#🍥|anime
1 messages · Page 33 of 1

it
hates faces
as @magic wave said, face restore is bad for Anime. It is only trained on realistic faces. If it finds the face at all in an anime image it will almost always make them look like some weird uncanny valley shit. It usually looks pretty bad even on realistic images.
But I also think its just bad to use it
good observation
how did you come up with it
Well, I've read about it fucking up anime faces a few times xD
These days we don't need it.
True
It was made for a time where it was helpful.
Just inpaint 
Looks like you're inpainting and have "Fill" selected instead of "Original"
We need an AI clippy that says this.
We need Automatic to set default values that actually make sense for how the tools are typically used.
Face restore doesn't work very well with anime model
i just updated the webui and the image browser became blank
any idea how to fix this?
update extenstions? close webui and relaunch
i didnt try updating extensions one sec
You're still running the old version of the image browser that was abandoned in early-mid December. It isn't compatible with the newer gradio version that WebUI uses now. You need to uninstall that one and install the new version which is here: https://github.com/AlUlkesh/stable-diffusion-webui-images-browser.git
also to add to this delete whole extension in \stable-diffusion-webui\extensions
spicey
thanks!
its working now
though its a lot slower
It has to build the database to begin with. Once it has finished that it should be faster I think and it also supports regex search to make finding things easier.
Aaand I'm out for the night.
interrogated this image and used it as img2img
the result is
not similar at all
giving it a second shot with .6 denoise
A person looking at the night sky, galaxy, stars, celestial bodies, planets,
lesse
hii
are you running SD locally?
Man I just woke up to hear that my model was removed from civitai.
The reason is that the preview image was SO GOOD that they thought it weren't even made using the model. I sent an appeal but it said it usually takes them 5-7 work days. Fuck that man.
damn, that's sad
typical mods moment
huggingface exists I guess XD
Like look. This is actual perfection
You can even look at the PNG info
Well technically someone could inject whatever PNG info they want
But still
breh
actually true, they could just download the png and check info XD
it is not that hard to download your model and check if it is right
Yeah I sent them this image and also this one on the appeal
Everyone is either too lazy to do that or thinks it's a virus because I first uploaded it as a ckpt file.
Relatable
Bro is multitasking here and on projectai
Oh shit what major?
artificial intelligence
so kinda learning what I learned myself a lot, but with some twists and updates I guess
also i have a bunch of programming too (atleast I had so yeah.
fr fr
I hope you make waifus real as your end goal
if i'm interacting with artificial intelligence, does that make me "real dumb"?
Yes

I mean AI is smart, so no??
well, i've been improving on my LLM bot 😄
Smarter than 70-80% of humanity
You can be the first one to make LoLiCon
they're pretty smart indeed
LoCon but with Li
dangerous language
Idk what you mean
good. keep it that way
👀
lovethat style
In like eighth grade, in one of the preparations for the ending ceremony of middle school (we ended middle school on eighth grade) I downloaded tor to my phone and tried searching for cp sites and then reported them to the FBI.
or prompt
XD
I fed my LLM the lyrics to the chalkeaters - bowsette
nice
have to sacrifice 1000 tokens just for context tho 😦
Tokens? You don't run locally?
i do. but i have to sacrifice 1000 tokens to "initialize" the LLM to get it to do what i want
unfortunately, i need to give it 2 examples
still, you have a max context lenght even localy, usually 2048 tokens
By tokens you mean tokens?
yeah, with lyrics it's about 1600 tokens
tokens being either a lettergroup or complete word
yeah, it works the same as SD on that side, it splits your "prompt" into "tokens" using a tokenizer
so i have a HUGE prompt that goes before i actually can dump the lyrics
Yeah ik
I ran Pyg6B for a while now
vicuna 13b, the 7b models just can't properly interpret lyrics
You mean the 7b vicuna models
Can't be sure about all 7b models
If someone fine tunes one just to do that
For example
true dat
Or a single letter, or sometimes maybe even more than one word?
yeah, i've seen double words become a single token in some instances
Guys is there a way to install pyinstaller to android?
Very rare probably
Never saw that myself
basically only happens with "words" which only consist of 2 words
A cool way to encrypt messages is to tokenizer them in the tokenizer tab on a1111
And then send the number combinations
Then to decrypt you paste them back into the tab
and you generate a random tokenizer from a common seed shared with each of your contacts, so that the number combinations can't be decrypted without that key, encryption of tomorrow (yesterday :p)
Oh yeah guizmus did you see my new model
I actually managed to overtrain a model on 1 image, so it becomes like deterministic or at least mostly deterministic
I am a happy man
Yeah they stupid
a 2GB jpg
They literally took it down
PNG
Worth it
if packaged as a hidden joke, it would have been better
And you can also do this
like you sell this as the most realistic model ever
with false pics
and there is only the kitten coming out
it would have still been taken down, but have created more confusion
I wanted to unveil it first. Now it's time to make a rick roll
package it as real model, add a model card, and release SD 2.3 on hugging face with it
HF will make it more realistic
Yes yes

Also gotta put a ton of work into readme
such a modular model bravo
add some graphs
people trust graphs
Also, it is made using a lora because I am not sure my gpu can handle dreambooth and stuff. The problem with that is that the lora is so overtrained that it is completely useless on any other model except the one it was trained on
So I just merged the lora with the model
when you want one girl but harem life says no:
I think I saw auto generated chapters on YouTube before but they also give such good titles?
holy crap
seems like niji v5
yes niji + editing with inpaint
i love the niji style of --style cute
it makes this masterpieces!

Kinda forgot to turn on dynamic prompt
Anyway, now I continue mommy posting
i love it too, i want make a model on this style
cute style
need to prepare dataset
maybe you can help me with dataset ?
I need about a hundred images made on niji v5 --style cute to train a good model
im not realy interested sorry to much work i have
Oh okay, no problem

Fair, you got the lyrics, I got the mommies XD
Also damn thats one cool pic :o
Anybody got any tips for achieving this specific anime style consistently?
Seems like a pretty standard style
I mean, you could try adding style tags and figure it out, or maybe try to finda style LoRA that can replicate it
rmoved that one, some elements too inappropriate upon closer inspection
I walk a lonely road
i noticed my llm prompter has a specific preference, so not figured out how to proceed from here 😦
model. for the love of god, can you please stop making cameltoe XD
1girl alone night walking midnight chicago middle street
i love all of you 
Nice

@dusk seal you won't be copying these prompts using interrogation though
these are my abominations 😛
but, have a taste:
It only takes one lone soul It only takes one in a thousand The absence of fear in your eyes No, that's not bravery As children we learn it's wrong To put out the light of another Our innocence lost over time Means to an end
| It's hard to hold your head up high But we must try
| Sway together in the dark It's supposed to be 'Cos I want to know the end And you never ever need to fight But you're fighting everyday And I don't know when your light will go out Innocent crying child The heart of your enemy
| My heart to heart with the light And we always get along Counting the stars in the sky Thinking why they have to die Just face to face we can hear A voice telling us it's wrong Counting the stars in the sky It was like a lullaby
| For sunsets I'll break the rules We learnt playing down in the heather The secrets behind all the veils We just use other words For freedom we make our charge For friendship we bare down on others Can't one of you just calculate Tranquility?
}```
Found this on 4chin
Amazing candy
I'd not mind getting stab stab by her
She needs medical attention
i suppose throwing in the lyrics directly with my verser scripts is more imaginative than letting an LLM interpret it tho
this guy is a fake, its he made it all up
chief. imma need that prompt for... science.
I think I heard some drama about that guy before. Idrc tbh
XD
Its uhhhh
SLE, mksks style, best quality, masterpiece, detailed background, highly detailed, intricate, detailed face, long eyelashes, detailed eyes, (cowboy_shot, standing, from behind, looking_back:1.3), (thick thighs:1.0), (medium breasts:1.0), (narrow_waist:1.4), (galaxy background, starry sky, stars, starry:1.3),
(one girl, 1girl:1.1), (solo:1.3), (light blue eyes:1.3), (glowing_eyes, glowing pupils:1.2), (white hair:1.2), (single_hair_bun:1.3), (hairpin:1.2), (tied_hair, hair_up:1.3), (thick_lips:1.4), (pale skin:1.1),
(white dress, ornate clothes:1.2), (long dress, elegant dress:1.1), (white_robe, mage_robe, ornate robe:1.2), (gold:1.1), (blue earrings:1.1), (glowing_earrings:1.2), (white elbow gloves, white thigh boots:1.2), (white garter belt, dress straps:1.0),
(light smile:1.3), (half-closed_eyes:1.3), lora:egirlAesthetic_v10:1
same i dont care, but i do let people know not to take any advice from that guy
if depression and art had a baby...
(KHFB, AuroraNegative:1.1), (low quality, worst quality:1.4), (loli:1.3), young, (animal_ears:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad hands, by (NEG-badartistnegOG:0.6), NEG-bad-image-v2-39000, monochrome, greyscale, (nipples, pussy:1.3), (clothes_lift, lifted_by_self, dress_lift:1.5), (side_slit, open_dress, open_skirt:1.3)
Changed a bit
so. many. loras and embeddings
And fabulous eyeshadow
Theres only 1 LoRA, and only negative embeddings 
you mean just art?
SLE, mksks style & "detailed background" are trigger words in Aurora xD
True 
There is 4 negative embeddings though xD
XD
jesus
laughing my ass off the more i look at it
I'm also using SAG, on 2 scale and 1.75 mask too
egirl aesthetic
lol
Yeee, I don't think it like fully works, but it does work to do what I want, which is make characters look like more mature and have less of a babyface :o
@magic wave removed all the weightings... this is a lot better 🙂
Fair, I like my weighting XD
my model already starts disintegrating at cfg 7... adding a weight above 1.2 usually starts fucking stuff up hard
O, fair then, Aurora doesn't mind my weighting addiction xD
mine is pretty damn sensitive ...
and why is it above 75 tokens, now i can't use cutoff 
did you see cryonicus' prompts?
i have. i think it's half tagged
works like a charm tho
make an image
get wd tagger to re-do it
then prompt that
(legendary powerful construct design of a futuristic war machine:1.26), synthetic mechanical fighter, mecha style, futurecore, (slim design, visible pilot:1.24), quantum technology, (plasma and pulsar weapons:1.19), fighting in the ruins of a (desolate barren cyberpunk mega city:1.23), (sand storm:1.16), ( cyberpunk art Style:1.24), (splash art style:1.21), (hypermaximalism:1.1), (digital art:1.20), (digital illustration:1.13), (cold and dark colors:1.3), hard lighting, volumetric lighting, (masterpiece:1.25), (best quality:1.25), (highres:1.25), (2.5D:1.232), HDR, Ornate, 8K
Negative prompt: (low quality:1.15), no quality, bad quality, crap quality, low resolution, lowres, no detail, low detail, (deformation:1.15), (deformed:1.1), (mutation:1.1), (mutated:1.15), (extra fingers:1.05), (extra toes:1.05), (extra arms:1.05), (extra hands:1.05), (extra digits:1.15), extra limbs, extra eyes, ugly, incomplete, text, title, description, logo, watermark, image artifacts, spaghetti body, conjoined twins, conjoined body parts, big breast, massive breast, major breast, big tits, massive tits, major tits, hentai, eroge, nukige, pornographic, porn, XXX, erotica, skimpy, slutty, nude, nudity, naked, bad_hands, line art, present day, past, bulky mecha, gundam, big mechas, large mechas
1.232 lol
blood red rain and dew fills the air, soft light, dim lighting, street lamp lighting, blood red glow (best quality:1.27), (masterpiece:1.27), (highres:1.20), (digital art:1.20), (grimdark illustration:1.25), (Nu-Goth Art Style:1.28) <Young neko ninja assassin perches on EDO period Lords Manor roof (best quality:1.27), (masterpiece:1.27), (highres:1.20), (digital art:1.20), (disaster art:1.25), (apocalyptic art style:1.28) (best quality:1.27), (masterpiece:1.27), (highres:1.20), (digital art:1.20), (digital illustration:1.20), (anime-art:1.25)
bruh his prompts
I did the like 1.17, 1.18, 1.22, and all that a lot, now I'm lazier, but I vibe with it XD
ok I found it
thx for the prompts. it will be used for further science. 
(Disaster Art Style:1.084), (Digital Art:1.14), insanely detailed and intricate Cybernetic fallen angels of horror, ([alisonbrie:anadearmis:0.60] as a synthetic humanoid alter:1.16), beauty filter, (featherless bone wings covered in blood and a net of fleshy tendons:1.18), (standing in a large canyon where the canyon walls mimics a gothic cathedral:1.108), (Dark Fantasy Art Style:1.144), (Tones of Black in Background:1.096), (Dark Scarlet Red Foreground:1.096), (Pseudorealism:1.072), Hard Lighting, medium key lighting, natural lighting, (dark mode:1.12) (2.5D:1.2), Cinematic Shot, Ultra-Wide Angle, (masterpiece:1.132), (best quality:1.132)
this is dedication

That do be some weighting dedication :o
probably used like randomize strengths at this point, no way he did it manually
Maybe? 
idk
@cyan burrow what is the matter with this weighing? you did it all manually? based on what? trial and error?
interrogator replication
completely disintegrates in landscape mode tough
Fair XD
POV date with
not sure what happened here. but this is a quest location if i ever saw one
I didn't want to wait for more images
oh... i didn't have my neg prompt on, that'w why it all sucked...
"A palace of broken glass"
Is there such a lora where you genderbend a male lora
Example Rengoku kyojuro lora + genderbend
Is there such a thing
no idea. Can't you just do that with prompts?
probably a lora for the character, but not apply it strong enough so you can genderbend it
^
I imagine the AI would get confused if you had male tags in the lora for the character and then also female tags either in the prompt or in a separate lora
my prmopts are all about confusing the AI 😄
I might try doing that with some FGO LoRA I have
Yep
you can literally do that with prompts
Normal|Genderbent
Look like a femboy tho.
add masculine
I removed 1girl and added 1boy
I'll try
my custom lora 0.5 , 0.6 , 0.8
Didn't change much
did you still keep 1boy?
yos
if so -> ditch that shit
it's not 1boy, it's man
i never use 1boy and 1girl -> that's poisoning your own waters
kay
that made it worse xD
you're the one doing sex-reassignment surgery!
It's for science
seems your model doesn't know about masculine 😦 go for muscle?
yeah, 1man seems to be completely useless in my model
it's either 1boy/1girl
i used 0, and didn't use masterpiece either...
ok
reduce weighting on your lora?
it's on 0.5
I mean, I use man too, I like just man or like 1man or something like that more than 1boy
might be that the base model doesn't know the tags really well
I think 1boy tends to make like teens, not men
i KNOW it does
that's why i fckn hate 1boy and 1girl
Got a jojo character instead
still feminine af
PERFECTION
you could try R63 as tag
it MIGHT work
(r63 = rule genderbending)
You have any female character lora?
yeah
Manly man
What the prompt doing? 
did you just fucking prompt misty, as an masculine old geezer? XD
XD
1boy works way better for me. Not really sure why
I tried to do lumberjack xD
oh god lmao
feminine man
watercolor, masterpiece, painting far from below , blue eyed curly blonde haired (mature:1.2)_princess_peach wearing a crown, dark, large_perky natural breasts, atomic , (in a shadowy_mountain_pass:1.0) (during a supernova:0.75), during the golden hour, by Krenz Kushart and Wlop and Posuka Demizu and Yoshitaka Amano , detailed, vivid , Hyper minimalist, Underwater Lighting , dynamic shadows,
This one is better
got this.
the hands are a little fucked tho
Christmas Vergil
<3
i shall fear this prompt for it has breasts in it...
Shouldn't matter unless it's a specific NSFW model :p
heh
OH GOD
... yeah.. about that...
GODDAMIT, it is NSFW but is pretty 😮
literally using a hentai model
DM ME
the fruit banana or the...
The uhhh
oh
should I train a model on this and then publish it on hf with misleading readme?
XD
the funny rod
another mix coming hehe
Its basically a POV BJ scene tho, I don't know how it got that horny, lemme take a close look at these prompts 
dm
Then the prompt ain't at fault xD
That's a really good prompt
Oh god
and... i can't post it... clyde hates it 😛
my model REALLY likes doing the caustics justice though
This prompt is activating Aurora's neurons really hard xD
try removing "large_perky natural breasts,"
this screams nsfw to me
Tis prompt is making like
already did that
100% hentai, like full on, head first, full send
large perky natural breasts = i just want nsfw but don't want to admit
genitals always suck in almost all models
I mean, the model I'm using doesn't default to nude breasts :)
got the golden hour, got the supernova 🙂
Yea, took me 2 months to fix this
I think its because this
Alright now it works, now its not horny anymore xD
maestro, prompt plz 🙂
(best quality:1.2), highly detailed, face details,((eyes details)), yellow eyes, long hair, spiky hair, (black hair), fingerless gloves, chill eyes, good anatomy, gorgeous, buff arms, solo, fit, serious, dark tan skin, minotaur girl, [giant horns], tall, fantasy clothes, (dungeon), ((dark skin), [giant ox horns], detailed eyes, gold hair accessory, highly detailed, (cages), exploring, ((fully clothed)), good fingers, round face, wasteland, fog, ruins, (vines), wide open eyes, flushed, shy, flustered, sleeveless kung fu shirt
removed the bs, applied come cut-off. now this is art ❤️. This is a good prompt ❤️
a battle angel girl, at the center of city, full body, dynamic pose, (storm:1.2), ruins, (bokeh:1.1), lora:BF4Color:0.45, (reflections:1.1), f stop 16, light particles, tracers, light rays, depth of field, Volumetric Lighting, close up, heavy rain, (Glowing Edges:1.1), Edge Detection, Contre-Jour, (Intentional camera movement:1.1), hypernet:sxzCinematic_800StrongEnough:0.65
Negative prompt: (easynegative:1.1), (verybadimagenegative_v1.3:1), (low quality:1.2), (worst quality:1.2)
Steps: 36, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1779036622, Size: 512x768, Model hash: 01a44729ee, Model: P_test1, SAG Guidance Scale: 1.3, SAG Mask Threshold: 1
What BS?
Except, it's not BS
my model does those out of the box -> intricate is something else tho
Can really feel that from below acting up
needed bigger breasts tho, i'm still seeing bits of sun...
it literally does not enhance my images anymore...
You sure about that

prompt only masterpiece, and see what you get.
Too lazy to upscale/inpaint the cursed face
black hair. cyborg body -> it knows battle angel 🙂
It's to capture a certain style of framing but sure I'll do it
I think it's kinda silly to just prompt 1 thing and base it on that though
I was trying to make femboy, but this works 
why do the femboy have breasts tho?
except. it's not. given that each token has global and local impact.
It's exactly doing what I said it was doing
But works for me 
true
wait til u hear males have breasts too
You know what I meant by that
It's making the framing a style that resembles modeling for a painting
A la mona lisa
Hence...not a bs prompt as that's exactly what I want from it :P
and you get bigger breasts.
Think I need that 2n5 LoRA
and generally, sluttified images: because booru 'masterpieces' are certainly not more pristine than their 1400's counterparts
it's good when applied as [masterpiece:0.4] tho, so it's general composition is kept out
can't use the lora's or hypernet. this is what my h-model does 🙂
I mean...all this is just your personal preference...I'm still not seeing why you brought it up?
try working without it for a while
I did
as someone asked me what bs in prompts is. i feel it's a noobtrap, and that's just how i feel about it
detailed, perhaps ok, but only IF you actually want details
intricate, certainly a tag on it's own which is different from image quality
but masterpiece, i still have to find a real application for it except for late adding in a prompt
2n5 LoRA couldn't get rid of the booba 
I find it a bit weird to be make these statements as the impact differs per model
Nor my Toxxy LoRA I think 
These aren't universal truths
nah, it's been there since the OG leak. unless the model has 0 ancestry of the redacted leaks, it WILL respond to it
Hi
Hi
I know it will respond to it...my point is that you are saying it like they are bad things
Which it isn't
I mean, I feel like its pretty much how the [REDACTED] is the best model to train LoRAs on, its because all popular models finds its roots on it
not per se, but has more influence over an image than just "enhancing image quality"
and now if it only did that, i'd be totally ok with it
but it doesn't just do that.
I get that
that's all.
But nobody said you should use it
I just took issue with "bs prompts" :P
because it will taint images when not going for "FEMALES WITH BIG BOOBS"
I'm new here how to use ai to draw photo
(especially when using a LOT of weak tags)
this look good I like
Look more like tomboy tho 
@warped gyro
Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝|tech-support for any problems you encounter while installing it locally!
... well i did at least
Also #1072220168534642768 is good
ugh. the forbidden word was mentioned xD
Voldemort?
nah, the one mentioned in chat earlier
@tropic wraith thanks
Prompted this
"AIGHT, NEW CHALLENGE
just copy paste chat and prompt that!"
got something weird
God this is so weird
@Eface
nah, it's been there since the OG leak. unless the model has 0 ancestry of the redacted leaks, it WILL respond to it
[17:50]Astranora: I know it will respond to it...my point is that you are saying it like they are bad things
[17:50]Astranora: Which it isn't
@Astranora
These aren't universal truths
[17:50]Smile ⭐: I mean, I feel like its pretty much how the [REDACTED] is the best model to train LoRAs on, its because all popular models finds its roots on it
[17:50]Eface: not per se, but has more influence over an image than just "enhancing image quality"
[17:50]Eface: and now if it only did that, i'd be totally ok with it
[17:50]Eface: but it doesn't just do that.
[17:50]Astranora: I get that
[17:50]Eface: that's all.
[17:51]Eface: nothing more
[17:51]Astranora: But nobody said you should use it
[17:51]Eface: that's why i said -> i can't use it, won't use it (bewerkt)
[17:51]Astranora: I just took issue with "bs prompts" :P
[17:51]Eface: because it will taint images when not going for "FEMALES WITH BIG BOOBS"
[17:51]nyi nyi: I'm new here how to use ai to draw photo
[17:51]Eface: (especially when using a LOT of weak tags)```
Oh the™️ [REDACTED]? :o
good shit! ❤️
This doesn't mean anything though?
a prompt impacts a picture
more news at 11
it doesn't mean a single thing. it's just a testament that this model is horny.
but creative too 🙂 i don't think i can get it to make actual small breasts
Which is good, but annoying at times :p
The last 5 gens were small breasts including masterpiece prompt :p
Sadly these were requested and nsfw so can't share em xD
good enough
This prompt is weird
this is a good prompt!

its a new model im working on
Thats a pretty nice style it has :o
Oh wao, thats pretty nice :o
and of course my ocd wont let me rest until the hands are purrfect
Waifu picker 
That will be hard XD
sorry but im still doing generation like this

Addicted to his waifu
well, got to be committed to ONE waifu
Gave this prompt a go with Aurora
I think its trying to say something 

got to do it step by step my guy 
GM 👋
Morning 
Gmorning Nubbacious 
my waifuuuuu 
Kinda cool tho :o
wow, best visual novel ever!
Masterpiece, late masterpiece, no masterpiece
i managed to do it with my other model
Even got a mascot 
You did? 
imma go snap you outta existence boi -> i have no idea what i was expecting xD
interstellar
Oh damn fair then :O
Hi
3d
hi blasphemy
this is not a generation channel
everyone here is generating on their own computers
Was meaning for Makima
cursed
good day to you fine Sir!
cyberspace dropaholic masterpiece -> since the dropaholic song from sonic frontiers doesn't have lyrics
but goddamn, does sonic frontiers have a fckn banger soundtrack
Luck, and also https://civitai.com/models/40199/aurora, its good 
also makima...?
ok I got it
I trained my own lora and with anythingv5 but hands is so bad
I am thinking how to improve it
Just luck and negative embeddings, thats what works best :o
the lips...
alright, I might take more hands negative prompt into it
so
T H I C C C
I usually use
(KHFB, AuroraNegative:1.1), (low quality, worst quality:1.4), (loli:1.3), young, (animal_ears:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad hands, by (NEG-badartistnegOG:0.6), NEG-bad-image-v2-39000, monochrome, greyscale
you made her cry
For negatives
Good 
negative prompt has addition network!?
Only embeddings
KHFB, AuroraNegative, NEG-badartistnegOG, and NEG-bad-image-v2-39000 are embeddings
ok I think it is another issue for me to survey
Is it working for you well?
idk....
or an attack chopper
how about cyborg angel
honestly the coolest skin/armor/flesh ive ever seen
This is like a Warframe but with the head of a girl.
yeah, it's like not a head, just some decoration stick on a war machine
that's creepy af
the builder species of that did it as a joke to dunk on the humans when they're gonna invade
Whats the prompt for that? 



genshin but honkai?

sooo

dont mind if i do...
sounds pog



what is it
Dark sushi
i like the effects & particles
but i like everything else of Aurora
yes yes yes yes yes
lets bump it up a notch
sheeeesh

damn model?
gonna try that prompt, thx
NOOOO my lucy LoRa never actually activated, had the wrong LoRa prompt for it....


actually looks like lucy
finally
even though the ones before werent all hat bad for having no information about lucy
DG3 is the best
Great
oni vtuber


stealthmetadata?
Just managed to run Pyg6B on phone with 8gb RAM total
idk what that means, also i come in here from time to time, and thank you ❤️
workaround to the stripping of metadata through discord
still not sure what you mean... are you wanting the prompt
yes, but metadata is convenient as to not clutter up chats with walls of text
Just take an example from projectai.
Actually if I am not wrong guizmus said they are working on a similar bot
What is that extension?
this
????
WHAT
wait
Yes
its gone
L bozo
rip





Not rip actually
why did it say downloaded for me tho
Now we don't need it
Well we didn't need it from the start?
I don't remember a time where they stripped metadata
they did
can confirm
yes
it was when you were super into LLM and left this place 
Idk I remember downloading an image and putting in PNG info and getting stuff
Am I misremembering.
that's before the removal of metadata by discord


quite solid
POWERRRR
am I only one excited
how long do you have to wait for a response?
long tim

like probably 1-2 mins
maybe even more
actually not too too bad
depends on settings obv
I can run it probably like a ton faster if I use oobabooga
not sure if can on phone
currently just using koboldcpp
still need better gpu for better generation time
They've stripped metadata for a about a week at a time twice now.
Yeah, no clue why they keep changing their mind on whether they should strip it or not.
Hello, I need some help, I would like to recreate a character (this one), I tried many things, img2img, controlnet, or just describing it very good but it always give characters very very different from that, any tips ?
Was wondering if anyone was willing to give me a tutorial.
I want to be able to create a deforum video using premade images.
I've tried Init image, but I don't see a way to use a second image.
I've tried video init but when I paste the path on my computer, and select Use Init. Nothing happens.
https://www.tiktok.com/t/ZTRT5HKaS/
The goal ^
that is loopback wave, and its really badly done by someone who didnt even test what they doing, for that video anyways
https://rentry.org/sd-loopback-wave
and
https://rentry.org/AnimAnon-LoopbackWave
bibibangp
Ok! I experimented with it real quick! I managed to generate 30 frames to create a video. However the end product is generated by the ai. Is there a way I can make it have a photo on frame 1, then end with a photo that I already have on frame 30?
I'm assuming I have to change the seed per the given frame that I want?
this will get you started
Total Frames: 60
Frames Per Wave: 30
Wave Offset: 0.0
Base Denoising Strength: 0.3
Max Additional Denoise: 0.55
Initial Image Number: 0.0
Video Encoding Settings
Save Video: True
Framerate: 20
Quality: 40
Encoding: VP9 (webm)
Create Segmented Video: False
Prompt Details
Initial Prompt:1girl, anime, big eyes, cat girl,fully clothed, kitchen, kitchen counter,(loli:0.4137931034482758), (milf:0.1724137931034483), (9yo:0.5862068965517242), (young girl:0.5862068965517242), blue hair, blue eyes, sfw
Negative Prompt:blurry, low quality, out of focus, black and white, blurry, low quality, out of focus, black and white, human, pink skin, brown skin, white skin, blurry, low quality, out of focus, black and white, nsfw
Frame change prompts:
0::1girl, anime, big eyes, cat girl,fully clothed, kitchen, kitchen counter,(loli:@wave_remaining(1,0)), (milf:@wave_completed(1, -1)), (9yo:@wave_remaining(0, 1)), (young girl:@wave_remaining(0, 1)), blue hair, blue eyes, sfw
30::1girl, anime, big eyes, (cat girl:@wave_remaining(0,1)),fully clothed, kitchen, kitchen counter,(loli:0), (milf:1), (9yo:0), (young girl:0), blue hair, blue eyes, sfw
SEED STATE
100::123::prompt
Set seed to 123 and increment the seed every frame
100::+123::prompt
Set seed to 123 and decrement the seed every frame
100::-123::prompt
And no you can not give a already generated image, its all seed based if you want control, you need to pregenerate like 100 images and pick the seeds
Heard, question, every time I generate an image. It is given it's own seed right?
Hey, did anyone test out vladmadic/automatic already?
current hype yes, I have no desire to switch, who knows how long this dev will keep this up, if its still around 6 months from now, i might take a look at it, right now its all hype based
mm nice wth, 1 star without any reason ?
you care too much 😛 models come and go, there is a better one every day lol
But the man put 1 star and did not write any comment
How do I know what he didn't like?
I don't want to stay on automatic, i really appreciate what he done for us. But he is just too messy for me. Every update breaks literally anything
why dont you ask nicely, why and maybe you get a reply, it is what it is, cant make everyone happy
it's an anonymous evaluation, who do I ask? xd
well depends, you can type a specific seed if you want, or just randomly generate it
oh anon, then yeah it is what it is XD
alittle bit sus

Aurora + 2 of my merges + Pika's new model
sus
Just assume it is someone who is anti ai art and down voting things as a protest.
not impossible considering how many people are against AI art xD.
Still pretty stupid that you can literally just 1 start something anonymously without leaving any review.
Switched waifus?
Oh I agree, if they are not going to take the time to say why is it really an honest review
similar just different hairstyle 
I want sdxl to come out all ready :(
Yoooo what prompts did you use for this?
thanks 🙂 that's
Anime, 2 puffs hairstyle, black hair, fluffy pink baggy sweater, blue jeans, brown eyes, dark skin: 1.5, sfw, 1girl, action pose, holding sunflower
Negative prompt: loli,underage
Steps: 30, Sampler: DPM++ 2M, CFG scale: 7, Seed: 3568356604, Size: 512x640, Model hash: 7e778ce711, Model: xperoEnd1essModel_v1, Denoising strength: 0.65, ENSD: 31337, Hires upscale: 1.5, Hires upscaler: Latent
train a lora on the character. how many images could you get of the characters?
is this the only image of the character?
yes sadly this is the only one I have
it seems not very possible to get what I want lol but no worries
you just wanna replicate the character?
yup
or also composition?
just the character, hairs, outfit,


















