very limiting between the prompt censor and so few gens per day. at least they aren't taking tokens when they reject prompts now though.
haven't had as much fun with Firefly
Here's dall-e. Prompt is literally just "vampire princess"
crazy
firefly only seems actually good for inpainting/outpainting from what I've seen.
the new model tehy released is doing better photorealism
and vector stuff inside illustrator is great
Here's Sonic the Hedgehog...
dalle censored this halloween night, horror themed setting, beautiful vampire with wings and witch's hat
that's fair. photorealisim doesn't really interest me so it isn't really something I look at. kind of gives me the creeps most of the time.
4th is nightmare fuel
Realism is lame
@finite herald trick-or-treat
that's cute
I love the Halloween themed stuff
FYI, these are great but not #🍥|anime
creepy-koopa
Meow bird
oh yeah.. i have to tag anime on dalle for it 🙂
Big Forehead = High IQ

how is that SD is often confusing some of the most obvious imagery, is it due to lack of training image data?
dall-e is very close to being able to produce usable sprite sheets
that's pretty exciting
for example i prompt for roller blades / roller skates and it doesn't give the right output 60% of the times
dall-e understands contex better then sd
gotta work on the fangs
inpainting fangs sounds challenging
anytime I tried to get fangs, I just got weird smiles
training resolution combined with an f8 vae is likely a big part of that problem. during training images are converted to latents which means that every 8x8 grid of pixels is represented by a single numerical value. this leads to small details being difficult for the AI to learn. consider that rollerskates and rollerblades are most likely to be present in full body images, that means they tend to be very small. inference also happens within latents, and then those latents are finally converted to pixels at the end of generation.
hmm i see and i was under the impression that a large training data set had to be fed into the model
that is also true
"vampiric smile" 😉
so in that case Ai industries need to collaborate and come up with all the data to feed it
kind of like a hive mind
even adding more data wont solve the problem of training resolution and vae precision though.
but how is dalle3 gets that detail right but not SD
it's the language model, gpt is "smarter" than clip
it can understand you better and thus you get better results
We can't know for sure because they haven't released a paper, but I think it's likely that Dall-e is using an f4 vae meaning those latents are only a grid of 4x4 pixels instead of 8x8. They may also be doing some multistage magic within the pixel space
training data could also be larger but the llm is the one doing the work
yes, the f8 vae is beneficial for efficiency since SD is intended to run on even low end consumer hardware. Dall-e doesn't need to worry about that since it's only run on their own servers with insane vram and compute capabilities.
Less money
OpenAI is also very good at preparing datasets which certainly doesn't hurt.
If you could "just use F4" for SD how much more vram would it need
not sure, but I suppose around 4x as much since the latents would be 4x smaller.
sounds interesting, training seems to be super inefficient so far though. unless someone is willing to solve that problem I doubt it will really amount to anything.
I ate those in Berlin last summer... and the curry ones too
what's that about?
The problem with trying to come out with a direct SD competitor is that it doesn't make sense to support it unless it has high adoption rate and it doesn't make sense to adopt it unless it has good support. SD has suffered from this just trying to get people to switch from SD1.5 to 2.x and now to SDXL.
yep
totally
Neither of them are much better than 1.5 tbf
You're just a bear
but so far i've read about wurstchen is that they are trying to make their product more generally accssible with far less resource demands
So are we good?
Also quite lovely.
on paper they are both significantly better, but 1.5 is good enough, while also being less hardware intensive both to run and train, and already has an insane amount of support so they struggle to get people to actually train a good model on them.
the training part is important, even doing loras or finetunes, sdxl needs ram
not everyone can afford 24GB cards
You have to compare base SD 1.5 to 2.1 and SDXL for a fair comparison. Trying to compare the models that we've been refining for over a year now to the base SDXL model for quality isn't fair or realistic.
Yeah but that's not the reality we live in
yeah the base models of each of those versions are an appropriate comparison but SD 1.5 models are more accessible in terms of training and use
I'm honestly not even sure where the limit of my 12gb card is for inference anymore with recent efficiency improvements. This is only using 6.6GB on my system.
Sdxl requires basically triple the requirements as 1.5 and it's quality increase is nowhere near that
I have 3080 10gb and it's very overkill for sd 1.5
if you want to train, it isn't.
Yeah for Lora training it gets destroyed
on 12gb I still feel limited when training.
But other than that
I can do it, but more would be very nice to have
well onlyr reason i got rtx 3060 12gb was to only run models and generate images and SDXL was very disappointing next to SD 1.5
Sdxl is basically a failure imo
even insane upscaling is no big deal with tiled VAE and tiled diffusion. I can upscale to over 8k on without using more than 8gb.
nice one! do you prefer Yanderei for upscaling
I'm not downloading 6gb models and waiting 30 seconds for an image just for an imaginary image quality increase lol
yeah, specifically YandereNeoXL_200k since there's a few different versions of it.
But it's nice for people who like Uber realistic images ig
That's the only real use case I found for it
we just need a good base XL model for anime

hopefully WaifuDiffusion can deliver
I dont think nai is gonna bail us out again
This is doing one image at a time right using 6.6
All sdxl "anime" models suck ass
They won't even give you anime image unless you put "anime" in the prompt lmao
2 at a time, base res 544x960 with 2x upscaling
not even using tiled vae
Hmm so a 3060 ti doesn't sound bad
batch size 1
8gb trash nowadays
Not even enough for gaming
For the games you want*
let me check my vram usage in Precision X1 to make sure WebUI isn't just misreporting.
Yeah this is just about bare minimum req for a1111
I mean, I could optimize that to get it a lot lower if I needed to. the resolution i'm genning at really isn't necessary either.
What does A and R mean
active and reserved
I need ~8gb just to have web browser and a1111 up simultaneously
alright looks like WebUI wasn't quite accurate, here's my real numbers on batch size and count 2, 960x544 starting res, 2x upscale (1920x1088 final res)
you should have hardware acceleration turned off in the browser that you're using while SD is open, same goes for discord. those can easily eat 1-2GB each.
Yeah I know
But watching yt without hw accel is a huge pain in the ass
So one of my browsers need it
yt doesn't even notice that it's turned off on my machine
I always have YT running and usually playing a 4k video while running SD.
use chrome with HW acceleration off for everything while in SD. save firefox to use when SD is closed and/or you actually need hw acceleration.
can confirm Firefox crashes too
Gradio hates browsers that aren't Chrome.
even most chromium based browsers have issues with Gradio.
I have 8gb vram and run everything fine
referring to earlier
hell I can even play genshin while generating
batch size 4 with the same settings as before
So you can do more images at the same time
batch size 6 is just barely doable, although it's quite slow since it's genning so many images at the same time.
I could use token merging on the hires pass to speed it up.
keep in mind, batch size 4 is essentially the same as genning a single image at 4k resolution.
Nubby my thing displays 3.4gb
why is that
mine is wrong too. it might not be counting the space taken up by the model. that or it's just because WebUI is made of spaghetti.
Vil 

masterpiece, best quality, 1girl, vampire, princess, circlet, fangs, pointy ears, white hair, sidelocks, red eyes, cloak, (bashful, shy:1.1), (blush:1.1), gothic fashion, corset, bat, full moon, looking up, from above, darkness, night, full body, mid shot
Don't be disappointed in me I'm dyslexic
meanwhile i'm busy genning a 4k image and only using ~85% of my vram.


if it's any consolation, it didn't come out particularly good.
wait, that one was .jpg. here's png
anyone tried this model yet ? https://civitai.com/models/156546/dalle-anime-model
not yet but I can confidently say that it won't respond to prompts similarly to Dall-e 3.
to be fair, neglecting to do updates and installing them later is the most common cause of Windows problems that I've seen so maybe that's for the best 
this model kinda looks similar to the dalle style
Who is this lady?
@supple raptor

I hate this dog
he's like that other hateable cartoon dog
real muttley vibes
That pink-haired lady looks like Sailor Mini-Moon or Madoka Kaname.
lol im not botherd with the dog, the problem lies with social systems and their conditions .... from economic model to relationships, also one of the concerning thing about AI development is human interventions, if Ai is not developed to function independent of human influence that would defeat the purpose and very definiton of Ai, but for now i understand Ai is still in development phase
I did throw madoka magica in the prompt
dalle is blessed
SDXL with base and refiner ....
P: (anime theme, 3d:1.2), 1girl with long flowy hair, blonde hair, blue eyes, sitting on a couch in the living room watching television,
N: unaestheticXLv31, (watermark:1.1)
(medium shot), (anime), (high quality, masterpiece:1.2), 1girl, short hair, blonde hair, blunt bangs, blue eyes, sitting on a couch in the living room wearing a crop top, and a miniskirt, bare legs,
unaestheticXLv31, (watermark:1.1), (low quality:1.2),
SDXL sucks at anime
what is nidy 2D?
wow it got the heart shape
found a collection of interesting SDXL TIs
Embeddings for SDXL There are multiple embeddings here, pls check out each one of them. Ditch those wordy prompts and get straight to the point wit...
done if anyone wants to see it
also fun to see my other loras arent overtrained
do controlnets and vaes work for sdxl at this point? i tried to use it today and got a good amount of errors at everything i tried to do haha
you have to use sdxl specific controlnets and vaes
Links please... im a lazy dude. xDD
i don't have the links for those, since i don't really use sdxl models

thats what they all say. 😂 😂 😂
I really don't use sdxl models much
even if i do, i would have to hop to comfyui due to my small vram 
what smol sus said, here is a link to SDXL compatible CN mods https://huggingface.co/lllyasviel/sd_control_collection/tree/main and here is the sdxl vae https://huggingface.co/stabilityai/sdxl-vae/blob/main/sdxl_vae.safetensors
same LMAO, i heard its much better at hands and holding things but my tests today havent been to sucessful
But could also be an skill issue hahahaha
thx
dont wanna discourage anyone but i spent the last 3 hours fiddling with sdxl and i wish i hadnt
so now im back to having fun with good ol sd 1.5
@finite herald
cho cho chotto matte
same prompt as that one on a new version of my model
Seems like fire is needed

nice Switch
oh these are really nice 
reminds me of something I did a while back
waiit
haaiir

Thanks 💖 . Looks like she is falling. I wish it had more quality though (typical SDXL low quality gen).
i wish i had such one 😦
A little different in style but I think you got it right (the flying messy hair) and quality 🔥
This woman is well-endowed.
a few more
What is the prompt for the flowing hair stuff
Let's try Lucy
There is no prompt for it, the model is trained on different anime artists, it's a mix of yoneyama mai and chi4. Prompt is:
art by yoneyama mai, art by chi4, Lucy /cyberpunk edgrunners/, solo, looking at viewer, long hair, black background, blue eyes, upper body, white hair, a masterpiece with best quality, detailed 4k
me stealing prompts and gpus @lyric zealot
add like another 300 before i started tracking 
I think you meant 3000 
That's a pretty good one
There was another one I liked but too much blood for this christian server 
👍I'll tag you there
Dragon Girl
The more the better
wait
I'm selling apples (I think) and somee potatoes (maybe) to afford a new GPU 
you should checkout comfyworkflows [dot] com 🙂
1girl, white mouse ears, arms behind back, close-up, hoodie, :3, claw pose, head tilt, white hair, green eyes, white background, trousers, ahoge, short hair, white hoodie, forehead, parted bangs
Cuuuuuute
what model is this
pretty sure it's counterfeit2.5
@primal wharf testing the lora. this looks about right doesnt it? was hard to find a sweet spot where it didnt artifact but still kept style
Nice🔥
Looks close. Will it be on civitai?
will post it in a few hours then
yes
Gens are eh because I just wanted to test the model
Hell yeah. I wanted to train it but i was too lost setting up kohya ss.
Thanks for training it.
Ya!
If you want to try out training in the future, you should try this trainer. It's a nice GUI frontend for Kohya_ss and it makes install super easy. https://github.com/derrian-distro/LoRA_Easy_Training_Scripts
prompt is:
1girl gloomy, dissatisfied, bangs on her face, disheveled hair, dark hair, eyes covered with bangs, gloomy, pale, head tilted, Jacket with a hood, space print on the jacket, stars on the jacket, headphones, dark background
The variations in style are from the model.
Did discord start to strip the metadata again?
No, it doesn't show the data because of Fooocus, dev does not want to implement it; the modified version has the option to save metadata but it still didn't work for me
oh ok
dude wtf
❓
How fking big is the possibility that it wrote my username without you including it
oh that whatever.
Interesting, didn't use much more than "The Frieber" tho
hey those are nice
Im a huge fan of the 3d styles 
When will you get tired of dalle3 ? x3
That's why i like PLMS xD
nevah. it can do any style i want & texturing, quality and accuracy surpass anything i've done in sd so far xD
waiting for next big image gen b4 i get tired of dall-e lol
First time i got 2 seperate eye colors
that's another thing, the accuracy is superb lol
sd doesn't reach this level of consistency (yet)
How fast is the Generation actually?
And doesn't that depend on how you use it?
talks about accuracy then shows the most generic imgs u could gen
clearly you haven't tried genning heterochromia with both the two colors and the sides that they're on specified without an overtrained character lora that forces it.
so a lora can do it then
cursed hands 
can't quite get the eye quality right
like 7-15 seconds per set of 4 pics from a single prompt
If you can't see the benefit of just being able to prompt for something like that rather than needing to train a separate lora for every possible color combination and position of the colors, then I'm not sure what to tell you. Dall-e 3 is undeniably very impressive.
it is as fair as comparing an img model that needs loras to an img model that has a LLM attached to it
What do you think CLIP is? It's effectively just a shitty LLM.
does that means i can run CLIP separately,oh man finally
also in the paper, it only mentions using the GPT-4 integration in regards to when you make a request on ChatGPT
they use a CLIP encoder just like sd
and i'm not using it through chatGPT, just raw inputs
technically, what's the difference between sd and dall-e?
sex
We don't know. There are theories, but unless they release an actual paper then there's no way to know for sure.
yea, this is all we have https://cdn.openai.com/papers/DALL_E_3_System_Card.pdf
and it doesn't give much
oh
break is good for different eye colors
maybe they are compiling an unique multi funtional model
sumthing like red eye BREAK blue eye and put heterochromia somewhere 
they do explicitly mention using a CLIP encoder at least for one step
Ah thank you, i'll try that out tomorrow ig
when i put 1girl with red eye i get a whole different sussy result
dal-e is only online?
nyes.
They certainly seem to be doing something similar to hires fix which is evidenced by the point in the loading bar that gives the unsafe image dog, which occurs roughly 1/4-1/3 of the way through generation when a low res image would be formed before upscaling and denoising again.
they open sourced some of it so people have made like mini dall-e's or like mimic dall-e 2 things
but nothing comes close to that, let alone dall-e 3
so yeah, only online, and the best way (least censored) way to use it is through bing
not chatgpt

it works for me Idunno 
LLMs are fun though
Is that free?
100%
What are LLMs?
100% free and unlimited use all day long from what i've seen
Ayo, gonna try out soon
large language models, like chatgpt
here you go https://www.bing.com/images/create
for what we know, the gui may have loaded different models, used for any specific necessity
Oh yea right
if ur on win 11 u can just press windows + c key
Wait, can i use while on phone too?
Dam lets go
ez gen while talking lol
I think that might require the windows insider build which isn't something I would recommend to the average PC user.
when you use an online gui, they can even load super models like chat gp, that almost need a nuclear dedicated to run it
no; they launched it for stable as of sep 26
i am on insider dev pro, however stable windows 11 has it right now.
yep, you can gen on your smart fridge if it has a web browser and a way to input text. it's integrated in the bing images website.
hmmm... Win+C just tries to launch teams for me.
u up to date on windows updates?
it for sure came to stable 26, i was on stable then
then i switched to dev after to get paint cocreator access
one fun thing to do too is call a LLM to generate a prompt, then run the prompt. And do all this in comfyUI just for fun
should be. I had 2 rounds of them yesterday
7 tries to get this to come through that stupid dog
do you checkmark the thing that says get feature updates sooner?
it could also be region related
Wait you can actually also use gpt4 with that app? Interesting
it might be US only right now
ye. bing app/copilot/chat all have gpt-4 too
use creative mode for gpt-4 and dall-e 3
try with an LLM vs non LLM prompt with same seed to see how it affects img 
nope, that's probably it. I'm a 5.5 hour drive from Microsoft's corporate campus so I don't think my region would be the reason.
try to move closer...
problem is that the prompt would be completly different. I submit a small prompt, in that example "best 1girl anime waifu girl", and the LLM makes it a real prompt fitted for that specific model
So the prompt that is then run into SD has nothing to do with what I wrote initially, and even with the same seed, results have no link
😎
kek
Any closer and I'd be in Washington 💀
Nubby knock on their door and ask for it

Heterochromeria is powerful. 😮
Still no Copilot after enabling latest updates 
no copilot like that can be installed locally and works correctly yet ?
in right click->taskbar settings do u have this?
it's definitely in stable since i was in stable first before moving to dev to get other ai features
Woah, gpt4 prompt overedit is really good
What’s that?
My prompt was just the Sentence till the : and just used an input image
Ah yea, i like having it come up with prompts like that lol
Oh i didnt think to upload a pic to have it add those terms tho, might try that
Also the Generation is accurate asf
Yh, it's seems like a rly good Feature, i mean, it made something similair to my render
my finger hurts
Also, if you don't like something you can tell gpt to change that
I like that
my honest reaction
hope the dog gets salmonella from that egg fr
nope. I'm on 11 Home, doesn't seem like it requires Pro though from the little bit of info I've found online so far.
pro only required for dev build, stable it's on home
...i know this because i upgraded from home to pro when i went to dev build and immediately lost copilot
xD
there's a bug stopping it from being on home dev insider, but on stable it's on home & pro equally
and also chat is like gone
like 100% gone
but i think i uninstalled teams too
I finally found a video that shows the version it's included with and I am behind that version. Trying another restart now to see if something will pop up in Windows update
i basically just did this in powertoys to quickly bring it up, caps lock = copilot, which is why i find it super convenient lol
Looks like I'll just need to wait a bit longer for 23H2 to come through for me. It's super annoying that Microsoft doesn't just put their feature updates available for manual download and install from their website like we used to be able to do on XP
but it didn't sound like one of those slow rollouts
they basically said on sept 26th, it's available to all
was available 10 am pst that day
u doing something to block updates? i'm sure ur not paused since ur checking for updates
they do actually provide a tool to force update
if for some reason ur not updating
this thing but i swear there was one for win 11
wtf then I'm behind your build version but nothing is coming though.
ur on the right version
22621.2361 was copilot ver
ur past that
hmmm
https://www.microsoft.com/en-us/microsoft-365/blog/2023/09/21/announcing-microsoft-365-copilot-general-availability-and-microsoft-365-chat/ here was the information
and that was exactly when it was made generally available
i don't think stable had the same bug dev build does
this is the bug with copilot on dev insider
I wonder if updating to october update .... messed that up?
was wondering how good it'd do sim racing hardware, lol
press advanced options and see if anything is here?
i think it's this one (someone elses screenshot)
Just a NIC driver update. I'll grab that, restart and see if somehow that was blocking it.
nope, as expected that didn't make any difference
just saw a news article
oml are they really doing
the slow rollout again
lol...
or no they're saying it's not available in EU
there's this workaround
C:\Windows\explorer.exe "microsoft-edge:///?ux=copilot&tcp=1&source=taskbar"
yea all this article says is
it's available in NA, UK, some parts of asia & SA
Ummmm, i wonder.
@supple raptor have you updated your MS edge?
I almost wonder if that's a requirement
cuz copilot stopped working for me once until i updated my edge, since it's tied to edge i'm guessing
Just opened edge and it is updating right now so maybe that's it
hopefully
Yeah, I just straight up don't have it yet. I found a video that shows the registry entry for it, and it's not there on my system so they must be doing a slow rollout.
Should be at Computer\HKEY_CURRENT_USER\Software\Microsoft\Windows\Shell\Copilot but Copilot doesn't exist for me.
ms sure loves to do that...
yep, took almost a week for the win11 update to pop up for me when I finally got around to enabling my TPU earlier this year.
Big red button = important
@nova remnant Have things improved?
No
berry important
Booba 

i wanna sleep there 
That's a lot of screens 
more screens
her dad
whoa this is cool, was the right img generated based on left?
grab the joystick
(Tight leather:2.4)
right is a screencap from the Bad Apple music video, left is a gen using the screencap for controlnet.
yesss
yo looks like you got more than 4gb vram now to me
Almost like I have a 3070 from someone

Im using smol sus gpu to gen stuff when I can

I love that mouse + that model
ignore the laptop 

Mouse girl very cute
I did a lot of them
Refiner is very fun to use
I've never used it
Some models look very nice with it
Like
You can use a model that gets shapes better then another after for better colors and stuff
better than without it? would love to see a comparison
is it much different than txt2img with one, then img2img with another?
It's just a second model that you run a few steps with. That's how it's being used if not with the specific SDXL refiner model.
The process is great. I love using it when using a model that closely follows your prompt, and then refining it with a model that has more pizazz than purity in the genning.
Counterfeit was horrible at getting my prompt right, I believe Aurora was the base model for that
But counterfeit looks amazing
Aurora or DarkAlfa as a base then Breakdomain as refiner is fantastic.
i have cheese here 🧀
i too have a giant pic of my face as desktop bg
it's primarily useful if you have a model that makes a style you want, but you can't get the composition/character right on it. Start with a model that can get the comp/character right then switch partway through to the model that does the style.
Un Lyucy
Looks like part of an opening to a sitcom. Will we have those in 207X? 
Probably. Sitcoms have been around since like the 40's
With as many things being thrown out the airlock recently that we've had since the 40's, that doesn't speak well for the longevity of shows with a like-minded format.
Would help press all the buttons

Big 
a quokka
Quokkas are the best
I'm a fiend for mouse girls
giant miniature space hamster
Best answer. 10/10.
mouse girls very cute
Bell neck second best prompt too
👍
but u said i was cuter
I'm surprised it can even gen quokkas 
installing comfy be like
Would pat all three of them
Using comfy be like
and then you realize the upscale node only upscales based on the native upscale factor of the model so you need another node to downscale your 4x upscaled image 

When the comfy when it when comfy when it do the node thing that comfy do when it do when it do
not yet i love doggie,both of them
@magic wave
more mouse girls for Lora previews
Wha? 
wanna hold my dog?
What preview tho? 
Any you do in the future include a mouse girl
👍
O, fair 
So well-trained with a well-cared for dog.
hes a good boi
she can hug me anytime
This style seems to generate better on 512 Adetailer
she can help me collect the dragon balls
more lora
heehee
needs more cheese
Soosi 
big
smol
@nova remnant
ip

rice is very good



Kotori here, to clean your room~
soup now
also ultra cute lol
eye
cleanest twitter user room
Haachama approves.
I found this clip to be Absolutely Disgusting. Enjoy!
Of course Haachama is into that degenelate lifestyle.
EDIT: If you saw an ad, be sure to thank Haachama's BGM - "Prelude" from Carmen Jones. This channel is not monetized.
Source video: https://youtu.be/om0EHrGdFYg
Akai Haato's Channel: https://www.youtube.com/channel/UC1CfXB_kRs3C-zaeT...
ohayo

i would mess it up
now do bonta kun
@native halo want some cheese
i wanna eat that 
Water
She looks like a red eyed Sailor Saturn, here.
wait a minute
What is this shirt called
I need to know
for a science project



black sleeveless turtleneck sweater
I offer you big piece of cheese for the knowledge 
thx



Something wrong is not right

omg nice
I want a model trained on this guy
I might do one myself but I didn't train a model for a very long time at this point
im hungry
square flowers >.>
matrix failed
oh
are these the geometric stuff you like doing ? xd
One bound oni pls
Well I'm about to leave to buy groceries so I'm not in hurry
guys
bro shit is trippy XD
what
u have exprience with stable diffusion ai ? wanted to replace sasuke with itachi in an anime art
which anime art
send
hm
i want itachi to do the same pose with the same angle
how are you planning to go about doing it?
might be very very difficult
If I were to do this
I would make a 3d model of this
and a face rig
and then use 2 controlnets
one for the 3d thingy
and one for the face
bruh shit is going to take a week at least
then I would inpaint multiple times
very difficult
I mean you could get close pretty easily
hm
maybe you could prompt it
In any case you need a model
What style do you want it?
the same style
Ok




