My new Gen Anime dance
https://youtube.com/shorts/UY2e54XFHoU?feature=share
#🍥|anime
1 messages · Page 7 of 1
sounds like a pain in the ass
it can consume 3-4 gb of vram so it's very worth
no lora needed
oh what wow
that model knows megumin
Models can know a surprising amount of characters :o

i'd rather like if they knew none at this point. use the model space for other concepts and stuff, the characters can be added with loras
Cool armor ma'am
this tbh
I mean, I don't feel like it bothers :o
That's not really how models work. Having characters doesn't take space away from other concepts in any meaningful way.
Omg turning off hardware acceleration is insane
it does bother me tho -> "memory" that is now used for characxters could be used for better concept understanding, like the ortographic view
and stuff like that...
😮
Still getting out of memory errors 😭
Well, according to nubby its not that meaningful :o
oooof
Well he doesn't rely on random bullshit quite as much as i do xD
I mean, you can still do random stuff :o
Well, then maybe you be addicted to the random XD
i do fear a bit tho, at this point it's just a matter of time before dead actors can be virtually resurrected...
We have 859.52 million parameters in an SD 1.x models. The number of anime characters that exist with enough images to be trained in a model is meaningless next to that.
I need Quantum computing Waifu Generation. i Need all unsigned long integers noise patterns calculated at the same time!
i shall judge each of the couple of trillion images myself. then move on to my next prompt.
i will not survive, but i'm willing to die on this hill.
Here, have this angry waifu that you can use as an easel to draw your own images instead.
at one point i did 😮
i'm not a good artist
:-)
how do you render in such detail?
There's Loki but what about high key frfr bussin
how many sampling steps ?
THIS ???????
bro i tried 150 steps lol and it was not smth this good
Modern samplers don't need many steps. Try any of the DPM++ samplers and somewhere between 15-30 steps is all you'll ever need.
i am still figuring it out also what makes it so high detailed
am using euler a
can i download it ?
more steps doesn't necessary means better quality
is it mostly prompts that affect the details, like intricate?
Nope
DPM++ 2S a will be most similar to that. Each step will be slower, but it won't need more than ~20 steps.
Prompts don't really affect the details tbh
my prompt game is ded ngl
i think the most detail comes from DPM adaptive
but it takes extremly long, compared to others
is there someone that uses abyssorange3 ? pls dm me
Have you tried switching the sampler that Unipc uses for img2img from DDIM to DPM++ 2M Karras?
thanks, impressive work!
I at least managed to connect the LLM to the game, but it takes about 30 seconds to get a response. how do I solve this? run on colab maybe?
interestingly pretty close, I guess this model knows the character
i am always impressed when the eyes are looking good
17k images on danbooru according to the autocomplete thing. She should be well known by most models.
so if you add (( it means more importance ?
warhammer inspired stuff is lit
crazy good
ai decided it's comic time 😮
im still getting messed up eyes lmao
do you all perhaps have any suggestion for that?
defmix
but i have a vae
did you download the negative embeddings in the negative prompt
2v
what do you mean by that? I do have neg prompts,
Original Hugging Face Repository Simply uploaded by me, all credit goes to https://huggingface.co/gsdf . Counterfeit-V2.5 (which has 2.0 as well) o...
ohh i dont have that
i thought easynegative is just a word to use in neg, didnt expect you need to download it
thank you!!
may i ask what else text embeddings would you suggest that are necessary?
you don't need to use other negative embeddings with easynegative, but using others in its stead will help yield a different style
also you don't need embeddings at all, try something like (low quality, worst quality:1.4) for example
i see, so thats how it works
very recently I've been preferring bad-hands-5, bad-image-v2-39000 over easynegative
(glossy, realistic:1.4), (3d, realistic:1.3), (low quality, worst quality:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad hands, by (bad-artist:0.6), bad-image-v2-39000
do we have an example of the training data of bad-image-v2-3900?
do we have one for easynegative?
cause I can understand the bad-hands-5 embedding intuitively
good Q
it'd be nice to know what examples of images we're avoiding lol
the ketchup mobile
she's growing, isn't she? 😮
Sexy santa
🙂
honestly i am struggling
hairy santa 🤣
Random waifu-less colorfulness
oh interesting, I always just gen'd as she was in the base models
Dj sona 😋
True XD
@supple raptor
Very nice!
indeed, works beautifully
Mine are... uhh..... not going in this server atm
how do you get this type of colors?
same thing, this pastel-ish colors are nice
my colors are always very saturated
gotta work on your negative prompt then
it might be my vae
I'm using anime2 which is the most saturated vae
same
what prompt did you use?
Negative prompt: paint_splash, (3d, realistic:1.3), (low quality, worst quality:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad hands, by (bad-artist:0.6), bad-image-v2-39000
Steps: 30, Sampler: UniPC, CFG scale: 10, Seed: 2978948191, Size: 512x640, Model hash: f16fca4cb4, Model: Anihex, Denoising strength: 0.55, Clip skip: 2, Hires upscale: 2.8, Hires steps: 12, Hires upscaler: lollypop```
what is lollypop hires upscaler
Whoa you mind sharing it (the prompt)?👀
apparently eye fix means it's for fixing eyes only, not that it makes better eyes
help
you killed her
masterpiece, best quality, highly detailed, saturated, colorful, chromatic aberration, acid, psychodelic, holographic, aesthetic, ceaceless, neon, rainbow, multicolored, disco, badass, electric, energetic, ambience synesthesia, glowing, hypnotic, iridescent, sparkling, data bending, cosmic, energy, iridescent, multicolor, magic, synesthesia, double exposure
Old prompt I used when I used AnyBerryMix
Still works great for hyper colorful stuff tho
Thank you <3

love this. i have a prompt that is very similar

badhandv4 > bad-hands-5
using both at the same time is not better than just using badhandv4
fuck clyde, dm if you want the comparison
bad-image-v2-39000 > easynegative
using both at the same time does help general quality
How are you getting the damaged skin and visible implants like that? I have been able to get them everywhere other than the face using inpainting, but anything on the face just really doesn't want to behave.
hair looks too good
this tbh
True 
someone do something with this PLEASE
automail mfs be like:
settings?
img2img from my beautiful drawing
Negative prompt: paint splash, (3d, realistic:1.3), (low quality, worst quality:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad hands, by (bad-artist:0.6), bad-image-v2-39000
Steps: 20, Sampler: Euler a, CFG scale: 9, Seed: 1946422283, Size: 1024x1024, Model hash: f16fca4cb4, Model: Anihex, Denoising strength: 0.9, Clip skip: 2```
wait, bad images like those work in img2img?
if you set denoise high enough it ignores most of it, in this case it'll mostly use the white background etc
also
what do you mean bad


We have Super Reactions 
i cant
get my sketch to work
how do you want it to work?
i want the character i drew
well I didn't really get what drew either, it's not how it works
I think the controlnet has some sketch model which replicates what you draw/sketch better, but I haven't used it
lmfao
i wish i could make simple illustrations too
some unprompted images + upscaled with negatives.
fear me
uahhh
you can just copy my settings and you too can become me
you shall become thou
Such is the way of things.
speaking of which what model do you use
currently anihex
https://huggingface.co/SirVeggie/hexa_models/tree/main
anihex is good at flat art
i see
not anime but i hit generate before i finished writing the prompt and got this 😂
lmao
i was awww so cute then saw that HOLE
what about hexa? more 2d?
gawd dayum
hexaw is less flat, but still pretty anime
my other good model is salutemix, which is more towards 3d
im goin try hexa then
here's a quick comparison
prompt: a girl sitting on a bench with some negatives
what about hands? anihex looks great
I don't really know if there's a noticeable difference in hand quality between them
WAIT A FICKING MINUTE
how do i do hair with two colors
VEGGIES YOU MADE THESE???
that's me indeed
to be fair they are mixed models, so only a part was trained by me, and I'm relying on other models to raise the quality

you dont do them accurately without lora
otherwise, multicolor hair is a good tag
besides multicolor, you can put the colors you want
the arrangement is out of reach without a lora though
veggie can you train a lora for me?
I don't really do requests, training can be time intensive
preparing a good dataset and testing different parameters can take hours
alright np
civitai:

is there a way to use controlnet in 6gb vram
i like it ngl
eye is messed but the hands and sword are perfect
the models are 5gb each, i instantly get cuda out of memory
Wot? You using the unpruned ones?
i did not see any pruned version when i downloaded
Whichever one you like the most
which do you recommend?
Not sure, I use one that is a app, but I lost the link for it
where can i get more hand depth maps for controlnet?
nvm, even the pruned models kill me
we can be quite blue at times
idk why it came out so blue
i didn't mention colors in my prompt
for me it's ~160s to load models and ~20s to load VAE
.
bruh
you running on cpu or something?
LOL
Im new at this World pls explain me how
damn
I heard some stuff didn't work well on 1660/1660ti like CUDA
are you sure you have CUDA enabled
you have a graphics card?
i still get cuda out of memory and cudnn something errors
Yes i have a GTX nvida3050
So beatiful
wait
local or cloud version is better i would really like to install and work with it
since you have a 3050 you'd be better off running locally i assume
Stable Diffusion AUTOMATIC1111 got updated multiple times these last few months since my first installation video, and it has received a lot of new features. So in this tutorial I will show you how you can install the most complete and updated version of the stable diffusion text-to-image Ai + GUI on your PC for free. You need to have at least 4...
Yesok but it's really cumbersome to install it and add the various models apart
Thanks
from what I understand you must also install some models which type you find on various sites right?
mostly civitai yes
it's complicated?
not really
model loading speed is almost entirely reliant on your storage speed. I have SD installed on a HDD and model loading takes about the same amount of time as you with my 3080 12GB
as nubby stated, model loading is entirely based on the drive it is stored on + cpu performance. With a 13700k and a gen4 ssd, my models load in under 5 seconds every time. I merge them in ~5 to 10 seconds.
GPU just makes image speed go brr
bragging 

I think model loading relies on gpu and not cpu
no?
nope
cause it allocates VRAM?
not even a little bit :V
background is very detailed ?
models are loaded into ram
as many models you are loading, you need th have as much Vram and RAM to load them
:VVVVVVV
HDD > RAM > VRAM
i have never felt that my rtx 3060 is weak until now
Good morning
hyper-detailed + increased weight
but the most impactful thing is the lora I'm using
egocentrism if she was a cute anime girl
man it's 5:27AM but I was just grinding making the visual novel I wanted
should I go to sleep?
no, I will just solve this one error
ok I think I solved it
but now there is another one
bruh I am trying to:
start = x.split("Option")[0].strip()
option_1 = x.split("Option 1: ")[1].split("\n")[0]
option_2 = x.split("Option 2: ")[1].split("\n")[0]
and even before giving it x it gives me unexpected EOF while parsing. why tf is this not working?
😭can someone help me to show me a tutorial on how to add the models apart
what do you mean by add them apart
just throw them all in the stable diffusion folder with their own names
stable-diffusion-webui\models\Stable-diffusion to be exact
thanks! i tried it but got really weird results lol i think because of my prompt. would you mind sharing that as well? 
time to go sleep
couldn't manage to solve the error
even chatgpt said there was no problem with it
they say that when you take a break you a lot of the time have a new outlook on things and therefore are able to solve things you weren't able to before
some blood
first time using sd, could use advice for fine tuning
pretty good for first time, are you using a vae?
not sure what that is
you have this dropdown at the top?
your images look desaturated and usually that means you weren't using a VAE
I don't have one to link to you, but I recommend the one in my pic
you'd drop the file into the models/VAE subfolder
the one @pliant epoch linked is OK too
theres also another good vae
https://huggingface.co/WarriorMama777/OrangeMixs/blob/main/VAEs/orangemix.vae.pt here's the one lotta uses, it's just been renamed
ohh what model is that? custom? so cute
i dont see it
that particular image is mostly terada tera lora
hm have to enable quick settings to see it on top
thanks!!!
go to settings > User interface > "Quick settings" and paste in sd_model_checkpoint,CLIP_stop_at_last_layers,sd_vae
then you'll prob have to reload
right i see it now
and when you put that file in the right place and click 🔃 you should see the vae in the dropdown
yep already got it set up
what does it do?
blue wamen supremacy 
Nice military meeting picture, no advice sorry.
magic
Will a VAE change the created outcome, or just its color balance?
Absolutely
mostly the colors, maybe very minor changes
I think vae helps render the eyes too
i see, thx for the advice
no vae vs av3 vs. kl-f8-anime2
in my case, I think the base model might have had a baked vae, not sure tbh, or just luck
you can see how kl-f8-anime2 has really nice colors
but sometimes it's a bit blurrier
VAE is what is responsible for converting the image from numbers into pixels that form an image. If you have an image than you used a VAE, but it won't look good if it wasn't trained on the type of images that are extpected to be output
a lot of models will look washed out without one
kl-f8-anime2 vs av3, this is why I prefer av3 because it tends to look sharper
but the colors are debateably worse
could you link me to both?
oh right i see
I've spent the last two nights merging models for my next TrexMix merge. I've settled on three variants of the merge - one for higher quality, one that's a bit more colored, and one that is very stylized. Thanks @supple raptor for the space girl prompt!
tried but haven't had much luck, like not even a close style 😦 sorry kinda new to it
hm looks like a watercolor
would need to know how you're prompting it
i kept it really simple like 1girl, blue hair , lora:teradaTeraArtStyle_v10Offset:0.6
hmm its anything3
hm
could u tell how this is done as well?
just looking over some old gens, anythingv3 was versatile
this model can't do yandere
Asuna (Small)
||shukezouma, negative space, shuimobysim, portrait of a woman standing , willow branches, (masterpiece, best quality:1.2), traditional chinese ink painting, closed mouth, demon horns, looking at another, wearing long hanfu, hanfu, arms behind back, (riverbank japanese house, raven flying:1.2), (Highest picture quality), (Master's work), (ultra-detailed), 1girl, fullbody, long hair , centered, lora:animeLineartMangaLike_v30MangaLike:1||
how do you get such nice backgrounds?
thank you! mine are lacking the anime face for some reason, weird 
I'm kinda messing around rn doing stupid prompts, could be that but probably mostly a lot of loras and the model I'm using 
fair enough, I need to actually look into loras lol. Where are you getting yours, or do you train them yourself?
civitai, touhouai discord, and huggingface are all good for finding loras
😔
easy to remember though, just open the site x)
Yeah, it's a minefield for sure.
they are pushing a lot too many unstable updates too...
I love new features, but 3 updates in a week with 2 days almost down isn't good
what? 25% daily uptime isn't good enough for you?
I don't get how it's OK for their investors either
I thought this was supposed to be stable diffusion 
civitAI?
just a random site from another company, nothing to do directly with StabilityAI
it was a joke about you saying unstable updates
I'm really curious about what their business model is. I can't imagine their premium membership is even coming close to paying covering their costs.
I'll be better at getting those in 1 or 2 coffee
Yeah, but what is the monetization plan that is bringing in investors and how do they keep them coming back with the amount of very sus, and potentially illegal content on the site.
what premium membership ??
I don't see any way to pay in there
nor any publicity currently
I think they are just going with the "insert yourself into the market by bleeding money" strategy
they'll add a "verified model uploader" mark for 10€ one day maybe
ho didn't see that
kek
Early access to new features
like they beta test those features on the paying users, sure x)
I really doubt it, but they could have a staging server with the updates 1 day before, to test them before rolling them out,
you know, like people tend to do in real business
I like to think its just kept up and ran by some horny billionaire to siphon off all the lewd models/loras easily
That's kind of been my theory
What if ChatGPT spun up the site on it's own when they gave it access to the interwebs 
does this mean a portion of it goes to model makers?
Civit doesn't pay anything to model makers.
or does it all just go to civitai?
well I'm not gonna make them money through my hard work
Nah, all the sites that pay model makers are super shady.
currently, there isn't any redistribution system to model makers. but you can set the "rights" of your model, like if you accept people using it to sell pictures or not.
Technicaly, this is meant to push people to contact model makers to negociate a price on their model
in practice it isn't
people just take models and don't give credit or get striked down
so what are civitai charging for then?
if none of it goes to model makers?
this guy did a nice mix, and got it deleted for using Deleberate in it
development/hosting costs mainly ?
this is strange, it's too little for such an infrastructure
It literally says on their website what you get for the subscription. Their operating costs are nowhere near free.
it's mostly some free cash on top of investments imo but I feel a little sinical
how do they identify what's in it?
scroll up... I posted a screenshot of the page.
What are you talking about?
don't mind me, just posting some monsters
Im talking to @wind girder
Well, anyone respectable lists what is in their mixed models.
they can just leave out the troublesome ones
I moved my DG 3.0 back about 6 merges because 3 of them or so were becoming troublesome
and then started over
there is no system for it, no
the only way they know if when people disclose
it's why I was saying, people tend to hide it more and more
I've also seen cases of people in the community just mass reporting things that they want removed for whatever reason and them getting their way.
probably why I might stop giving credits too, especially since they can retroactively change their models agreement
like, on all my models, I use the same rights :
that doesn't feel legal...
I'm sure people who want the model for a mix just don't care
YES
this was a big surprise for me
they changed all my licences at one point
well they can do it because I might follow their agerement and it might change
allowing selling of the models
It should limit it to the upload and only way to change it is to get a new hash ket
I just don't share on CivitAI. 
key
I do, because I like the UI and it brings some new users around here
but I hate the current situation
so if you identify it with the right key in your model, thats the agreement
yeah theyre stirring the shit
they couldve just made a patreon for models
and merge models have to give a certain percentage to each model they use for the merge
unless its free
They really turned me off when my first experience with them was finding models that were made by people I know that Civit took it upon themselves to upload without even attempting to contact the author of the model first.
yeah
still one of the best sites to find models though
huggingface interface isn't that user friendly and can scare away the average user
or well put them off
they rehosted multiple models of mine before contacting me to know if it was OK and if I wanted to upload the other models I had
directly imported from HF using a tool they made
man, SD really did improve since I last used it
Did you know that cows are secretly members of Discord anime communities and participate in weekly picture-making contests? These contests are no ordinary art competitions - they're actually a cover for a bizarre eldritch ritual that the cows perform every weekend. The cows use their innate ability to distort reality to create horrifying, otherworldly depictions of their bovine brethren, often incorporating tentacles and eyes in their twisted creations. Unbeknownst to the human members of the Discord communities, these contests are actually a way for the cows to summon and communicate with their dark lord, a powerful eldritch entity that feeds on the energy of unsuspecting anime fans. Don't forget to post your donation submission in #1087493722645725184 

https://civitai.com/models/6209/dark-gemini-v21 this would be mine currently, I'm not sure if I want DG 3.0 on civitai antmore
anymore
but is there even any other sites where they would have as good availability?
I mean, if you want people to actually download it then that's the only place worth while to upload it. I just don't really care how many people download my models. I only upload them so I have somewhere to link people to when they ask what model I'm using.
I see
civitAI has pretty good discoverability all things considered
eh... I
I'd say it was good for discoverability when it first came out. It's a bit too flooded now and search often doesn't find things that it should
after DG 3.0 I'm going to make a battle focused model that can actually do things like chop people in half, slice them up, lop off their head etc
It's still way better than HF for discoverability, but that isn't saying much.
I dunno if I will release it publically antmore
maybe on a patreon
then I can make money to afford better gear and train better models
HF has the same issue as GitHub
Yeah, HF is basically just GitHub but specifically for AI.
90% of everything on civit is hentai or porn and 99% of it is identical to each other with only a few standouts in quality
speaking objectively
It's largely just AOM, ChilloutMix, or PastelMix clones.
hassan,improved porn etc clones





if they could fix the DL counter on the models on HF that would help a lot...
Very true. The only models I've ever seen the model counter move away from 0 on is ones that very obviously have thousands of downloads. I kind of wonder if the only time they count them is when someone clones the entire repo instead of just downloading the files they want from it.
from what I get, downloads are counted when you clone, yes, but also when you use the diffusers from another Gradio interface or even at home. it's the cloning of the diffusers that is tracked
and that is just stupid because it's around 10% of the trafic, max
Yeah, that's very dumb.
I can see it on my Experiements model, that has only ckpts in it
None of mine have ever been downloaded 
I'm a little surprised by the numbers today tbh
it's around half or 1 third of what I get on civit
Oh dang
hands ;(
yeah for end users, civitai wins
mmk models/loras where? Im joining in >.<
Model's cetus mix, loras are ligne claire anime and mostima
hmmm so can I get cutoffs into regions or only into prompt sections with certain prompts that shift between each generation?
Star pla-
lets see if I can make something of this pow before it shuts down
my generator giving me this image instead of my prompt which is somewhat far removed from this
What were you even trying to make
the grim reaper standing on a jutted cliff
continues to deliver disaster, gonna see what is causing it
does anyone know how to animate a png with ai?
based64?

da model
smol why no decoration?
there you go
didin't know that works with server profile as well
good job discord

it's gonna be gone after awhile right?
also soundboards
yep
finally, i can play anime girl moans in vc

the revolution is already full steam ahead, thinking about it is so exciting
it funi tho...
I just changed the whole thing for luls
YEEEEES
I can already make anime moans with replica studios
Yup, but i hope it doesn't cause more trouble than benefits
wait you talking about discord plugins?
that is true, one biggest thing to worry about is AI used to hack
WAIT HAHAHA
SEND FILE PLSSSSSS
we are basically defenseless, malwarebytes is still not ai base, either it is a good thing or a bad thing
the biggest worry is probably not ai used by humans but ai using itself going rogue, what a fun provoking thought lol
ai making new ai omg hahaha
At this point in time AI is very much a narrow use case thing
AI babies and its family lol
Once men turned their thinking over to machines in the hope that this would set them free. But that only permitted other men with machines to enslave them.
dune is real
AGIs are a long way off, and I'm not convinced that they will happen in this current iteration of development
i saw a nvidia new graphics card, it sounded convincing
that's probably one of the turning point
The first AGI is definitely gonna come from someone doing nsfw research
100000000%
if it exists, humans will try to fuck it. thats the best motivation
yea i started this for porn too lmao
the porn industry is single handedly advancing our technology 
well, no shame buddy
i must say lol
now one thing we must think is how to utilize this technology on a single entity bases
for money lmao
if not we are doing this out of hobby alone, of which is not a bad thing aswell
money doesnt work with this
the first person to make a good, monetizable use case for AI will be the next silicon valley billionaire
something more communist would be the only way to sorta regulate anything
that is true, but among the users we are in a way advanced in the curved, i calculate only 5% of humans will ever use this technology
I see a lot of parallels with the early days of the internet
if you are from a country that is not advance, it is possible to monetize it, sadly i cannot think of a way yet
https://www.youtube.com/watch?v=sitXeGjm4Mc what did i just watch....
Dw Elon will probably still do it first
this seems like a scifi movie trailer
I would rather not Elon tbh
someone else deserves that money ngl
dude's already earned enough
xD
True
can sell 2d and 2.5d prefabs
NFT industry is plummeting because of this lol
Let's keep it that way
lmao NFTs have always been a speculation nothingburger
always were
facts
it was bound to collapse
what you said i realized how much a speculation can earn you 
now we have to speculate more lol
don't, for every success story you hear you don't hear about the thousands who loses everything
indeed
so for now, back to making art
on that note, we have found image singularity in a sense
Uhh
when you see headlines where people have their NFTs stolen and as a result have to shut down their game, or tv show etc
its hilarious
absolutely deserved
This is the hardest pic to ever drop on the internet, feel free to screenshot 💀
Alright gotta go now
Last one
is it possible to include blip, git and boorugan clips in a lora?
or do I need to pick one or the other?
do you know a prompt for a close up shot?
close-up
thanks
If you mean captions. You can put any text you want into the captions files. It's important to note that over captioning can be as bad or worse than under captioning though.
In general it works best to have consistent captions for different concepts to allow the model to learn them more easily. This is why booru captions work so well for SD training.
anyone knows how can i like improve the quality of the pic? upscaler just makes it bigger not better
run it through img2img again with a low denoise
What do yo udo when an ogre says "Smell my finger" ?
oh and about an hour until I can test my first try on my hellsing ultimate 1 Lora
only has characters from episode 1, so if it works I will make next epoch episode 2 and so on
with the info from previous in the back
and soon I will also generate this (not sure about weights or if it will work)
The young woman birthed by the the marsh and stuck in it forever Her limbs were made of roots and vines and her tears released only tar Her skin was made of bark, her eyes were amber and sap Her smile had massive thorns for teeth and in her gape was a person Cause she was hungry and he stepped on her roots dark fantasy art, horror fantasy art, nu-gothic aesthetics, dramatic angle, cinematic lighting
que? that's not a marsh

I just randomly made this image after a failed attempt at something else and it vaguely fits what you wrote before
can sd do this?
sure it can
lol why wouldn't it? That's a simple pose
hi guys, i made this awesome chatgpt prompt for visual novel stories: https://sharegpt.com/c/dJbZVFv
This is a conversation between a human and a GPT-3 chatbot. The human first asks: write me 15 short 1-sentence visual novel story ideas, based on these rules/factors:
- It IS NOT allowed to sound like an average dumb story.
- Do NOT, I repeat DO NOT add clichés!
- It needs to be non-nsfw, but slight violence is allowed.
- it needs to be a c...
I don't know which artist's style it belongs to
well me neither, you can only try to check out for similar ones
or use google imgs to find author
eh I tried, maybe a bit too realistic and not enough lens flare
Good evening
thinking of using these as examples for my DG3.0 model, but might need some more waifu ones and combat ones
these are also in the running
got a few more that havent been upscaled yet, trying to showcase the improvements
uploading the beta file now for the early draft system and in5 days it should auto-release
gives me time to fix more examples right
and see if I can get the right lora made that will add a lot to the model
how do you put a text in a photo?
Photoshop 
i overclocked my gpu
idk why i didn't do it earlier
webui just launched way faster
didn't measure how much but it's more than 2x
My gpu isn't real
reminds me a lil bit at inuyasha and the miku there^^
Gn all 👋 , this is for the Americans who will be awake soon
deff not what i wanted
The rizz
rev animated?
what do you mean by that?
Spunch bop
The model
oh it was darksushi
the model is pretty good, specially when it's with a complex artistic backgroun
it seems there are really bad models with a close up shot
it cant figure out what a close up shot is lol
Im getting less and less sure if I want mt stuff on civitai
why?
since they will own the right to redistribute, reproduce etc anything they want at any point they want, however they want
you grant Civitai a non-exclusive, royalty-free, perpetual, irrevocable worldwide license (with sublicense and assignment rights) to use, to display online and in any present or future media, to create derivative works of, to allow downloads of, and/or distribute any such file or content. To the extent that you delete any such file or content from the public portions of our site, the license you grant to Civitai pursuant to the preceding sentence will automatically terminate, but will not be revoked with respect to any file or content Civitai has already copied and sublicensed or designated for sublicense. Also, of course, anything you post to a public portion of our site may be used by the public pursuant to the following paragraph even after you delete it.
that extends beyond the selections I make of how I want my model to be shared
in my case, I don't allow derivatives
without direct permission
in particular if it's to be sold
No other host site howver?
I dunno, huggingface I guess? Doesn't matter, I can use my model 😛
Yeeee but doesnt solve youw anting to upload it tho :p
I only want to upload it if they don't have agreements that basically allow them to use the model in their own generators and then charge money for people to go into that one
like if they're just a host for it
then all they need is a non exclusive royalty-free timed license for the duraton the model is uploaded. To display and upload copies of the uploaded file to distribute it as per my settings
Nice
Nvidia?
yes
If I am not wrong, I think it automatically overclocks according to some metrics and stuff, and if you do it manually then you risk shortening the lifespan
Well probably unless you have like an ultra cooler PC setup and are not consuming too much voltage or whatever idk I am not a pc geek
pretty sure better gpus will be accessible to me when this one dies
They will probably cost a house though
i mean something like a 2080 will be cheaper
already?
Probably
damn
Just a guess
50xx series will probably be released next year
gotta teach the model what it means to be attacked and to attack properly so they can actually slice off limbs
well did lose some fingers
Borger girl
big flower
Since i'm running out of inspiration i just throw random prompts and see what happens
Such as crab girl









