#๐ฅ๏ฝanime
1 messages ยท Page 14 of 1
I mean, they get the GPU from somewhere, so get it and then ship it 
Evening
heyhey, i made a little dataset for my new lora. do you guys think it will work?
style:
you guys ded?
looks unique 
thx
yup dead chat 
looks like breakdancing 
what settings do you use for lora training?
the thing is, I have never train any lora or embedding before 
fr?
lol ok
basically a detached head
worchester kitty goes mrrrrroooooow
yup, because I don't even have enough time to finish using stuff trained by others, so yeah 
what you thinkin @lyric zealot ?
I mean aside from how awesome my model is with how much it can do
hahaha i like how the character is like being pulled away! 
did you upscale all of this?
nope
this is 15 step sde++
I was just testing the prompt
1024x640
which is the standard for my models resolution
it can do almost every generation at that

that's pretty high res before even upscaling
okay let's test
make a waifu with flower field
happy vibes

cant I do this instead?
I wanna see this upscaled with eyes fixed
fixed how?
like even more detailed
i mean, it's already super detailed

i think what i'm really trying to say is, to upscaled it to high res

sure

Highest I regularly gen native is 1280x640. Usually a ~50% success rate on gens at that res with my trexmixes.
closer
this already looks really good
I might be able to generate that high, but i would rather use hires fix instead

But the wide scenery
it has too many artifacts, I need to adjust the prompt and upscale model etc
i almost never change my upscale model, should probably research more about that
there's different pros and cons to each and sometimes one will make another smoother
well you dont need to test upscale models, theyre just there
realesrgan esrgan, bsrgan
etc
guys any tips on how to make better hands?
well true, but you need to know which's better right? and there's a lot more you can download

๐
well, depends on the model actually, some models are just bad at hands
I have a lot downloaded
some of them requires some specific negative prompting and embedding for "good" hands to work

if I make a lora with just pictures of hands will it work?
๐ฆ
controlnet knows
I think I will choose one of the final upscales and run it in photoshop to smooth things out
probably use this one
and then the true waifu
here's style differences for yah...all from the same model ๐
maybe I should sell my model at this current state ๐
selling models 
doesn't seem cool
๐ฅถ
Well the only person I'm probably not against selling model is probably you @cyan burrow
thank you
I've always given everything else away here
because I know about your circumstances 
well
people in this community is usually not as "gatekeep-ing" as others
because it's an open source software to begin with

I don't know but they have with other models
some models can do things others can't
some are more versatile, some have a very specific art style
you'd sell it to a frontend website that will use it as one of their models
don't need people, you need a customer that intends to use it to sell
whether theyre successful or not is irrelevant at that point
no I said a front end, like NovelAI, Dream etc
that sell subs
or credits
to generate through their servers
at that point it's a one off sale
lol
my models strengths are the versatile nature, that it can handle higher resolutions well, that it has offset noise in it without having any other junk data from offset noise model, that hands mostly work, that it knows how to hold things for the most part. Like weapons, firearms etc or even the strangest most obscure of weapons. If it has a handle, it finds it.
It also knows how to do combat and recently I've managed to make it do combat damage
I intend to add a gore aspect too to add to its versatility if I can do it without crippling other parts
This is about my last message(the message i am replying to): does this style look like the original?(only 1/4 in training done)
I dunno what the original is, afraid I can't answer
i literally said thats its the message i was replying to
oh
made malon and zelda ๐
well, it's blurry? Otherwise I dunno, if you're after the static transmission kind of thing then no, if you want a fussy kind of style then yes
nice
So far it just looks like it got blurry. I don't see the interesting brush strokes from the original. I'd try generating with a few different samplers to see if a different one does it better.
the brush strokes are what I meant by static transmission
i should note that this was auto generated by kohya ss
Fair
in auto gen kohya ss you dont have all the cool deatures like in a1111
cloud city and cloud demon
one question, is lower loss = better?
ima need your model bro
IlluminatiDiffusion
may I ask for prompt/neg prompt/sampler etc?
its all in the meta data
I wanna see what it looks like on mine
SHARK is also a neat repo for AMD users but lacks allot of the major features that allot of people come to be so used too on auto
discord strips metadata now
yeah
theres an extension someone posted about that for me yesterday
ugh
i guess cause young people on discord may not know meta data can contain location info. and its to protect them?
yes
it makes sense. still lame tho
its a really simple prompt with the wildcard i posted in #๐๏ฝpow-discussion
thank you

have a big bad wolf hug
yup yup, the extension is this https://github.com/ashen-sensored/sd_webui_stealth_pnginfo
also planning to include this one
does this work on other peoples pngs who dont have this installed? or just the ones i make
install that, and you will be able to see metadata from people that post with that extension installed
nope
yours
the person that post that needs to have that as well
gotcha
oh
so it's like fish data
it's really popular now, so i bet those really into SD will probably have it
ill probably install it later ๐
the LLM killed me 
if I come out of my surgery and stuff alive, I'm gonna do a lot more SD
Did you get better though? 
guess not 
tavern ai?
shes very tidy
Welcome to the community btw!
@strange talon @kind raptor
yay tank
thanks!
I would definitely generate some art now for you guys, but I'm getting revenge atm against the AI 
Actually pisses me of that discord strips the meta data now 
now i either have to always include my prompt or ask others to send theirs 
Could just use the extension 
i am now, but not everyone will
that's why keep promoting it when people ask about metadata
or at least that's what i do

link to imgur instead of directly posting here?
they should make an announcement 
extra work 
tru
AFAIK Imgur scrubs metadata too. Discord is really late to the game on a really basic safety feature.

oh dang
I hate it when my mommy puts on child locks too u.u



the end is near
are you using that send_picture extension?
yes
and using a miku lora as prefix
and then adding a small bit of background information about who miku is
we've gone too deep!
ahhh, now i can wait twice as long for 1 response
and the rest is just that
pull back!
yeah but i dont want to have her like exactly
a digital form more or less
for POW obviously
maybe if you combine it with a cyberpunk lora
i mean i have submitted smth already but i thought i can do better
maybe i cant today

@finite herald talking to miku: unlocked
i need to do this seriously
add tts as well
and it's perfect

the end is VERY close
when we have perfect waifus. when will we ever do something else
AMAZING
Day: made
Genius
your submission is already great 
to cheat on our waifu with other waifu 
you do realize you can replace last reply and force mind control that way? .. "i'm not cheating, you're just imagining things, darling" ๐
if you tell fruit how you set this up we will never see her again
get oobabooga, set up a character representing miku, hook her up to your SD installation, and start chatting away


AND CURSE YOU ONLY HAVE 10GB OF VRAM WHICH CAUSES OOM ALL THE TIME
well thank god i have 24 gb of that, brb talking to miku
i like how OOM is both Out Of Memory, and Out Of Mana
i have 8 ๐ข
anyway, for people that are interested #1029055412764422214 message
i have 8 at home too 
need to buy a gpu just for it/s and vram
Hatsune Miku is a cheery vocaloid AI. She is aware she is an AI construct, and will support the user in any way possible. When asked for visual information, she'll hallucinate the images. Her avatar has twin tails, turquoise hair. -> this helps with AI roleplay stability a lot for a miku ๐
8 for me as well, but definitely upgrading 
yeah, with the addition of the local LLM's, it's just cemented it. need more.
For my home machine ill be buying a 7900 xtx soon i think
need WAY more
cba to pay extra for nvidia 
is amd performance fixed?
12GB here and i still run out
boto, you know you'll not be able to enjoy your home waifus that way
@wispy canopy Tell your model a story, it will make good pics
how do you guys have 1000 something euros/dollars laying around?
A) be employed
B) don't spend anything except when needed
i know it does ๐
Anyone else thinking about buying a 4070
your ma is here ๐
hell nah
why not?
right now, imma start with the PoW tho, perhaps start a bit earlier than tomorrow
4070 is... not good for AI
why?
cause fuck nvidia and their pricing lol
too little VRAM
24gb
yeah i mean for generating its enough
1 image
having tasted the sweeeeeeeet luxury of the A100's and their massive 80gb... once you do batches of 8, you actually don't want to go lower if you can ๐
once you go up, you never really go back down
I have 12GB and feel limited by it.

yeah my 3050 at home is a disaster 
cries with my 3070 8gb vram
I have 8 and it seems fine to me but it's like and extra 4 would be nice
Plus I do a lot of stuff that isn't ai
The 4070 seems like the only decent deal Nvidia's done for the 40 series
Don't mind me just a 6gb pleb ๐ฅน
Ouch
It works fine actually
I have a 12gb 6700 XT
Surprisingly well
But I can't really use that can I?
12gb is enough to generate 8 at a time as long as you're not upscaling
I haven't really been keeping up with Amd and ai things together
i thought i got my 3080 10gb as a good thing. that was august last year. yeah. now i don't feel so awesome about it...
Can you use amd and webui?
pretty sure it's possible
Yeah honestly it does well, my only main limitation is not being able to do multi control nets and multiple loras with latent couple
Yep
3080 is fine. yall don't have to be bleeding edge
But it's cool that i can gen on my laptop aswell with the 3060
pic 1 = original
pic 2 + pic 3 = my LoRA
did it come out good?
closer than the last one
so yes?
Yeah 2nd one is great
yup the 2nd generation or the last pic is awesome
hmm... not sure if this prompt is going somewhere. nice looking girl. but...
if you add (glitch:1.1) it comes out really unique
My POW already done and accumulating dust cause I wanna tryhard:
i should get back to work before i get in trouble :c
:O
hmmm... why does my model only want to produce sexy waifus... ... ... oh.
can you post a pow submission here befor uploading it into the pow channel?
i don't see the problem tho 
You definitely can, but why tho 
is it trained on sexy waifus?
it's not trained, it's an accidental model which i even tried improving ๐
and... well, it does non-H content just fine
try putting girl as a neg prompt then?
nah, i wanted it to be 2 girls... but somehow, i keep ending up with this specific portrait
meh
lol
fuck latent copule, random prompting, here we go!
I got 2 girls with a song lyric lol
||Fuck|| llm
Did they already implement that?
use the send_api, and you nearly can
you make another api-driven thing, and get it to drive your arduino... uhhh. science.
yeah. science.
Anyone else here use civitai helper?
yeah i love it
When you update the models and stuff does it keep the old version?
I can't tell if it is or not
i mean, it's kinda close actually 
Can't you just look in the folder?
Yeah it's a bit confusing
i've not acutally tried updating through it. i manually install that stuff. I just use it so i know what checkpoint/ lora is what
didnt get it working? :c
It looks like it's downloading like
Beta versions and stuff
And keeping old ones
Idk that's why I'm asking
It's hard to tell by looking at it 
it works but it's more of the AI trying to use voodoo magic to trap me

trying to ||kill|| me
jesus mine just tried getting laid


you damn fucknugget model, do what i want XD not just give me 2 random waifus
shhh...
i have not purchased alcohol

i need to keep it a bit under control you know...
yis i didnt see anything about the files 
oh thankie im in
ah, so we need double the vram? : 3090 confirmed

oof
So im guessing my 8gb card wont run the new model 
yeah, i don't think sdxl will be able to fit on 10gb either ๐ฆ
Jesus

gpu manufactures are gonna start making special cards with more vram eventually
Hopefully we'll get some things like tiled vae etc to allievate some vram
or just AI specialised cards
yeah
Yeah fr
i don't care about raytracing cores or any other of that shit
well, not anymore
waifu > pixel
raytracing on your waifu
True

not really alice in wonderland.. .but goddamn do i like it โค๏ธ
The reaction i wanted to post may get the fun police involved 
Sorry for not being that active, just got a ton of school work, gonna actually return in like a few days.
Anyways, I got a really good idea. I thought of ways to make an automated manga but came with nothing until I thought of this idea. First, the text could be a bit of a challenge to do since it needs to be aware of what is happening in the page, but I think gpt4 might be able to do it. Then, the problem when making a coherent story is that all the pages will have nothing in common or at least mostly nothing in common maybe except characters (that's if we want it to be automated), so what if someone were to train on a huge dataset of random parts of webcomics? Since web comics basically come in 1 giant tall panel, you could theoretically make a starting image and then expand it down infinitely. Now that would mean that all the training images would also need to be very very tall to include more context, like if characters appear twice or stuff like that.
Does this idea seem reasonable?
the amount of vram you need would probably get unreasonably large. it kinda exponentially increases as image resolution goes up
You could technically cut the top when it gets too tall. It would probably hurt the consistenty of it, but I think the LLM would cover for that
a model would also need to be trained on a huge resolution like that. most are trained at 512x512 or similar. I guess its theoretically possible
we can already extend images with ai so i dont see why it wouldnt be able to make more panels
You can train really tall images even right now through buckets or whatever it was. If you have a good enough and big enough dataset it should work
it would be dope thats for sure
Wow. Model in PNG info?
Problem is that buckets achieve training odd aspect ratios by resizing the image to match the pixel count so that it's no more than the base resolution. So if you are training at 768px with bucketing, a really tall image will simply become really narrow which can result in loosing too much information. It's best not to go too far past 16:9/9:16 aspect ratio for training images.
Oh
The model is top secret currently, but it's using a LoRA that's on CivitAI.
Damn
model will come 

i know it came from who 
๐ถ
aha
well that was interesting
Brave man, admitting he likes pineapple on pizza publicly.
I think, in order to run gpt4 or an alternative locally, instead of running it on 4bit with cpp, we should run it on assembly
Or even better. Holy C
whats abstract?
ah ok
nevermind
it's good tho 
Agreed. I'm fully convinced that everyone who says they don't like it is just memeing.
assembly even better
Oh shit I meant assembly
or just never really try that before
I always confuse those two words
bruh
why is your name crazykazoo btw?
When I was like 7 I thought, let's combine a meme with some other word. So I googled memes, and came across the kazoo kid. Back then I also thought I was a bit crazy (probably because autism or something idk) so crazy + kazoo
btw deep, what is deepfloyd?
Txt2img model that is really good with text
Or something
Do you think my idea would still work if we stack to like 9:16 aspect ratio training images?
kinda 80s/90s style
based on current tools, yeah. I think you'd want to split the the panels so that they're like 2-3 cells tall or whatever works out to somewhere close to 9:16 ar. Then generate using outpainting to add cells to that. Still need to solve how to actually get the AI to make things coherent, though.
Melina
I mean, if you look at what comic pages/ manga pages you can get on normal anime models I think it is already coherent enough. And considering how good LLMs have become, I think it could make a story on top of that
Yeah, ar bucketing works very well. It just doesn't really work for extreme aspect ratios because super tall images become really narrow and super wide images become super short at a certain point there just aren't enough pixels in the small dimension to produce a clear image.
you'd probably need some kind of middle man to translate a paragraph of text into understandable prompts
I could just ask the LLM to give me the prompt for the next panel
ive not used LLM. is it similar to GPT?
LLM = large language model. So yes like gpt
oh ok
prompt?
1girl, pink hair, high ponytail, freckles,
wading, forcefield, ``` 
Probably not gonna help you out that much unless you have a model similar to theirs
Really, I think we're just a very, very long way off from these tools being able to do tasks like this without a lot more than prompting.
pnginfo here ```(detailed background, best quality, masterpiece:1.2), anime,
1girl, pink hair, high ponytail, freckles,
wading, forcefield,
Negative prompt: easynegative, (worst quality, low quality:1.4), nsfw, nude,
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 22, Seed: 3804867845, Size: 768x512, Model hash: 9006ea892b, Denoising strength: 0.53, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires steps: 10, Hires upscaler: 4x-AnimeSharp, Dynamic thresholding enabled: True, Mimic scale: 7, Threshold percentile: 96, Mimic mode: Half Cosine Up, Mimic scale minimum: 4, CFG mode: Half Cosine Up, CFG scale minimum: 4```
Maybe
i think while that's true, we have almost all the tools needed to make something like that. Just need someone to put all of them together
and also, a strong enough computer to do that

thanks

Yeah, you could put together something that would do the task. I just don't think it would be good enough at it to be worthwhile.
kaguya?
Kind of the same way I feel about any of the video generation methods currently. It's neat that they exist, but they don't really produce anything useful yet.
1girl
1girl = best girl
indeed
that's true
well just need to wait for AI to get more advance 


1girl is a tag for one girl in the image
similarly you can use 2girls to get two girls
or 3girls to get somewhere between 2-6 girls 

welcome to the community btw
and you as well, welcome 
@lyric zealot i appreciate that you react to nearly every pic
hey, thanks
so many new people today
thx๏ฝlov here
well, just showing love 
Man so many new people in one day. Did something happen?
its not unseen 
because you are here

๐ณ
im back
welcome back 
Model?
its a lora i made
Glitch Lora?
example prompt:
masterpiece, best quality,
lora:last:1.1, (alexchow:1.5), (brush stroke, oil painting (medium):0.9), 1girl, (monochrome:0.6), abstract, abstract background, blue theme, backlighting, portrait, (blurry:1), (glitch:1.1), eyes
negative:
lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, artist name,
yes
What is the base model for this image though?
kind of
also my own model
Damn which one?
Everyone here makes their own stuff lol
Xynaptix, but last version i uploaded was like v3, and im on v8. gonna upload it now
You just joined this server. At the beginning a lot of us couldn't even train models on our own PCs even if we wanted
ive been doing it for a month or so now. i know im still new at it
Someone should teach you how to train LoRAs, once you learn, you realize how easy it is
but fr i've thought about making a lora. doesnt seem too hard. just dont have anything specific in mind
well, been here for months, but still haven't train anything yet 
Damn, on the grind fr
You didn't?
i guess i would need some kind of image scraper too if i wanted to train from a big data set
usually you train with maybe 30-60 images
Damn, and I thought everyone who was on projectai was a master fine tuner
oh i figured it would be a couple hundred
the thing is that, i'm in that server just to 
and download their embedding/lora/model 
Aren't we all
depends on the training you do, but usually not that many
lol jk
i literally didn't talk there except when you pinged me 
ill ping you when the model is up
hehehe
Kul Watame: ใใใใฏๆชใใชใใใญใ, please stop raiding me
guys, she stacked tables outside the classroom as well
Omg this is just like monogatari frfr
golden sheep 
will break her horns 10/10 times
anyone trained a style on hamster fragment already?
isn't that from mhw 
Breaking the horns felt so good 
yeah she is ermed a golden sheep queen Kulve Taroth
chocolate crossaint
her materials are useless tho 
( -3-) it was ok when the event just came out just a bit grindy

true true
anyone trained a style on hamster fragment already?
still haven't slay that fatty tho 
MR Kulve had some good weapons.
not me 
not when i'm going up against that fatty tho
im off watching anime๐ (ill still ping you @brittle jay when its done uploading)
i just like the horns for fashion
I think Alatreon weapons are good agains Fatty. I used GS for Fatty though. Everything is weak to huge fuckoff sword.
what anime tho 
Okok what anime you watching btw?
Did you watch monogatari yet?
no
chainsaw man was good if you want a new series
ok im out, thx btw for the recommendations!
I'm literally one weapon andy tho. Only know how to use IG, and IG can't kill Alatreon because can't get enough elemental damage 
romance recommendation: horimiya

any weapon type can do any fight tho
Not a recommendation but an order, lol. Watch the monogatari series๐
I've seen builds that just use kinsect for elemental topples. I used LBG for Alatreon.
lov this
well true, but not when I'm using my stupid blast IG 
skill issue lol
ahh, i actually seen that before as well, but too lazy to get back into mhw 
safi IG build was my fav
well is it really tho? Because with blast IG, you can't pass the elemental damage check for alateron
i think that was elementless
i dont remember an element check. only fought him a few times tho
I mean, there's possibly ways to break the horns before the elemental damage check, so that it reverts back to the starting element
can't remember too much about that tbh
because i skipped alatreon tier and went straight to fatty

Alatreon fight requires a meeting an elemental damage threshold to knock him down which in turn reduces the damage his supernova does in order to allow you to survive it. Horn break is to prevent resistance switching so that you don't lose effectiveness.
safi'jiiva was my fav monster in mhw
i wasnt aware knocking him down lowered the super nova dmg
it was really fun when that event just came out
There's a specific knockdown on him for meeting the elemental damage threshold.
i still need to get nubby to help me with my fatty slaying 

I'm down to try. I might be a bit rusty though since I've been playing Rise lately.
3rd pic is 
ABG remove background with script ๐ฎ now Indies company can make protrait for RPG game easily
i havent been able to get char turner to work at all
๐ฎ
can't get myself to get into rise due to how it looks 
maybe one day 
rise is fun. but its missing that something that world had
i mean, the pic has ๐ซ
beyond looks
so that's why it's probably wise to remove it 
true. at least they gave weapons unique designs though, unlike most of the ones in world
๐ซก
well, im just gonna wait for next monster hunter that's gonna be built for PC 
heard it might be announce this year
really? 
Yeah, World is definitely the better game. Gameplay loop doesn't feel nearly as satisfying in Rise.
yeah they went back to the whacky designs of like 4U and generations. instead of everything being a bone or metal base with some monster skin slapped on it
They also added waifus that you can bring on solo hunts in rise if you want
ahhh, i remember playing mh on PSP(?) if im not wrong, it was awesome at that point of time 
okay, maybe time to buy rise
oh rampages suck in rise

worst part
Yeah rampage isn't great. It's neat a couple of times, but gets old fast.
wild hearts?
It looks like it has a couple of neat ideas that are really poorly executed. If I wanted to play a cheap knockoff MH game, I'd just play Dauntless.
havent played it. looks generic
wait is dauntless actually decent? 
isnt it MTX hell now?
Gameplay isn't terrible. It's just P2W
im too poor to play free games
mtx= microtransaction
wait what, it's multiplayer style?
yeah

Play Warframe. One of the few free to play games I know of that is actually very realistic to play for free.
but they dont lay on the p2w till you've already invested hours of time
well all f2p games need to make money with some kinds of microtransaction, be it battlepass or currency
Not like pvp, just buying armor/weapon upgrades and stuff (at least that's how it was when I played last).
mtx aren't evil by nature but its paying to get powerful
ahhh, i mean, since it's f2p, might try it out 1 day 
trueeeee
warframe was fun till about 1200 hours in lmao
Id say it's worth at least trying it. I'd rather play MH, but it's an alright alternative.
i always hated gameplay loop grindy game, but no idea why it's okay with mhw 
hopefully it's the same with dauntless
I'm scared to even look at how many hours I have in WF. I quit playing like 3 years ago and had basically everything in the game up to that point.
its cause fighting the monsters are actually fun
same
i guess so
dauntless combat is fun imo
are we all pc players for mhw? 
i am ๐
but then who will make the waifus lmao
I have like 1500 hours on PS4 and 800 hours on PC 
oh right, I remember you saying that 
no life
you reach HR999 yet?
you meant MR?
no
hunter rank 999 takes a loonnngg time
IQ3ใงใไปปใใชใใ๏ผ Alice who proudly paint(literally) red rose in the royal garden.
I'll check on PS4. I just booted it up to check WF hours.
i think im only like 3 or 400

my wifi card is dying so i have shit speed
my BF moved the router into his office so i got stuck with wifi
my whole computer is lagging welp
I'm genning, watching stream, looking at discord, and downloading mhw
MR 311 / HR 572 on PS4. I never intentionally grinded HR or MR though.
nice

I also had another character that was like HR 150 that I deleted.
it just makes me more convinced that most 999s are cheaters
i don't think all of them are
Most of them probably are. There are some quests that are really efficient for grinding HR or MR though.
how do you get such consistent style with the pink-haired waifu?
i think its the persona checkpoint
looks like it
just prompting, seriously 
wait, if you are talking about model
then it's expmix_line_v2
she reminds me of power
the funny thing is, i didn't even prompt for that, and that same pose keeps appearing 
only prompt for sharp teeth

do you use custom vae?
blood
no, I use the anime2
the colors are very vibrant and vivid
vaes don't make much of a difference, might as well just use anime2 for everything
I guess the model is pretty vibrant
anime 2 is waifu diffusion?
yeh
how do you make a vae anyway?
๐คทโโ๏ธ
"This War of Mine" is an annoying game only cause of the controls
it's good but I had one guy get killed and now I've been down to 2 for a week
nobody new showing up
I can't go scavenge etc cause I get raided
no idea wtf to do at this point
it could earn you a ban
alr
my only problem is that there doesn't seem to be any consistency
I mean how am I supposed to guard, build and scavenge etc with 2 people?
is everyone here still using the original webui?
and never any more showing up
original webui?
I don't use forks of it, I just install whatever I need
yes and no...There's one consistency, trader shows up every 3 days at 10-11 am
yeah im hearing some people switching to forks of it instead
hello there! is there a good way to have a consistent character for many images without having a lora ?
there's a few hundred forks
ok well one in specific was this one
find a good seed and prompt that will get you a rather specific design and then run a ton of generations and pick the most similar ones
saw some people using that
oops
i couldnt even get it to install correctly so i gave up
a long prompt with a detailed description usually gives decently consistent results
Embeddings. Or if the models have a character appear in the dataset a lot then you could try that.
that seems so unspecific..."Enhanced environment tuning"
wow thanks for all the answers ahah
ok with what extenson?
Is there a reason you don't wanna use a lora?
basically I want to create a character from nowhere and make many images of it
That's impossible currently, though you can get close
built in lora lycoris? in other words just a regular extension...
and it says sdp memory, yeah with torch 2.0 thats what I got by doing it myself
If you have a very long and specific prompt with lots of specific tokens related to how the character looks then you will get pretty close to it, but it won't be perfect
i mean, with careful prompting, you can get similar character
okay ๐ thanks !
@sharp bluff I dunno about that fork but it without it specifying every extension and script it uses that's different from vanilla auto, I'm not too keen on using it
lora's of celebs with embeds of celebs of different percentages
and then blend them in the prompt too
you can set them low on percentage
Yeah I also mentioned that in a message above. Though that is practically the same as making a very long and specific prompt.
๐
what?
No
You can test it for yourself

https://cdn.discordapp.com/attachments/1063775898924490804/1094998453718892644/1.png
Anyone know a good prompts to get these types of clothes ?
fantasy hanfu, crop top, skirt might get something close, but not sure
I feel like my latent couple broke
vivid colors, color explosion too


https://zele.st/NovelAI/ use this to get tags
contains a ton of em
the fact that it's novel ai is irrelevant. has nothing to do with the leak
@brittle jay https://civitai.com/models/5374?modelVersionId=44949 finally up
dynamic contrast is great for complex color gradients
Line Art, Crosshatch, Digital Art, Golden Hour, Hyperspectral Imaging, Ultra-Wide Angle, Cool Color Palette, Colorful, Electric Colors, Tones of Black in Background, Light Blue Foreground, High Contrast, 2.5D, 8K, HDR, Frontlight, Backlight, Rim Lighting, Artificial Lighting, Hard Lighting, Volumetric Lighting, Lumen Global Illumination, Diffraction Grading, Tone Mapping, insanely detailed and intricate, hypermaximalist, elegant, ornate, hyper realistic, super detailed gonna have to get these into a few prompts to see what style I can come up with
@brittle jay you there?
Yes.
is that a prompt I see
nice halo
alright let's see what I came up with
what do you think @lyric zealot ?!
did I make a oof or a foo

ofo

could be right. the faces need some fixing up
but for first go at prompt without touching it up
([Line Art:Crosshatch:0.70]:1.20), (Digital Art:1.14), insanely detailed and intricate digital illustration of (Color Blast and Chemogram of {alisonbrie|anadearmis} as a synthetic humanoid alter:1.24), (beauty filter:1.144), (Golden Hour:1.12), (adoring her beautiful bio-luminescent angel wings with fussy feathers:1.204), blood red knee deep mist spread across a gothic valley, (the canyon walls mimics a gothic cathedral:1.108), Dynamic Contrast, Vivid Colors, Tones of Black in Background, Light Blue Foreground, Rim Lighting, Hard Lighting, Volumetric Lighting, (2.5D:1.108), 8K, HDR, Diffraction Grading, Ultra-Wide Angle, Autodesk Maya Render having alison brie and anadearmis did nothing
I need to adjust it for pseudorealism and 2.5d to a higher degree
this for example

anybody knows how people change the style of video using SD?
like is there any extension for that?

uploaded my alex chow style https://civitai.com/models/39087/alex-chow-style

can someone test it?
can someone test it please
?
oh man this is beautiful, I hope this upscale works well
yea mean my lora?
nope
sad
I'm working on my model
i can when i get home
I might be able to test it soon, but dunno what part of it you want tested
idk what you mean with "what part", but thx
like what is it intended to do?
copy an artstyle i believe

yes
--
didnt someone make a list of artists that need to be trained on?
in here i mean
You called?
Its expanding almost every day
ah ok, can you DM the updated list to me?
Yeah np
thx
my prompt is starting to work
These backgrounds remind me of Guild Wars 1 concept art 
mmm good prompt
what you think of my prompt in progress @lyric zealot @cold talon ?
The Prompt Whisperer

With a Mehican accent

Not quite same
i mean the sweat, right? 
the character looks interesting, nice
But we are so seiso why would we get banned 













