#🍥|anime
1 messages · Page 218 of 1
the reason is because that lora is a SD1.5 lora, and the model you are using is SDXL based
nope. i am not use SDXL. this is SD 1.5 model Indigo Furry Mix v12.0 Hybrid.
Nyaaa!
oh noooo!
"somoone's sick?"
oki-doki! our clinic has the cutest nurses
get well soon
and woke up sick?
no
I don't like fish that much
If it's anything related to good, is probably because I didn't eat enough
student?
No, I was not feeling super good during the day to have big appetite

here is the result. nerves, malnutrition, and the absence of fluffy rabbits nearby!
No bunnies are definitely the biggest problem

are you asking for like.... actual base 1.5 model?
I can get you a link if you want

maybe just download any anime SD1.5 model from civitia and add LoRa too? for example Anithing or BluePencil or BlazigDrive or Aurora. And put it in model folder.
oh I know but he didn't specify so I gave him exactly what he ask for

Any model in civitai with this detail works
this looks like and act of blasphemy, you should report yourself to the closest reprogramming center
I find it funny how some 1.5 lora work on SDXL anyway
maybe some blocks are in common
yea
i don't understand why 1.5 have 26 blocks, and xl only 20, when merging
being the model bigger, you expect more blocks lol
I guess it's just less but bigger block overall
makes sense
and the question was about the SDXL model and SD 1.5 LoRa?
yea he tried using 1.5 lora on SDXL
hence why he ask around for 1.5 model

so I shrimply gave him one
he's a pervert!


in the prompt it is written as a white blouse and a black skirt, a bow and a ribbon,
just look PNGInfo
it's light. It just looks big and heavy. maybe... I'm not sure.🤔
looks like sd3 will be available only for api, for now at least
we'll just wait.
Pirate cat
can we do inpaint with sdxl?
Yes
BIG belt

yeeeep!
API released
Is it exclusive or can I use it too?
I have no clue 
@lyric zealot?
Where can I access it?
but cost 6.5 credits per generation
Check #📣|announcements
you can pay to get 1000 credtis tho
how?
since it's api, some
needs to be done
Do you get any for free?
25 for free
What you mean
for any user that sign up with google account
It's the same thing
Nice
yeah thats what I mean its the same, doesnt work inpaint
so basically about 3 generations
You're doing it wrong

I see i see, so how do I need to do it?
I forgot the right options for inpaint
SENSEIII
@lyric zealot be useful

what do you want
I'm assuming I need to run it in visual studio or such?
i guess so. Im doing it on vscode
inpaint SDXL
Same
if you know how to inpaint in SD1.5, then isn't it the same thing?
ouh do I need any specific sdxl model for inpaint? like sd1.5 inpaint?
I love vscode

it's my favorite
it doesnt work with sdxl for me
its just blurring image
thats it
def generate(prompt=""):
response = requests.post(
f"https://api.stability.ai/v2beta/stable-image/generate/sd3",
headers={
"authorization": f"Bearer sk-YOURAPIKEY",
"accept": "image/*"
},
files={"none": ''},
data={
"prompt": prompt,
"output_format": "jpeg",
"aspect_ratio": "16:9",
},
)
if response.status_code == 200:
with open("./lighthouse.jpeg", 'wb') as file:
file.write(response.content)
else:
raise Exception(str(response.json()))
def main():
generate(prompt="anime girl with a t-shirt saying 'SD3'")
main()```
for people that wanna quick and dirty way of testing this
^
just replace with your own api key
and put your prompt
but there's more stuff you can do
what are you trying to do here?

remove the ribbon
?
It's doing the blurry thing
yesh yesh
Instead of generating something
souka, I'll try
give me a prompt to try on SD3
with text

I have no idea
it saves to lighthouse.jpeg
Where is that located?
Figured
Generate a cute anime style potato that says potato over its head

I wanna share the images I made on the SD3 bot but no one said im allowed now 
Thats a rather big sus then
yup
the text is perfect as well

looking forward to adding more text in future generation
I have a list with stuff that doesnt work super well let me see if I can give you a task 
So I found that fish still generates with too many fins for example, or if you generate something infront of a face the face gets distorted
Also the model doesnt know kim possible for example 

It's really damn good
I wonder when they'll release this as a safetensor file
Im not sure if im willing to test more by putting in money 
3 generations are just not enough
but damn the price is expansive
probably cos it's on their largest varient of the model
Oh yeah people interacting like hugging fe is also an issue
cant we change fonts or smting?
You can try 

How large?
nvm
Try prompting it
the 8b varient
i mean that's what im guessing tho
don't take my word for it
In GB lmao

How can i code to sequence the lighthouse so it doesn't overwrite?
if we think about it kinda logically
SDXL is about 2.6 billion parameters
and that's like 6gb
so maybe around 15gb?

I'll never be able to run it locally
the easier way is to make it so that it save it as prompt
or something

look like colored scetch.
i don't know why
i can still generate
without credits
is it intended
or is it not
same
still sucks at hands
(I'm going to abuse the everloving shit out of this)
The moment they made the API available the SD3 bot died 
crazy
so close yet so far

Make it do some anime style
Like death note or something

cool thing i learned. so whenever my lora is not giving me stuff i like, put another one at a low scale
there's also negative prompt as well
if you want to add it
somehow works idfky
isnt dat small?
I did

what neg are you using?
ok
i can probably code it
so you can use it on the discord bot

I think sd1.5 and xl are consistently better, but when this is good it's great.

i think currently we just don't know how to prompt and stuff
and also
the text is way better definitely
That's true
a blue cube on the left, a red sphere in the middle, and a green cube on the right
this is wrong tho

It's so close
Try again but put above water

Stone, metal, glass

a blue cube on the left, a red sphere in the middle, and a green cube on the right, above water, an anime girl is sitting on top of the green cube saying 'SD3'
kinda funny

a blue cube on the left made up of wood, a red sphere in the middle made up of glass and it's see-through, and a green cube on the right made up of metal, above water, an anime girl is sitting on top of the green cube saying 'SD3'
prompt
a blue cube on the left made out of reflective metal, a red sphere in the middle made out of glass, and a green cube on the right made out of marble, inside them are images of cat faces
is it just me or anatomy looks strange asf
It's goofed
i think so too
Not just for you
no nude in dataset bili:
When it releases I'm sure it'll get better

Oh boy
NSFW models are criminally underrated
people are doing much better than we are here
They're not just for porn

Sex sells
How is that relevant to anything I just said?
It that with sd3?

What's your negative?
(worst quality, low quality:1.2), ugly, 3d, censored, abdominal muscle, artistic glow
im just throwing my old SDXL negative
and test
bro is SD3 truly only API for now?
open ai amirite

maybe theyre cooking maybe they will open source later
i mean cmon that anatomy aint doing shit
bruh
guys where can i test it
this reminds me I need to update my comfyui file paths
I just have it increment i by one every gen
it's so simple
true
i don't know why i used timestamp
I overcomplicate things too
did it block the boobas?
yup
just prompt this (mammary glands:2.5)
1female, mega mamms
1900: maybe in 2024 we'll have flying cars
2024:
I'll take this over flying cars
👍
You lack sufficient credits to make this request. Please purchase more credits at https://platform.stability.ai/account/credits and try again.

well it was fun while it lasted
(Masterpiece), (Best quality), (Ultra HD), (Super detail), (Whole body :1.2), 1 girl, Chibi, cute, smile, flowers, outdoors, holding the camera, sitting on the roof looking out into the distance, with mountains in the background, amber, warm yellow, sunset, artistic sense, Quadratic style, white clothes,
(Masterpiece), (Best quality), (Ultra HD), (Super detail), (Whole body :1.2), 1 girl, Chibi, cute, smile, flowers, outdoors, holding the camera, sitting on the roof looking out into the distance, with mountains in the background, amber, warm yellow, sunset, artistic sense, Quadratic style, white clothes,
Error: This is not a bot channel
definitely need to finetune
there's currently no bot available here. Check #1047610792226340935
ik
but man the direction
i am not very optimistic
yk when half of the announcement is "safety"
Ong
Hmmm, well if the release it with the weight
It's like a poorly trained 1.5 model
Well to be fair, when the base sdxl comes out
It’s kinda like this
I still remember the sdxl0.9
It was bad at that time
sd3?
yeah
Thanks
I could
they took it
I'm going to make another alt
Lmao
Every account is like 3 generations if the credit works properly
Is it worth it for that much work?
would crack
nah, but i got 86 gens in so i don't really care
what prompt?
you know that meme where eggman says "I have 70 alternative accounts?"
that's literally me

these look great, prompts?
(Masterpiece), (Best quality), (Ultra HD), (Super detail), (Whole body :1.2), warm colors, sunset (anime style:1.2) 1girl, highly detailed, absurdres, insane details, solo, long hair, underwater, full body, wallpaper quality, black track shorts, black crop top, best quality, masterpiece,
It's really a roll of the dice


model ?
I sent the first message
💪

how could you
i saw your fart tho

It was the mods
My life will never be the same after this

there are people who've ran 13b text models on their 8g 3070
yup
And sd3 goes up to 8b
That is true
Last time i offloaded to cpu I bluescreened
or they quantized it
I wish there was a way to load chunks of the model when you needed it

It'd probably be slow though
i don't think i understand the way model works enough to know if that works
Same, it's only hypothetical
But sd3 goes from 800m to 8b so there's probably a 6-7b model for my 8g card
yup, from what i know (can be wrong), there's like 3 options atm
so one low tier, mid and top i guess
so you're on the same boat as me?
yup, but we still better than @nova remnant

what card do you have?
a cat with only 4gb vram 
Rx 580
I just use online services and play Poe
Smuggered
sd can run on amd cards now?
I mean
From what I know you can grab the AMD branch instead of Nvidia one and run it
Poe?
An
A bit scuffed on some things but works
Rich
i guess it was when i won the #cow
actually didn't win but 2nd place 
I run amd, I'm just stuck with using SD1.5. SDXL requires voodoo magic regarding zluda and SDnext and I have no idea how to make it work
damn
like this voodoo
?
I don't think that can run SD in my opininon
world's strongest amd gpu.
🙏
I got 3.2k hours since legacy league and only killed maven once.
I get bored and quit usually around the start of red maps

I play a bunch every league

this league is pretty decent, much better than last one
still not finished but wtf is that head
wow a JET head now that's new
imagine the headbutt
million years of pain
I'm suddenly reminded of Reze
What's the deal with the "score_9" etc in pony prompts? I've never seen it actually explained where I could understand it
When the model was trained images it used were given a score this was so that later people could control the quality in a more numerical sense and it gives greater control
so putting a higher score like 9 makes it try to generate images closer to things scored as 9? is 9 the best?
yeah
should i put the other scores in the negative?
But you don't only want to put score_9 or you will be limited
well the best place to learn about it, is to go to the civitai model page, and read what they say about it
What I want is to play this character, but what happens? That when I do it, since the series is called Ranma 1/2, and at the same time Ranma is a character, what it makes me is Ranma, even though I describe the characteristics of the character, it makes me Ranma with the characteristics described of the other woman. https://civitai.com/images/1029666 any solution for that problem?
apparently the best place to learn about it was to listen to what orax said because he had the most understandable explanation so far
while I dn't need all that
This is what I do
Pos
score_9, score_8_up, score_7_up, score_6_up, source anime, masterpiece, high quality, BREAK (prompt here)
neg
score_4, score_5, score_6, source_furry, source_cartoon, source_pony, monochrome, (censor), 3d, realistic, render, CGI, text, watermark, cartoon, sketch, greyscale,
well, fine by me
Smuggered
I'd like to see a difference test including score in the negatives
And without

same, was just about to look for that
help pls
And I don't think masterpiece high quality exists really on pony
I'm not sure on that myself it seems like it helped but 🤷♂️
shiney
you wanna use the lora you linked, is that what you are asking?
First one's prompts:
p
score_9, score_8_up, score_7_up, source anime,
Woman```
second one with these negatives as well:
n
score_4, score_5, score_6, score_7
^
makes it look better imo
I still think first one look better
sure it look more grumpy but 2nd one's nose look wrong
there back to anime character 😄
there's a new sd3 channel 😬
how much VRAM does SD3 need to run anyway?
no idea, I didn't even know it was out
not fully, api only rn
right on
same difference to me
difference is pretty slight
one has better eyes I think, the other better other things
or use the parrot with the version I have, or be able to make that character without parrot, so that it detects it directly, do you have any ideas?
what model?
Apparently the 8b parameter version will require approximately the same VRAM as running a 13b LLM from what I've been told. There will be smaller versions which will require significantly less. The architecture is more flexible for offloading as well so even if you don't have enough VRAM you should be able to offload some to CPU/RAM like you could with an LLM.
I have an 8g 3070, how much does a 13b LLM require?
lol hopefully I can run it with 12gb
without offloading and with quantization it'll just barely fit in 12gb
Can you even offload an SD model?
With SD3 you can. It's a lot more like an LLM than diffusion models that we're used to.
Praise the lord
Will it cause a significant performance decrease?
yeah, cpu is still worse for the task than GPU is, but it should be a lot better than the experience when you run out of vram on current SD models.
By what margin will it be slower by?
Between debilitating and insignificant
depends on the specific CPU and GPU as well as probably the speed of the system RAM to a lesser extent.
ok
im gonna be honest, i have no idea what you meant
create this character https://civitai.com/images/1029666 without using lora, because it confuses me with the name of the series, and it makes me the protagonist of the series, the series is called ranma, and the character is called shampoo, when you put ranma series, it makes me ranma not shampoo
If you want to generate something without Lora, then first you need to check if the model you are using knows the character or not.
If not, then you will need to use a lora trained on that character
The problem is that the lora that exists is not compatible with the version of stable diffusion that I use
Yup. So you can do a few things. You can either use the same kind of model for the Lora, or try finding a Lora for the type of model you are using, else, you will have to train yourself
And for training, I know next to nothing about it. So yeah
Your best bet for now, is to use the model that the author of the Lora used, so you can pretty much get the same result as the author of the Lora
https://civitai.com/search/models?baseModel=SD 1.5&sortBy=models_v8&query=shampoo Could you tell me if any of these would work for my version?
But how do I use it if it doesn't work in my version?
What gives you the impression that the LoRA you want to use isn't compatible?
but I don't have that version, and if I change versions all the time just for one character, it gets uncomfortable
isn't good in the bad, maybe 🙂
1.5 is a model version, not the version of Automatic1111. Those are different.
what version of model do you use?
It doesn't appear in the list, and someone above said it's for version 1.5, and I have xdls or something like that
The model he referenced was a 1.5 model and would be compatible with the 1.5 LoRAs.
sdxl, ok, it's different from 1.5. you need an xl lora model
The model used in the link for the image:
https://civitai.com/images/1029666
is a 1.5 model.
That's what you posted.
If you're using SDXL, then no, the LoRAs won't work.
You would have to use SDXL LoRAs.
Yup, so basically the model you are using is SDXL based, and not SD1.5
but I can't find it for that character
And sadly for you, there’s no SDXL Lora for that
Then use a 1.5 model.
There are several solutions here.
this model have data of shampoo too
Or that ^
How do I do that?
you simply need to add the right tag to the prompt
How do you use a 1.5 model?
Guys, I’m leaving this to you guys, imma go sleep
You download it, change to that model, change your image size to something like 512x768, and then apply the LoRA.
nighty night
peace smol
which model?
Whatever 1.5 model you want. Just go on Civit, go to Models, filter by SD 1.5 and find one you like.
1.5 lora are only compatible with 1.5 models. take the model you like more
maybe chose between the most popular once
and newer
newer models usually gives better results
but parrots only appear if I do that
what parrots?
das lot of limbs
cute

@supple raptor what was dat xl model ur using for upscale NAI images? kohaku xl?
beta or delta
realcartoon3d I believe
Thanks
yes. delta is what i've been using, epsilon (newest version) might be even better for the job, but i havent tried it yet.
lora
you need shampoo and parrot lora at the same time?
if parrot is xl, you can't use it together with shampoo 1.5 lora
It doesn't do it right for me, it confuses me with Ranma again
oh, it's pony model. pony is not compatible with all xl models
pony is 100% compatible only with pony lora
And what model do I use then? I'm lost
first try the right tag
to call shampoo, you need to copy the right tag from the civitai page
in your case, the right tag is "shampoo (ranma 1/2)"
use this instead shampoo ranma anime
if you still fail, try with the model https://civitai.com/models/260267
thanks, I'll try
In that case, is the lora you put compatible?
yeah, that lora is based on that model
pony models are very good, but too much trained and so not compatible with most of the other xl models
what is this VAE (319.14 MB)?
vae is a model that works together with the main model to gen images well
without it, the result is a big mess
but how do you use that?
if you use auto, the drop down menu next to the drop down menu of the model, on the top left
place the file inside the vae model directory
thanks
Is there a way to see in civit, the prompt that generates the image, without having to download an image and look at the info?
when you click on the image, usually is on the right. if they are not there, means the image doesn't contain them
thanks
@vital raptor @steady grail How could I make a character hold a bomb and try to manipulate it? Are you planning to deactivate or activate it?
create a image of goku super sayan 3 with full power charged design for tshirt printing
gib prompt
evenin'
good idea.. get to work.. yourself 😄 oh and there are no bots here
thick tail
@nova remnant

:D
I love goth girls 😍
I need to improve my creativity of goth prompts
gonna finish this episode and then bed.. tired.. blah. thud
submitted the post - accepted the post
Look who's here
My favorite nurse prompter

are you still sick?
it good
one hemisphere has already gone to sleep, the other is just waking up. how small is the Earth
you almost guessed it right. with hair color. and height.
hmmm where lost her horns?!
or just a very small dragon. In any case, this is not quite the right dragon
Hornsss
hmmm
my video card can never keep up with you. it sad,
am i cute pie too? 
of course
I'll need to do a test to verify it
A physical test

X3
Reminded me of Yoshi
Never enough dragon girls
Friren + horns?
and random pointer ears ^_^
I should try floppy ears
Don't know if pony can do them
try

Doesn't look as cute
Furry x3
Most of the time they are coming out normally
she look cute
It’s just not clear who she is?
cat selfie 
She hides herself
We'll never know 
floppy ears spy-girl
grabs
Is she upset that she didn't catch the crab?
shes mad at u cuz u cant swim 
i can swim
Spy from bunny land
😵💫

Miss knight I'm not innocent take me to your house as punishment
👍
always amused Illustrations of girls knights in iron panties
There is a tag for that specifically
Cx
It's called bikini armor
such cool armor is drawn! and iron panties on her.
I would suggest the artist to try out his creation
I only put clleavage on the prompt, I'm innocent 
now remove her armor 
Genshin?
da
not everyone into anime yk
UH HUH apparently fanart-the-model xl 3.1 doesn't know touhou
trying to generate some touhou for waifu diffusion info payment but nah it doesnt know
w h y. Like i dont know shit about blue archive or arkknights or something cuz i dont actively engage in anime games or shit. But touhou, every one on this fucking ball of dirt at least knows reimu
but the creator of the model apparently has a completely different opinion
apparently
like cmon man, everyone, their grandfathers, their pets, their monkeys, their godzillas
i say "oh you know reimu?" everyone knows
its that popular
until recently I didn't know about any of the Touhou characters
Oh dang, how'd you prompt the ears?
1girl, standing, dress_shirt, black_pants, black tie,
that's the whole prompt
Yeah they intentionally removed basically all of the Touhou from their dataset. 3.1 is a little better, but it's still not good at them. kohakuxl delta/epsilon are much better for touhou. Otherwise you're going to want a LoRA for every Touhou you want to make on animagine. Even Pony usually does better at them.
Bro why
How am i going to make images of touhou characters to pay for more wd info 😭😭😭
hmm its opensource afterall
maybe they work on kohaku?
It'd be worth trying them on kxl. They might work well enough.
pls @supple raptor accept this mid ass pic as payment for your great wisdom
I offer 1 reimu in return
tPonynai3
That my best of the best generation! my SD broke again
I offer you feet... I mean Marisa
I think it's much harder to generate good feet than hands, what I made look like chicken feet
mmmm
first or second?
both, but i kinda like it
first ones pastel is a bit muted
maybe try it LoRa
https://civitai.com/models/35847
well lot of OG member doesn't show up anymore like @wispy canopy

and this channel is quite tucked away
oh hey there he is
btw. i very love model helloArt Oil
oh? sneak me an invite in DM
how do i uh use the lora
You have already indicated the lora and weight. if lora has a trigger word then you need to write it
also pay attention to the fact that the model and lora must be the same version. for example both SD 1.5 or both SDXL
for this is lora type it lora:OilPaintStyle_v1:1
yeah, i love that lora too
interakson
I love interrupted stuff before it get a chance to start giving it interesting feels
like half asleep shit you see in corner of your eyes kinda feels

It's on pony
"bunny girl, (floppy droopy ears:1.3)"
I only know those character names because nubby spams them

Not you 
You're gonna send tomatoes there too 
Don’t be jealous I know you want it to be exclusive in your DM but I get around ya know

You can’t tie me down
Oo, will try it out
whats dis model?
nah lol
droopy
help?
maybe u can use image to image generation?
It doesn't work for me, it doesn't do anything for me
agree
canceled image
goes back to watching tokyo drew Nezu at the Shrine's Azalea festival [LIVE]
and doing art
literally same answer as before
canceled images
hands 😭
👍
My colorful Hoodie Club selection.
@native halo
are you hurt somewhere why you looking down
those belts are gonna cut off circulation bruh
Like Jesus Christ (not present in this channel) how do you get a belt buckle that tight around yourself
smol

(tight belt:2.5) 
its not working
stupid model
we need sd3 so we can "tight belt:2.5"

tight
@native halo
not enough
she can inject me anytime
"preschooler" lmao
the main thing is not to tighten the belt on the neck

now this is beltiful
is that not what you prompt for all your gens here?
Fruity, the mod 
don't call me the m word
🤣🤣🤣🤣🤣🤣 noooooooooooooooo
you mean the style ?
I'm just saying your whole history is gens of kiddies so was curious your prompts
yes only kids 👍





