#π₯ο½anime
1 messages Β· Page 231 of 1
Ok explain why the hell they would put that as 4o's capabilities when it's another model generating
Is it that 4o can perfectly describe a face in words?
because gpt-4o is still responsible for the vision part of it.
So it's pretty much img2img
Idc if its not literally so (which it is)
it does the same thing
notice under model evaluations that it does not show any evaluations for image generation performance. it only shows vision understanding evals
how do you evaluate image generstions. and even if its not it does the same thing
in other words you can show it a picture of a banana and it's more likely than all of the other models tested to recognize it as a banana.
i dont really care tbh
it has that ability, I can give it a picture and the website spits out a picture of the same style or same person inside the picture
image generation evaluations include prompt adherence and visual quality of the output.
image generation evaluation is a lot easier and more reliable than text output evaluation tbh
Ok I still very much believe it has pic gen output. And even if it's not like I said it does the same thing
Very cool stuff
in the sense that you can ask GPT-4o to generate an image and you will recieve an image, yes. however that image is being generated by Dalle-3 just as it currently is when you ask GPT4-Turbo to generate an image. text model reworks your request into what it considers to be an ideal prompt for the image model, then sends that to the image model (Dalle-3) which spits out the actual image.
angy chibi inttrupts with a boot to the head
No? You can input an image in too. That's the cool thing
sighs. blah tech spam isn't anime
Dall e 3 can't. Turbo can give you a prompt that maybe close but never the exact face for example
4o can
yes, and in this case the vision model looks at the image and considers that along with your text prompt when generating the prompt that will be sent to Dalle-3
oooh she more angy
So you are telling there is a prompt that perfect describe my face, not only that, dall e 3 listens to it 100%
Well aint that even more impressive
with enough attempts at generation, yeah probably. a face really isn't all that hard to describe.
capabilities wise at least then stuff looks like this
D3: Text -> Image
Turbo: Image of your face -> Text -> Image of not your face
4o: Image of your face -> whatever -> Image of your face
Ok π
Man actually telling me that you can copy any one's face on dall e 3 as long as I have the prompt for it
we've had humans who's entire job is to create a sketch of the suspect of a crime based on a witnesses description for at least decades. a computer that's been given vision being able to make the description shouldn't be that surprising.
Omfg this is ridiculous. Not to mention, Dall e 3 doesnt follow prompts that well
that's largely because it doesn't use your prompt. it uses the prompt that gpt4-turbo made based on your prompt
that's why you can get the dog on some attempts with a prompt and not on others without changing your prompt. because sometimes gpt makes a prompt that the filters then decide isn't allowed.
sleeps
beautiful art style!
Am extra sad today π¦ attempted to do a batch over night of ~300 (It has succeeded before). Apparently, it crashed, like pc crashed while i slept π¦
(300?! What the heck were you doing?!), well, i was using a wildcard plugin for many different stances, landscapes etc
SO yes, it was randomizing the positive outputs
I'm wondering if i have a memory leak somewhere, since Python can end up just using 18-25 gb of ram, after some gens
cool
i use wildcards a lot, i use an extension called unprompted for it, it allows basically programming within the wildcard files for better control where needed
Ended up getting this tbh https://civitai.com/models/45448/full-feature-fantasy-prompts-characters?modelVersionId=243705
100+ detailed dynamic prompts with fantasy, sci-fi, video game, mythological and historical themes. Each prompt includes setting, background, pose ...
That one was used for all the images above
oh , those were wild cards ?
Yeah boi
cool
Cause i have 0 imagination xD
Now is just a matter of finding out what promts and lora's works well together xD
i have a little, at random times, but mostly i can't think of anything
learning from japanese sd prompts rn
Its great just sitting, staring at the screen, for 30 min, only coming up with 2 promts
i just generate the same 3 prompts for days at a time
I just put on a batch of 300, and let them gen over night xD
Until i run out of regular ram, cause Python hogs it all, and then crashes the pc overnight π
lol, 300 would take me ~20 minutes
Gib your old 3080 xD
send money
MONEY!
Gosh darn it.... The wildcards are slowly moving away from anime, to more realistic
And i even have the anime lora on π¦
Me tink
And we're back
Hey @green temple Do you happen to know of an extension that allows me to rename lora's, IN the webUI?
idk
HMMM
Pretty cute

She's just on a mewing streak, but you wouldn't get it smh
true, first time I see those words

Get with the times old man
Beard is too small and needs more wrinkles
That tattoo, looks very nsfw 
Needs to be lower
π
Cool angle 
feet 
Yes. Feet. 


RIGHT! Anyone got any face fixing stuff recommendations?
Cause this yielded me much sadness #πο½general-with-images message
yes
the one and only
Adetailer
just seach in the extension tab, it should be called adetailer or after detailer https://github.com/Bing-su/adetailer
believe this is the github

And then...... Lora's??
no
adetailer will inpaint, during the generation, the faces it detects at a higher resolution so details will be better
So obtain from here? https://civitai.com/search/models?sortBy=models_v9&query=adetailer ?
not sure why you would get an extension from civitai
when its in the extensions tab to just click install...
you only really want the face model
Well
the others, never heard them be any good
face model comes with the extension
there are some others that come with it, not worth using I think
Spoilered, cause technically not anime, but that extension worked great! :3
as long as its art it should be fine to post here
no one will hunt you down (maybe) 
Upscaling the img, and jesus christ, the improvement
(Still being genned)
Heres the final img, @nova remnant π
Call back your hitsquad there, i spoilered the non anime pics!
adetailer works on my pc now
use "BREAK" in your prompt too it`s so helpful or... you dont prompt i think ? right
And I think I need a vae lmao
I didn't seem to need BREAK, running it through img2img
It's not a hit squad, it's STALKERs in some highly radioactive area.
So radioactive that she doesn't wear any gas masks or suits lol.
So shes a Child of Atom?
Or was she a regular girl, before the radiation?
Regular girl from another world. A medieval fantasy world.
steals your gpu
No. Bad cat
My adetailer is taking 2 minutes

counterfeit is cute
next one should have vae lmao
no , bad smol
SIG with an AK gas system
but the gas transfer thingy is missing
Or a 2nd barrel for a small EMP charge cartrige, for the obvious mechanical threats in the wasteland..? xD
Mah webui broke π¦ Can't launch it atm. Gotta wait for Cs to be back on π¦
what is cs
Not what, but Who. CS1o π
Idk
If i knew, I fix. But i no know, so no fix
make her smoler 

she will disappear if too smol

./setsize 0.001
perfect 
Pocket... Elf? (Cause of ears)
πΆ Sitting here, just browing Lora/models, waiting for Cs to come back on πΆ xD
Remember to be careful when you put your hand in pocket
I ain't got any pockets atm π
I'm bored now, π Idk what to do
kill with fire

Don't you dare hurt sweet toofless
sweet... tooth
I wonder if sd3 can gen him
Imagine begine able to gen anything atm 
Do you use forge

Still can't launch my webui xD
answer me 
Do you use Forge or something else
Using Auto1111
why not forge
Cuz Auto1111 was ez to install, and quick to install
And it was the first ui i ran into
Am fine with Auto for now
yeah I use it, way better!
I think your gpu is 11gb it should improve speed for that one
Like its literally the same UI but better speeds and vram stuff
It is 12gb, but... I also wanna be able to use my pc while its doing stuff, so capping it at 8 is fine
and the same way to install just use a different link to pull
yep yep yep

Mebbe later. Need to get auto back up first.
go do it
You said you were bored, make a new install on a different folder :b
No no. CS1o isn't on yet xD He be my Go-To support xD
ponders a theme of art.hmmmms
listens to Nightclub 80's πΊ Retrowave Cyberpunk β¨ A Chillwave Synthwave Mix for The All Nighter
Speaking of. U seen this adorable big shit from Palworld?
nope
don't have palworld
theres a 1 button install
step 3 profit
Step 0.5, Recieve link from a Cat

Now i just need to find out, how to pull stuff from a different folder than the default, so i don't need to move anything of the Lora, checkpoints or embeds i has
I just did it as a branch. one sec
... A branch?
just move all 3 folders into the new one
should have that on there
That'll end up taking 3 hrs. @nova remnant
I backup my Lora's on a near daily basis...
moving should be super quick
Nop
Slow as fuk
Copying is what would take a while
are you on a 2005 pc or something
Its instant if its on the same drive
I mean, yes.
The drive might as well be
why you using a hd
cause s soft drive means it melted?
Huehueheu
No. Cuz harddrives was expensive, when i built this pc
So i used my previous drives
Which were also previous, in the previous pc
I have mine on a external SSD
I TRIED having mine on my external.... But that took 8 min just to load 1 checkpoint
works just fine
with forge it's much quicker to load to another model
unless you mean move from HD to HD
Still need to figure out how to do the thingymabob
Tbh I think Forge would be better for you in that case since it has better VRAM/RAM optimization, but you do you 
We'll see.
Got Forge up and running now
1000/1000 interweb connection
Don't know what it is like for other people but I saw people were saying it was better than A1111. I tried it out and it tripled my generation speed, might not be the same for everyone however
wait... @nova remnant ... This mean i need to Re-DL adetailer etc again?
Ancient pc, so we'll see π
I think it comes with extensions
controlnet net and stuff
might also have adetailer
No ADetailer
I think it's gonna be worth it in that case, it's supposed to work better with lower end GPU's
"Lower end": pfffft
@coarse brook x3
It WAS high end, when it was new
Not anymore! Get with the times old man 
Ugle
I'm not obsessed I swear..
YOu just don't have a keyboard, and can't change the promts
lmao
I generate from my phone so I'm too lazy to change the prompts not gonna lie π
I c. Illuminati keeps your promts the same to train THEIR AI to mess with u
As long as I get my waifus with cool looking swords and armor they can do whatever they want 
This lora can be used to generate images of chillet from Palworld. It is trained on a variety of art styles, including both SFW and NSFW. It was tr...
On a side note, it's genuinely amazing just how consistent NAI is with generating accurate character outfits
ooh no . I can not post this.. hides it
hmm nah
CAT LIZARD GIRL -HISS -RAWR
:
oops did I kill chat?
Still trying to fix my webui xD
hmm ok...
shrugs
1wall, solo...
Didn't know there was a promt for an irish girl in the snow, with pompoms

I HAVE MY WEBUI BACK! WOO!
Using blender, I made a screenshot of an anime character in a setting. I then put it through img2img and got this:
I zoomed in, in blender and made a close up render of the face. This kept the same angle but the face larger. That allowed the SDXL controlnet to maintain the composition of the face (if I did it direct, then the expression is not maintained). I put the render in img2img again and removed background and got this:
Finally, I overlay the face output on top of the background output and line it up the best I can. Then I manually erase things like the T-shirt along the collar from the face image and hair that sticks out from the background image. It gives this:
Yes I improved by workflow to make it look more 2d (I added lineart style in blender and so change the model and some nodes and prompt), this is only for the solution how to use accurate SDXL controlnet for the face too.
That's why I didn't fix the image 100% and also why I left the hands messed up. With this the demonstration shows a workflow for manga can be made
Inpainting or something can be made to fix hands and such. Also with Segment Anything+ Grounding Dino it's possible to get fast and accurate masking on anything you want in the image, such as red T-shirt. Then you can denoise it specifically to add more details like more folds and things. I kept the colors simple here to be more consistent and also because it's not important at this point
This is the original blender render btw
I have changed my workflow now to also be more consistent. For example I use the comfyUI node iterative upscale with pixelk upscale. So first I put the blender render in regular img2img but no upscaling, only style change. Then when I get what I like, I do the upscale process
The background image I used BluePencil XL because I made it a whole before. But the face I did with anima pencil X. But now I only use anima pencil XL. Anima is a mix of BluePencil and Animagine 3. BluePencil is good with img2img but not 2d looking. Animagine is good to generate 2d images but img2img ability is very bad. So anima is the perfect model for me. Also, because it does 2d, it's so much easier to edit manually even with bad drawing skills. I only used a mouse to fix it, no drawing tablet. If you zoom in where the collar meets the skin you can see the manual adjustments
To show what I changed while experimenting with workflows, this is a render in blender. You can notice there is lineart generated inside of blender itself. This gives it a more 2d look but still very 3d. Somewhat like a webtoon
With img2img style change but not changing the image size i get this:
Then with upscale it looks like this:
Or the other way around I can't remember lol. But I did not do maximum canny controlnet power. Maybe I should do that because you can see some small structural differences. But anyway you get the idea. I used this type of image because it is more complex so I thought it would be better to do that to a more difficult image so I know it can work or not
I did not prompt for the mouth
I mean... It is a fox girl. Maybe she be eating something with long legs?
Also if anyone is good with promts, i asked a question i #πο½prompting-help π
Model childrenstories_v1customer: prompt: pop art, vector art style , eyes and ears, invisible background
entered random mode
Is (something:1.3) random enough?
oooooh
updates avatar
oh it gets better
hmm tho not pink
snowcre?
very anime-ish? yes?
oooh
2 faces or 3?
something smug
@lyric zealot smuggered
When u r trying to create a badass looking warrior.... But no matter your promt, the AI won't cover the tits
Looks safe, ye? NOP! Next preview: BOBS IN YOUR FACE!
bob?
Bobs, Boobs, Boobies, Tittes
ya know boobs upside down is poop
do rating safe on positive

and dont prompt for them...
What is cmd for dat?
Am not promting for em to be shown π¦
But the rating_safe, rating safe, whatever the promt, worked π
Nvm. DIdn't work once i got into img2img π¦
What is your prompt
huh.. lcm karras
ok more anime , finally
lower steps and lower cfg and it gets wild!
This is really cute β€οΈ
thanks
there my new av here, a angry snarling panda Snow Leopard, yeah the zen munk panda kitty lost his cool when news BS over-rode his tiktok feed.. grrrrr

Major panik folks. I have an image that's being genned.20 steps... It's stuck 19/20 done, but only 5% on the first step
But the preview looks amazing.
Hoping it doesn't crash... If it does, I do have a screenshot of the blurry preview
Both vram and regular ram is sitting locked in at 95% use atm
It's struggling
RIP, what could've been
@coarse brook rate my waifu gen
you're denoising way too hard
Are you on 0.8 or something
do 0.35 to 0.5
more than that is Unecessaurus
In the end of the generation is when the vae is being applied and its the slowest, so pretty normal to get stuck making you sweat
π
Well, first of all, it hadn't actually finished rendering/genning, cause it crashed
And i don't really do img2img atm, cause i actually like the glitchy look π
Time to spend the next 6 hrs converting her from a 2D drawing, to a fully fledged AI character
Vae... Isn't that only when you do have vae's installed? Cause i never knew what they are, so i never really looked into them
models have vaes in them usually
otherwise you'd be looking at bleached images basically
I c. Well, my Rule of Thumb is usually when an additional 60 sec has been added to the timer, its a sign it has frozen xD

Cause it adds a second, every like 10 sec
I'm talking about hires

I don't use hires xD
What does it do besides upscaling the resolution?
basically makes the image better
small details clearer and stuff
using the lower res image as base
ok loving this new model I found, mostly for furry but new vae too so it works.. now what shall I try to make?
tries smugpunk
hahaq love it
MIKU BUNNY
smug rabbit
somehow.. I don't trust her
yawns
hehe
sorry didn't mean to hog the posting
merges the new model in with some older mixes :D
bakes in a VAE
tries same prompt with new merge
it works
rawr?
goes back to bed... thud
I seem to have found a fix for the "stuck at 98%, but console reads 100%.
Open a new tab, (On the same window, and go to new tab), wait a few sec
Click on the WebUI cmd window. press enter a few times. Wait for new activity to occour. Press enter a few times again, and that should have cleared up the gen xD
prompted for miku hatsune there but.. well.. wtf?
ahh there she is
sigh, hate it when I can't sleep even tho i'm exhausted
your prompts are very long
i know
and seems the end result isn't the prompt
medium view, Standing, 1girl, solo, (kawaii :1) (beautiful detailed skin : 1.2) (finely detailed beautiful hands)
beautiful detailed white hair, fox ears of fur, pink hoodie, green eyes,
beautiful detailed girly room, curtains, wall, wardrobe, art by Akihito Tsukushi lora:more_details:1
tho I am using a merge of anime furry and another model to get that π
i'm gonna go pass out now.. thud
but before I do.
Learning how to smol
why do they come out looking smug?
only 2? i see 4..

π
yoo what model is that
damn NAI really evolved then

you're about to get mugged
Give me an Head pat or get Mugged!

You wouldn't do something like that in front of this cutie?
Here's another one 
i wonder
It doesn't matter how much I try to lower the size of them they're just always big lmfao π
the model is biased
I mean that's how big they are in the anime/games so I guess it makes sense but still π
yeah, but she doesn't have that bee belly
{{{{{{{{{{{{{{{{{{{small breasts}}}}}}}}}}}}}}}}}}}
I'm not complaining tho 
kek
xl model?
What's funny is, the artists I use don't even draw swords (as far as I know)
the mugger got here head pat, now is time for the next victim
head pats are never enough
I think they impact more on body type than what they are doing
its more of what they do anyway
She does, specifically this outfit of hers
guys, what's the booru most of the anime models are based?
Danbooru I'm pretty sure
dan
That's where I go to search for prompts for NAI/Pony
Yup, but still it is quite impressive how good the weapons look with their art style
ty
Still takes quite a few rerolls to get an accurate pose, but that's mainly cause I maxed out the token limit cause of her ridiculous outfit π
nai can BREAK?
i try a thing
Imma try out her other outfit/weapon, it's much cooler
what about without them
You can send it to me for a free review π
you got your bright painting and dry elf
who can draw the sword, will become king of britain
forgot the chokers in the prompt with the last one 
eheheh
sometimes i really like when regional prompter messes up π
Unironically had a couple of gens with that and I had no idea why until I realized I had "uncensored" in my prompts, didn't know it did that without other prompts but it makes sense I guess
Does anyone know if there's another word for a polearm? I can't generate this dumbass weapon π
You can try naginata, lance, spear
pointy stick?
Ah yes, of course, how did I not think of this π
Yup, lance seems to do the trick
lmao
looks like unsing nirmally break to gen first a bg and than the character is a bad idea
you get a fantastic bg, but bad characters π¦
why i need to add portrait to get something decent?
Portrait is one of my favorite prompts honestly
adds so many details to the image
Finally got this stupid weapon to work
Big armor
without it, i just get blobs
Cx
i wonder how it works on comfy
Yup, her full set has some sort of lion headgear but it looks nicer this way honestly
Who needs head protection anyway
i'd blame it on whatever model you're using. i never use the word portrait in any of my gens; the better models don't need like 2 paragraphs of text to make work usually.
Pretty much yeah, it all comes to down to what model you're using
yeah, the double paragraph was an experiment to gen a detailed bg and a good character, like i do on comfy
but still, without portrait, the result is meh
not too good, but still passable
Try adding fighting stance to your prompt
let's see
Just her standing there like that looks a bit too stiff
yeah, not caring about the stens, it's the face that is alway so lacking
fighting stance, attack, slashing, those prompts usually give better composition, but then again it all comes down to what model you're using as always
compared to when adding portrait
Try using ADetailer, that should fix it
i suppose i should use it again
If you're on Comfy idk what ADetailer is called there but I know there's a version for that as well
nmever mentioned face details in these ones and they come out fine
What I like to do sometimes is, I generate with NovelAI and then img2img with Pony/AutismMix, might work with other models as well
hands r another issue but those can be fixed with inpaints
I love that anime style, looks amazing
I usually just fix them with photoshop π, I think it's much faster, but that works as well
if ur gud enough at mimicing the hand-drawn style lol
that is
the style the hand is drawn in
Shouldn't be too difficult
lol realized i wasn't even using masterpiece, best quality stuff for these ones
But then again if the hand is completely mangled then yeah, Inpainting is much faster
Not bad, I'm guessing the results weren't that good without the portrait prompt
yep
Mmmm warriors
Thats the.... ehm.... Something something armor Lora, ye?
Ok there frendough. What checkpoint and lora u used for that style?
nah, i don't use lora
Pony
hotarulab

Mmmm Pony
Put a cat in your profile picture
It'll make your gens better
(ζ₯ζ¬θͺθͺ¬ζγ―εΎεγ«γγγΎγ) "HotaruLabs" series are experimental(intermediate) models, used to develop my "Breed" series checkpoints. 7 types are uploaded. To ...
I tried to Anime-Fy my cat... But i couldn't figure out how! π¦
Nai 
No nai! REE
Use controlnet
Here. Go animefy the cat only
same here with Nai
nai is just top tier for now.
Nai is $$$
i'm pretty sure i actually save money using nai cuz my monthly pc electricity usage lmao
@coarse brook go shopping for art style loras if you want a different look x3
almost
(ζ₯ζ¬θͺθͺ¬ζγ―εΎεγ«γγγΎγ) "HotaruLabs" series are experimental(intermediate) models, used to develop my "Breed" series checkpoints. 7 types are uploaded. To ...
I am
But shopping for free ones
Civitai your best friend
Don't mind the background I totally wasn't too lazy to change prompt
Its not π
Its a friend, but not best
I type too fast for the search bar, and that triggers me immensely
cat entered the adventurer guild
i didn't prompt a double horizon, that's laputa <.<
the Pu**Y cat?
Civitai search is horrendous
PuTTY
im2img
yea i only find out about new civitai models by reading metadatas from interesting images, or it having been mentioned somewhere i follow
Ye. Have to type 1letter at the time, wait for it to search for the word + that letter
So is HuggingFace's
you usually land on huggingface links from somewhere else to be fair
i only understand it cuz i used to upload rvc voice models there
Lmao
I find the HF UI fine, better than Civit at least. But searching is crap, or more like, it lacks basic shit.
Civit is obviously no better.
ummm...
I mean..?
hehe
inpaints whole image
What's there not to understand? 
It's just git repositories, and probably the easiest interface to git repos ever.
Everything
Pictures. We want shiny UI with pictures
cause it allows you easy access to go git it and git'er done?
If you don't use huggingface, it looks alien

Ok ok ok. I'm liking the outcome of this HotaruLabs model
i mean the only real difference is there's a different page for files & version than the about page? it's actually easier to download individual files off hf
Yeah, there's like a cover page, instead of just showing the files and the README below it.
I get that it's confusing for non-techies, it's not very clear what to download and stuff
You mean github
ye
can we go back to tortoisesvn days?
please no
tortoisesvn was much easier than git
no SVN
lol
also i never realized ash blossom & joyous spring has a tail
i don't even remember what i was using svn for back then, but i used a lot of projects that required svn
ash who
the char's name is ash blossom & joyous spring

shes t-posing on the second pick
Infamous Ender cat that only steal your slabs?
prompt was a square hat
lmao
okay.
Ain't you gonna make my cat into something scary
Everything is catgirl if you are brave enough
well no but maybe
Well I asked for creepy, not scary
Me wondering why my new gen takes such a long time.... Forgot to change from Batchcount 12, to 1... (Was taking bio break)
I mena, all things considered it could be a lot worse
true
Sadcat emoji indeed. Not even mah cat π¦
this model is not creative enough <.<
ummm.....
Well, I used Fill instead of Original on that one.
@coarse brook smuggiest cat in the world
AH HA! done
I could almost train an Ai, with the amount of pics i has of her xD
grabby 
ok im done
Rawr
shooo many cats
cute
whai r males still so hard to makeee kekw
took so many inpaints on eyes and still messed up ;_;
Just make femboys, ez.
But please don't
lol
most masculine anime character ever
was just doing something related to persona 5 and thought i'd try ren amamiya, but the eyes were frustrating so i gave up. i should've used makoto instead.
one day males will get more representation in fan arts XD
challenge unaccepted. made a vaporeon instead.
Whatcha doing with that peppermint? Can I have it? π³
already consoomd.
for some reason adding the vaporeon ruins her hair, cuz it's supposed to have the lines of pink.
What are twists of pink, if not a long line of pink?
Unrelated but feeling like my life hit a new low bruh π
I just pitted two chatbots GPT and Gemini against a dice throne rip off game
i think gemini just still barely edges out the lead in nearly everything, espec with latest update, but gpt4o isn't available on copilot yet
Can anyone give me an ELi5 & TLDR version, of what a Hypernetwork is..?
Its barely smart enough to play the game
And gave me a bit of an entertaining game
ELI5:
Imagine you have a box of crayons (Stable Diffusion) that can draw anything you can think of. A Hypernetwork is like a special filter you can put on the crayons to make all your drawings look like a certain style, like a comic book or a famous painter. It doesn't change the crayons themselves, just how they color.
TL;DR:
A Hypernetwork is a small add-on for Stable Diffusion that changes the style of the images it generates, without altering the core model itself.
do people even still use hypernetworks anymore? it was a big thing for a while but i think everyone moved to loras & controlnet
I c. Fanks.
Well, I c some stuff on Civit sometimes I like the look of, but then they're marked at Hypernetworks xD Soo
I am still having major issues understand what, and how to use controlnet, in ways I seek. Most of the guides I've looked at, dabbles in more realistic images
Whereas, I'm more in the 2D, 2.5D area. Anime etc.
Such as this
as far as i understand it, controlnet's main benefit is you can set the scene using another picture as reference, so the character you make is in the same kind of pose for instance
to do things that otherwise wouldn't be possible, are hard to describe, or if you just want a hyper-specific scene
Aha
I bring forth a new question
Using Wildcards. Would this work as a combined promt?
! NOO! π¦
Everything was looking like it was going perfect. Shield orientation, hands, everything! π¦
so use the same seed and do it again
It didn't save the seed, cause it crashed generating the image π¦
But just ran the same promts again, and got this, so we can call it a happy incident
Which.... Looks very, very similar... Odd, its on random seed, and random styles. 
What... What is causing hair to become like this? Too strick of a CFG?
Unrelated. But currently pitting three chatbots in a dice throne rip off. I gotta sleep bruh
i still dunno wut a dice throne is
Its a board game
i've been playing munchkin lately a lot more. probs only board game i've played in the past year aside from 100% orange juice
Well ive not played dice throne myself
no cash bro. the only complex game ive played is here to slay that shit fire
i also definitely have trouble making ai generate anything that looks like a board game
that looks like ygo to me
yeh, rosemi's played it a number of times & duelist of the roses was a yu-gi-oh game
ik
i think mimikyu put the wrong costume on
Tries to make cool viking. AI gives me completely nekkid person 
that's just ai in a nutshell.
i made an ultra complicated workflow, till the last week, it worked perfectly, but now, it just gives me landscapes π¦
Can you use a a lora, thats made for an older model (SD1.0), on a newer one, such as SD1.5, without any major hiccups?
There aren't any SD1.0 LoRAs. SD1.4 and SD1.5 are compatible. SD2.0 and SD2.1 are compatible. And SDXL0.9 and SDXL1.0 are compatible. These pairs are not compatible with other pairs at all with the exception of the VAE which is only cross compatible between SD1.x and SD2.x versions. SDXL can't share anything with the others.
I read it wrong. Is SDXL 1.0 π But If I'm using the (Screenshot), does this mean they're compatible? I strong assume so, but just wanna make sure, since my brain read it wrong
Sort of. Pony-diffusion-v6 is in fact an SDXL model, however, it was trained using silly parameters which has made it differ from the base SDXL model so much that it doesn't really work with things made for normal SDXL models. That includes things like LoRAs, Controlnets, and even merging it with other models that weren't trained from Pony. Attempting to do so will often lead to completely broken images. People who are extremely experienced at merging have tried everything to make merges with it work, and it's always a downgrade from the base models. If you want to merge it, stick with other models that are also based on Pony.
But this isn't the basic v6 model though afaik?
Yeah, I have no idea what that merge is. The name implies that it's Pony V6 merged with base SDXL which is weird and also won't work well.
I mean, its not doing a bad job
It probably isn't what the name implies that it is then. It looks like it's Pony based. If you're not frequently getting images that are nothing but colorful noise then I highly doubt it's just Pony and base SDXL merged. Unhinged filenames aren't uncommon on Civitai.
Aight. I mean, if it works, it works... Right? xD
Yep, if you're happy with the model, that's all that really matters.
Tru tru
You'll want to look for LoRAs that are specifically made for Pony to use with it for best results. You can try others, but they'll be much less likely to look good.
Oh i am, just found this neat pic, and wanted to c if i could somewhat replicate it, https://civitai.com/images/13132472.
That is a nice img. Backgrounds are a weak point for Pony so you'd probably be better off with a different model for something similar to that IMO. They show the model and LoRAs they're using in the img info if you want to try those out.
Yep, when it says that, then it should work with any SDXL models that aren't based on Pony.
:Hmm... Was wondering why a Lora i had DL'd just kept ignoring my promts... Even overwriting [[ and ((... UNTIL i checked the activation promts in its settings... I had the Lora main promt (The <LoraPack>) twice.
Whoops
hmm?
the workflow is stright up overpowering 1 of the 2 conditioning
no clue
shrugs, I tried comfi.. to many moving parts
that's what i like about it
holy canoli!
hmmm?
yeah that doesn't make sense
uhh oh
bbl gonna go lay back down
last one.. thud
sneaks one last one in and runs
no problem if skull with smol but not with big π πΆ
Here, fixed
x3


