🏞｜general-with-images

[split valve]

much higher steps resulted in softer image with a few more details but maybe overworked

wow 1.0 credit

15 steps

1024x1024

[unique pecan]

Temple

[wispy nest]

pov of the robot everytime you ask it to draw her pregnant and barefoot

[long forum]

quick question....i thought 1.5 was supposed to be released a few months ago was it never released? are we still on 1.4?

[wispy nest]

no, 1.5 is out

[toxic condor]

The detective

[wispy nest]

2.x is still spazzing out through its growing stage

[lament anvil]

#1034941531762733167

what it is?

[vague oasis]

that version was released months ago by runwayml, were now at 2.1

[stiff wolf]

hands did me dirty lol

[dense tapir]

Never heard of them/

[wispy nest]

https://tenor.com/view/stoned-cat-stoned-cat-funny-huh-gif-19222132

[split valve]

This is SD, i wanted to take a historical painting and try to make SD turn it into a photo...

I split the original painting into 4 tiles and used SD at max res to get all this detail

then stitched them all back and photshoped it

here is the original painting

[dense tapir]

Painting is so close to real SD didn't have to do very much

[split valve]

yea SD did not do much because i made it copy it at 99%

[dense tapir]

Fantastic painting

[split valve]

but if i turn the img2img strength down it will stray from the oringinal and try to do less detail

i could try it though

and its pretty cool to take a really high res painting and split it into tiles, that way SD gets more resolution for deail

[dense tapir]

well, the new depth map thing doesn't do that

[split valve]

i treid to negative prompt "painting" as well

[dense tapir]

what I do not get about the depth thing is that it is 512

[split valve]

i dont even get what the depth to img is

[dense tapir]

I finally do

[split valve]

it looks like style transfre

[dense tapir]

well, sorta is but you prompt guide it

this isn't me but it explains it where even I understood it.

[split valve]

it looks like the depth map obtains depth ie forground and background and shapes, so it can feed that data to the img to img

[dense tapir]

yes

[split valve]

so the style transfer will be more accurate

[dense tapir]

yeah

really cool changing fore/back ground at will

[split valve]

i was thinking there could be a fine tuned SD model that turns historical paintings into photos

and that can do tile splitting to get the full detail

[dense tapir]

We are barely touching the tip of what can come

[split valve]

quick ban it

extradite these criminals to China

[dense tapir]

to work making Iphones

[split valve]

yes, as slaves

[dense tapir]

chained to their desk with suicide nets outside

[split valve]

yup

omg just thought

if i split the original high res painting into enough tiles, theoretically SD could add more detail

or infinite detail

sort of like the ultimate upscaler

its just a pain to do with photoshop

[dense tapir]

yeah, need some automation

[split valve]

needs to be an option built into a future version of dreamstudio or the other UI versions

it could also restore paintings digitally getting rid of noise and scratches and faded color

[dense tapir]

Well, I am so damn sick of seeing browser based Ux.

gradio is shite tbh

besides browsers are slower than a real UI and devour ram

[split valve]

Somebody should make a mega blender addon with it

i already have the rendering addon

[dense tapir]

never liked Blender as my thing was old old Maya (pre AD) and Cinema4D

[sharp sigil]

Hello

[split valve]

if i can make some money next year i want to move on from the browser version and run it locally

but i shall need a new nvidia card

[dense tapir]

yeah, I am saving now for one but ffs I think they are going to screw us over on the 4060

[split valve]

for sure

[dense tapir]

I have my limits of acceptability.

[split valve]

im sitting out the 4X series

even with my GTX 1070..;. im not going to be scammed with their shit

[dense tapir]

they said 379 USD and 16GB of vram

[split valve]

expensive.

[dense tapir]

already the scalpers have 4090 so a used 4090 is now worth a lot more than when it first released brand new.

[split valve]

my gtx 1070 was a similar price back in 2017 and had 8GB

six years to double the Vram... thats a joke

[dense tapir]

I agree as my limits keeps going up due to them. $199 to $239 to $299 to $399 now. I refuse more

[sharp sigil]

I am attempting to turn this sketch of an eyeless armless monster into a realistic figure. I've tried several prompts and erased empty space ut it doesn't see it as a full body creature. What do you suggest

[dense tapir]

the reason is because they are protecting their pro cards so they limit the ram

[split valve]

let me a trty

nope cant do it

[sharp sigil]

Thanks for trying. I think I'd need to edit the image in a sketch app very to define the lines. It's a cellphone picture from 2012

[split valve]

yea i tried a bunch and it kept giving blurred result. you need more detail

[dense tapir]

Clyde

@sharp sigilhttps://i.imgur.com/YpxftFU.png

@sharp sigil

[sharp sigil]

Was that the generation? It's having a hard time with it.

[dense tapir]

Both are

The problem is it is a low contrast, highly noisy pic

[sharp sigil]

The skin would be smooth and wet like an albino salamander or something nocturnal.

[dense tapir]

The prompt it

*Then

All I prompted was "eyeless armless monster"

[sharp sigil]

Yes I think could edit the drawing to be just line art

Is img to img on discord simular to midjourney? I post a url link to it in the prompt?

Didn't work. I don't want to waste my credits.

[wispy nest]

while I only use my own "local" version of the AI, what is it you're after? to get a character based on your drawing with img2img? It might be hard for the ai to "understand" what is what with only two colors. and the prompting still matters a lot. From my knowledge and experience, img2img is just a way to slightly control where what goes. But where you got a head the ai might "think" is the flame from weird lighter ;P

[dense tapir]

@sharp sigil

It is fighting it

[split valve]

im working on an 8X tiled version of that Rome painting

[dense tapir]

It demands giving it arms

[sharp sigil]

Yes like my original character sketches, finished with sharp lines and color and scaled up to be printed or used in short stories and make book illustrations

[wispy nest]

it depends on the quality, and time, etc, but it can be done, I tested it some months ago and the results shocked me in that the ai really did "understand" a lot. In the end, it's all about the knowledge how to do it. I wouldn't recommend someone to start out using img2img, I think txt2img is more important as it "teaches" more what your words mean to the ai, and how it even works, etc :)

[sharp sigil]

This is a sketch I colored with a paint program nearly 10 years ago. Could it see the articles of clothing and make a model to wear what I designed?

[sharp sigil]

I've been using midjourney and stable diffusion for about 3 months just using my mobile device and discord. I goal is to complete my art portfolio and print the finish products

[wispy nest]

yes, and no. You can make the img2img image in "steps." You let the ai make changes to the image above, then use that image to make the next, and then that to make the one after that one, etc. The ai might change stuff, but then you just doesn't use that image when you try again. The more specific stuff you want, the harder it'll be to get it correctly made. but it is possible! :D

[sharp sigil]

There's somethings I welcome to change from the original and can be better than what I'd do. I definitely see it as a collaborative project

[wispy nest]

you're free to PM me with the results if you want, might be easier to help. That is, if you even need it in the first place :3

[sharp sigil]

Thanks!

[vagrant magnet]

you could try isolating each limb or section with layers and masks and then inpainting

[split valve]

SD 4K res. 90% likeness to the original painting

for reference here is the oringial

if i went with 16 tiles SD could have gone beyond the resolution of the oringial painting

prompt was Digital Photo Rome on a clear day. | pencil painting ugly harsh undesirable dust dirt draft sketch blurred amateur signature text watermark artifacts logos fog mist haze hazy: -5.0

and the SD image is clearer and sharper

probably really worth doing a 16 tile version

took a bit of photoshop work to try to blend the seams

[rain gazelle]

Neat

Give the bigger one a try

@sharp sigil - let us see how your img2img comes out from your drawing!

[split valve]

sure... just paypal me some money from the credits on Dream Studio

imagine a 32 tile version with so much more detial

[sharp sigil]

alright !

Runway model lmao

[rain gazelle]

might want to run that one through a bit more 😄

[wispy nest]

made this

[formal flume]

dream booth + deforum

Did a rotation around the Y axis and a translation on the X axis to rotate around subject a bit

[worn coyote]

Hi,does anyone know how to avoid this? using 2.1, at like 80% in the top image, it looks good but then it finishes and turns horrible, I'm not sure what the name of that effect is or if it's possible to get rid of it with a negative prompt? I've tried a lot of samplers and step/cfg configurations and it always seems to be there?

[rain gazelle]

@sharp sigil , I'll give it a try. Here's the first one it spit out.

[sharp sigil]

Hey not bad. It got the plaid skirt. And several pieces of clothing simular

[rain gazelle]

next few were ok but cut off kinda weird, so ill try a few more

matches the pose a bit more

[sharp sigil]

It's like a short fur or wool vest on top of a pink chiffon blouse

[rain gazelle]

hmm ok

[sharp sigil]

It did make a pretty close match to the general shape of her pose. The shoes are cute off due to poor scanning of the original. But it's almost 14 years old sketch

[rain gazelle]

will also try non-anime style, but that can be hit or miss for me 😛

[sharp sigil]

Something I have done is take the anime result then make a realistic photograph

[sharp sigil]

This was another drawing I attempted to make into a standing figure

[rain gazelle]

I think I've maxed out what i can do. LOL

[sharp sigil]

Now that's my kind of anime style. Classic turned out very close

[rain gazelle]

Gonna try a realistic one, but not sure how it will go.

[sharp sigil]

Hey you did light years better than I have in minths

[rain gazelle]

oh .. oh man.. oh no...

the realistic one is ... not good

lol... need to tweak

[sharp sigil]

Oh nooo I can imagine

[sharp sigil]

I'm saving these

[rain gazelle]

go for it

these are horrifying 😄

[sharp sigil]

Thats great!

[sharp sigil]

It's got the gist of several articles of clothing but also not

I gotta get ready to do some errands. Thanks for all of your help!

[unborn osprey]

I've worked 4-5 years on a raymarching fractal game (realtime audio driven 3d fractal worlds). and always found it to be a super zen "cloud watching" experience. I am absolutely amazed at what SD and the likes bring to this experience. Here is a Friktal screenshot I printed and has been on my wall for a couple years now. And then an output from SD w a prompt describing what I see in the fractal clouds.

another example... Friktal SS:

SD output "Cute Owl in Hyperspace"

image to image w ~50% new image weight

[split valve]

check my older stuff for loads of fractal renders https://www.deviantart.com/sono2000

[unborn osprey]

looking... thanks for the share

[unborn osprey]

stunning stuff

[split valve]

i even rendered a 3D fractal in an glass orb in blernder https://www.deviantart.com/sono2000/art/volumetric-fractal-snow-globe-II-873012177

[unborn osprey]

really badass render

[split valve]

ive just finished my 16 tile render on stable diffusion get ready for this.

here is the original art work https://cdn.discordapp.com/attachments/1004159122335354970/1054447463693557790/d976f9d4bf6a2d64e7483f66ad2e9562_1.jpg

i did img2img with stable diffusion at about 90% strength

here is the result

[unborn osprey]

still amazed at what ai art engines are able to pull off... and this is just the beginning

https://youtu.be/l6Bqj0Dl55U?t=114

^^-- looking to use AI to help create hyperspace "entities" / life for the game. looking at ways to animate and bring to life outputs like that Owl/Jaguar overlaid in the fractal. Will share more when I get something working.

[long forum]

you have a link by any chance on how to update/download new

[vague oasis]

sd 1.5 is here: https://huggingface.co/runwayml/stable-diffusion-v1-5/tree/main

"v1-5-pruned-emaonly.ckpt"

[long forum]

ty! idk why i cant see that, it had all the way from 1 original to 1.4 in the index and i just couldnt believe it was still on 1.4 lol

[vague oasis]

yeah its cause it wasnt released by stability, but by runway

[long forum]

oh im looking at the compvis model not this one okay

any clue what ema is?
would be a good idea to google stuff before i ask and look like an idiot lmao

[vague oasis]

Ema is for generations and non-ema is for finetuning

[long forum]

ty

[split valve]

What we need with SD is a system where you can input a large image and it breaks it into tiles and then assembles the tiles when its done

[rain gazelle]

Today's favorite Tifa so far 🙂

[hasty kettle]

Hi! Anyone knows a prompt to do something like that?

[wispy nest]

incredible. can you explain why its a tile render? not familiar with that lingo

[split valve]

since the image i used was a large painting of approx 7000x7000 resolution, i could break it up into tiles using photoshop. Then in dream studio i fed each tile one by one into SD keeping the same seed and prompt and with the image strength of about 90% and after they were all 'remixed' and done i stitched them back together in photoshop with a large canvas using layers.

And with a cleaner painting not full of dirt and dust and cracks like this one, i think SD could actually add more resolution and detail than the original with enough tiles.

[wispy nest]

https://tenor.com/view/stoned-cat-stoned-cat-funny-huh-gif-19222132

[split valve]

8 tile remix of a public domain painting

i did image to text for each tile... probably should not have as it made the tiles not match that well and i had to do a lot of photoshop on the seams

SD 1.5

[wispy nest]

why 1.5?

[tame forge]

My new wallpaper, all done from scratch with stable diffusion 1.4 model, SD Dreambooth training, inpainting, InvokeAI outpainting, and Topaz Gigapixel for enhancing sample images. Probably will use Photoshop to fix her neck and a few details

[dense tapir]

I tried to enhance using gigapixel and a few other things. Looked great to me but the AI saw stuff I didn't so came out terrible. I then used some REALLY garbage images (only ones I could get) and Dreambooth floored me.

[tame forge]

How strange. I had the exact opposite result. I had a bunch of images that I thought looked good but the output images all looked blurry as hell. I decided to use a Gigapixel to sharpen them up and holy crap the difference was night and day. No more blurriness.

[dense tapir]

Interesting. I mean I decided to just try it and forget it if it failed due bad/lousy source images and what came out was just WOW.

@tame forge Most where rubbish images (done for 1.5 as I refuse to touch DB until 2.x settles down). https://old.reddit.com/r/StableDiffusion/comments/yyc4l3/by_your_command/

[tame forge]

They are definitely overcooked by gigapixel. GP is much better for faces.

And yes DB 2.x was the disappointment of the year but I'm pretty sure something good will come around soon.

[dense tapir]

well, 2.5 vs 1.5 I will then make the final decision but 1.5 vs 2.0 was unfair for people to decide like that due to everything, pretty much, starting over.

If SAI can't get its act together by 2.5, or go to this "for pay" pro nonsense to actually make it viable then yeah, I'm out.

[dense haven]

what am i doing wrong with outpainting?

[dense tapir]

Only do one side at a time not both.

Then play with all those sliders

[last rose]

No upscaling or anything, using HD model. Happy with these.

[granite heart]

I made a thing! https://emoji-diffusion.lol/
it let's you mash up 2 emojis together into stable diffusion

works really well with animal emojis 🐇 🐯

[dense haven]

My characters' heads keep getting chopped off even though I specified ((no crop)). I know outpainting exists but it's super unreliable. Any solutions?

[frigid radish]

@dense haven try like

Full-body

[dense haven]

How can i force it to draw a hand? It keeps wanting to draw objects

[split valve]

It sometimes gives more realism for certain things. Although the other 16 tile large painting i did was using 2.1

SD 2.1

[split valve]

ok i am now addicted to doing large remixes of paintings

8 tiles for high res

as long as they are public domain i should be fine

even so i dont think there is a law against copying somebodies painting in real life.. as long as you dont try to sell it as the oringial

[dense tapir]

I personally don't care if PD or not. If it is on the net it is fair game as far as I am concerned.

Exactly

[split valve]

i mean yea, if i saw a piece of art if i want to copy it or trace it then i can

my dyslexia gets worse

i think covid fried my brain

[dense tapir]

I had the alpha and it sucked. Worst I have ever been sick in 50 years.

[split valve]

ouch,, well on the new year day 2020 i had a very high fever /chills not sure if it was alpha

[dense tapir]

Got over it though, but damn. 103.7 fever for a couple of days which is the second highest fever I ever had.

[split valve]

worst chills ever. but i woke up in a major sweat, had a shower and then i recovered

[dense tapir]

Feb 2020 got me. 1 week of near death like and 2 weeks of weakness.

[split valve]

what i had probably was not alpha then.

i had Delta....

[dense tapir]

felt like an elphant was on my chest and I don't get sick easily

[split valve]

ouch

well the Delta i had was not nice, but was not that bad, razor blades sensation in the nose, and very dry nose, lose of taste and smell for weeks.... night sweats for a while and a bit of a cough..

really not that bad

[dense tapir]

I got something again in Aug of 2020 with a slight fever of 101 for a day then POOF. Nothing like that Feb crap.

[split valve]

omicron is like nothing... im sure i must have had it and passed it off as a cold

[dense tapir]

Well, if you had the real deal herd immunity is a thing

[split valve]

oh yea i was testing positive for over a week,

im immune now im sure

[dense tapir]

reason people had chicken pox parties for the kids so all would get it since nothing bad as a kid (unless aspirin) but as an adult look out.

[split valve]

yup... i had my cousins round to catch chicken pox when i was about 4 or 5. lol

[dense tapir]

I caught it when some kid in my 4th grade had it so 9

[split valve]

the immune system is real...

[dense tapir]

yep

[split valve]

gives you long term variant indifferent broadspectrum immunity.

T-cells

[dense tapir]

Which is why HIV is so bad because it eliminated T-Cells then full blown aids that just opened you up to everything since your T-Cells were pretty much gone

most times you perished of X due to complications of having AIDS.

[split valve]

right thats so bad.. Scientists must work on an immune system transplant

[dense tapir]

Well, modern medicine is more about big pharma getting rich than about cures.

[split valve]

yea.. Thats why we have not come very far.

if it were not for big pharma and their profit we'd have a cure for most disease by now

[dense tapir]

I think the last thing we cured was back with Salk but then they used liquid polio vaccine which is not as good so here it comes back.

I 100% agree

I mean, let's look at something that baffles me. Stem cells they are using for various stuff. Stem Cells they are growing cow meat to eat (not me) BUT if stem cells are all that (which I have believed they were for 25 years now) then why are we not growing new kidneys, etc... with them?

no rejection drugs needed and millions of lives saved.

[split valve]

well my theory is that we just dont realize the sheer complexity of the human body and its biology

we discovered stemcells decades ago yes...but to use them in useful ways... scientists and doctors just dont grasp the technology yet

[dense tapir]

I think we do but $$$$ is the important thing not saving lives.

[split valve]

the body is ridiculously complex on the cellular level

i think if they truly understood stem-cells there should be all kinds of markets for them to make $$$$

[dense tapir]

Well, how in the eff are we about to sell lab grown meat from cow stem cells if we don't know? My elderly mother is about to get stem cells for her legs. Six months and pretty much done. 4 injections I think it was.

[split valve]

think of the hair products they could sell if they cured baldness

[dense tapir]

Precisely

[split valve]

The way i see it, its levels of understanding.. its like graphene...

[dense tapir]

BUT more money selling other things

[split valve]

we can use it a little bit... but we are not masters... the human body is too evolved.. it tends to reject things and each person has their own frequencies and genetics

its like if we got hold of Microsoft windows 20.. from the year 2050... we could use it a bit, and learn to do a few things, but we just wont know how it works

[dense tapir]

I am more cynical than that and feel we could but why do it if big pharma, and hospitals, etc... would cease to make all that moolah?

[split valve]

and imagine it was coded by super strong AI

naa. scientists are meddling with shit they do not understand.. they always think they do, but the level of complexity just keeps baffling them

like the mRNA vaccines

they thought when they deocded the human genome that it would be the holy grail... but it gave them nothing...

then they went on to decoding the human protiome which is how every protein the human body folds...

they will decode all that with Alpha fold,. and realize they still know nothing.

human body is advanced nano technology

[dense tapir]

Well, gonna bow out of the convo before mod jumps us.

[split valve]

until i see bill gates or jeoff bezos reverse aging and living past 150 whilst looking 30 or 40... i know that they still have no clue

yup..

im going to have a go at a new painting remix

[dense tapir]

not sure what you are doing. My new thing is emb bashing

bizzare effects not what one would epect or get if just layering them.

@split valveDo you know of a site like Lexica but with 2.x stuff?

[split valve]

nope.

[dense tapir]

:/

Perfect

[grave tinsel]

Can anyone make some images for me

I’ll pay cash

[split valve]

why cant you make them in Dream Studio?

[grave tinsel]

Idk how to use this

I have never done this before

[dense tapir]

@grave tinselhttps://discord.com/channels/1002292111942635562/1047760914008522782

[split valve]

it's not exactly difficult

its basically a magic art button

[grave tinsel]

Yeah but if I sent some images can you do some for me

[split valve]

i would normally but this evening im going out

[grave tinsel]

It don’t needed be rn

[split valve]

if you paypal me some credit for dream studio

[grave tinsel]

Cash app?

[split valve]

i dont have it.

i have some crypto wallets

depending on what you need £1 should probably cover it

DS is not expensive

[grave tinsel]

I am in US

Ion do that shit

[split valve]

no paypal?

here is an 8K stable diffusion remix i made today

remix of a public domain painting, at 91% likeness.

had to stitch it together and try to photoshop all the seams...

[rain gazelle]

I run sandboxed. Left a generation running. When I came back, some process started under rundll32, which is.. weird.

Going to see if I can replicate, but thus far not successful.

[split valve]

sounds complicated. watch out for viruses

[rain gazelle]

well, it's a clean sandbox, so the only things in it are python, git, and my SD repo.

It was the end of the day and I was just shutting everything down, and I saw it right before closing. 😦 So I blew it away before I could investigate.

[split valve]

probably microsoft uploading all your private data

[rain gazelle]

No, because that doesn't run in the sandbox. 🙂

Probably some function in SD calling a needed DLL for a specific thing I did, but don't remember doing. 🙂

[split valve]

what do you mean by sandbox?

virtual machine?

[rain gazelle]

Sorta like that.

but not a whole OS, it's an abstraction layer so it uses hardware directly

not sure SD would run in a full hypervisor...

maybe one of the professional vmware tools with full gpu passthrough, I dunno

[split valve]

i dint realize VMs were limited like that

[rain gazelle]

yeah, they use their own "virtual hardware" layer

[split valve]

what about Amazon AWS could that run SD?

[rain gazelle]

cloud instances with direct hardware access can

That's how colab and runpod work.

I don't know much about aws.

But like... local Virtualbox... that isn't gonna do it. 🙂

[split valve]

so its all about direct hardware access an no virtual layers

[rain gazelle]

Sort of. A virtual layer that does proper hardware passthrough might work?

Not something I've tried.

[split valve]

can it work on CPU

i wonder if they could build an ASIC for it.

[unborn shuttle]

has anyone tried Web UI Designing using Stable Diffusion????

[rain gazelle]

no but that sounds interesting

try it

[unborn shuttle]

I never actually tried it, I thought if anybody knows it then I could get some help with prompts so that it can generate UI designs.

[wispy nest]

sick and tired of using SVG? ;P

[old fiber]

can someone give me an example prompt + negative prompt? My SD images are so ugly compared to what I see here

thanks

I'm using v2-1_768-ema-pruned.ckpt

[wispy nest]

most people use embeddings here I believe so it's hard/impossible to get those results without them :/

[old fiber]

oh, how do you load embedding in stable diffusion webui ?

[vague oasis]

what kind of images are you trying to generate?

[old fiber]

just basic landscape image like midjourney, and portraits

are there any public embeddings I can download ?

[wise garden]

https://huggingface.co/sd-concepts-library

[vague oasis]

give me an image and i will try to recreate it with dreambot

[obtuse wagon]

My first test of Point-E. Image generated with Stable Diffusion, Model generated by the new Point-E.

I used this Colab (https://colab.research.google.com/drive/1jXHyRY9cQHT6JOqbhRyB-jc9Y2-1aDKR?usp=sharing) made by https://www.reddit.com/user/dismantlemars

[errant leaf]

#1047197565365538826

[wispy nest]

This had been my latest project.

[gritty cedar]

@errant leaf It basically does not work on 1.4. have been downloading 2.1 and trying anyways while I wait.

[errant leaf]

1.4 is very different from 2.1 so I'm not suprised lol

[gritty cedar]

best result so far

not ... terrible

not that good either

[errant leaf]

compared to what the embedding is supposed to do...

[scenic carbon]

fan with a procedural drawing in top view::1 vector digital, flat Miyazaki, Monet, hd, rule of thirds, symmetrical, palette, colorful, with a procedural drawing and one black hole

[cyan snow]

dude this model is awesome

[gritty cedar]

I actually want to drive that middle car

[sonic vessel]

3k upscale

https://images.seek.art/3eef9cefa4259d4d4706207425557ac7.png?auto=webp&optimize=high

[split valve]

random question. What tv shows are similar to Altered Carbon?

[sonic vessel]

Expanse

[split valve]

hmm i watched a few series of that...

probably worth a re-watch

[cunning geode]

my mums making fluid art
two examples
https://cdn.discordapp.com/attachments/856030898805145611/1054893116987035668/20221220_234715.jpg
https://cdn.discordapp.com/attachments/856030898805145611/1054893219130917025/20221220_234737.jpg
second one looks like a person
https://cdn.discordapp.com/attachments/856030898805145611/1054893962873286656/20221220_234737.jpg

[pallid ruin]

It looks nice 👏 🤗

[simple star]

Thoughts?

[dense haven]

is there a way to get consistent eye styles for my characters? i have been including 'eye highlights' in my prompts without much success

[dense tapir]

https://i.imgur.com/TohFd1x.png

[analog goblet]

hi guys

[hardy anvil]

How can I generate an image like this without any embedding?
Image
parameters

orange fox by Andrew Alex, beeple, volumetric lighting, fur, chempunk-96
Negative prompt: blurry, disfigured, ugly, out of shape, cartoon, animation, extra limbs,
Steps: 20, Sampler: DDIM, CFG scale: 7, Seed: 729875313, Size: 768x768, Model hash: 4bdfc29c
chempunk is the embedding btw

[cyan snow]

wtf

[cyan snow]

ok that is enough internet for today

[split valve]

My thoughts is they are going to have to backtrack again and grant copyright. Because nearly everything in the future will be generated with AI tools to a large extent, and millions of people are going to expect and demand copyright...

[sterile kiln]

Nice

[dense tapir]

Thanks

[rain gazelle]

If that comic loses copyright, then I'm going to demand access to the YouTube algorithm.

As if companies haven't been using AI-trained data on MY viewing data for years.

I'll also be requesting the netflix algorithm.

And the google search and ad serving intellectual property.

I never opted any of my websites into any of the search crawlers, actually. So, I'll be getting full access to them all since they'll all lose their copyright protections.

Right? Isn't this what will happen?

laughs hysterically

[wise garden]

:3

https://tenor.com/view/kyubey-madoka-magica-kyubey-madoka-magica-gif-26136775

[cunning geode]

I've updated the model card of my "Darkest-Diffusion" model on Huggingface again with additional information + an Img2Img guide of taking an output from the vanilla v1.5 Stable Diffusion model (second image) and applying the Darkest Dungeon artstyle to it using my model (first image)!

[split valve]

human avatar hybrid

look at this art theft

[dense tapir]

Yeah, my emb I trained this morning but I want to go more than 5k epoch

[split valve]

what is an emb?

[dense tapir]

embedding

[split valve]

what is embedding

[dense tapir]

https://www.youtube.com/watch?v=MLz0iM0M7Fk

[split valve]

you know... if these artists on artstation realized they could fine tune SD with their own art they wouldnt be so quick to call an AI ban

they just need to see ways they can utilize SD as a tool

[dense tapir]

Yep, but they are artists so are left brain thinkers and all this is "scary" right hemisphere thinking.

[split valve]

its a tool at least as versatile as photoshop

[dense tapir]

They will come around in time (well, most will).

[errant leaf]

**CarHelper embedding has been added to #1047197565365538826 ** 🙂

[split valve]

https://www.youtube.com/watch?v=raaC9IAeONg

The art snobs thing they are winning.

They dont want anyone to study their art, especially computers

They think they have special rights of protection from AI

And they think they can stop AI

[golden umbra]

This is probably just delaying the inevitable, image generation is likely to be fully solved next year

[split valve]

These bunch of idiots thing they are going to stop AI... just like the "stop oil" protestors

useful idiots

[obtuse wagon]

1. Generate spaceship model with Point-E
2. Import into Blender
3. Remesh model with lower polycount to smooth out ugly bits
4. Sculpt a couple parts to improve shape.
5. Mirror across to create symmetry
6. Create camera and align viewport to camera.
7. Use the Dream Texture Projection plugin for Blender to project a texture onto the ship.
8. Create a plane and align it to the camera view behind the ship.
9. Use the Dream Texture Projection plugin to project a starfield texture onto the plane.
10. Reproject to create endless variations!
    Done!

[split valve]

yea this is Blender renders, the fish were created with Stable Diffusion, yes they are 3D yes its a 3D scene

yes i am leveraging AI to make even better art in 3D and sell it

Stable Diffusion is a tool

Stable Diffusion in-painting

[sleek needle]

working on some dreambooth dragon portrait models

[obtuse wagon]

What's your current method for generating 3D models?

[split valve]

i use a program to generate AO, normal, roughness etc maps from the SD image, then i use displacement modifier and mirror modifier, on a subdivided plane with the textures set as its material.

[split valve]

look what happens if you feed SD texture like image such as this, and then feed in a prompt that has a lot of artist and keywords

see what its doing

feed it a grunge texture like metal rust, tell it to match it by about 70% but use a prompt

[split valve]

creepy or what

this is the image fed to it. a wood texture

wow

wood textures so useful as a guide

[tropic shell]

I like it

[tropic shell]

I really love all the eyes. Super great

[hardy anvil]

What's this style called?

[dense tapir]

Looks sort of like a redshift

[gritty cedar]

can I post girls kissing (0 % nudity) just a little... passionate

anyone know?

ok read the rules prob not

maybe if you mark as a spoiler, give a warning and 0 nudity, some suggestive stuff should be allowed

Actually just make a NSFW channel

[dense tapir]

I would post that regardless if I were into that.

if it is SFW, which it is depending on the kiss, then bugger off puritans. just my take.

Interesting but this new 100 step (took 10 mins locally) emb I just made turned all my people older.

[wispy nest]

https://tenor.com/view/elf-santa-is-coming-merry-christmas-christmas-excited-gif-15879777

[frigid hill]

lebron james as blade in action

[lucid river]

the captchas in colab have me questioning if the pixel is included

[molten river]

SD running in the metaverse! Prompt with friends, try it here: https://www.simulacra.io/ai.html

[gritty cedar]

I have been struggling to make objects out of gasses, fluids and plasmas, they never even get close. like a heart made of fire or water, or a person made of electricity

any thoughts?

[errant leaf]

for water and fluids, refractive and transparent can help

[white pivot]

anyone has a really strong photorealistic prompt ???

for SD2 ?

please ?

I experiment multiple ones

but Im not satisfied

[split valve]

if using 2.1 you're not going to get the best result from just a prompt, you have to give it a starting image and work from there

you have to feed the images back to it

[gritty cedar]

yeah, go on google grab the pose/ expression you like and experiment with different noise amounts

im going to do pirate stuff for a bit

[white pivot]

and from SD2.0 ?

[wispy nest]

Karlos is something to behold

[crimson flower]

Dreambot gives a thumbs up.

[quiet knot]

Santa and Mrs. Claus 😄

Uhm. What? 🤣

"A photo of Batman dressed in a santa costume"

[errant leaf]

Elonites
..
what

[wispy nest]

https://iqfy.com/ai-vs-artist-controversy/

omg its real

[errant leaf]

to be fair, it's on the same website that has bangers like 8 insane Tumblr moments that went down in history, Is cuddling a form of cheating? and These are the signs you've been abducted by aliens, according to UFOlogists

[wispy nest]

whose side are you on? the bloods (being the cult of Tay), or the crips (being the Elonites)?

lol

[errant leaf]

the article is so full of reasoning flaws and contradictions it's not even funny anymore

[gritty cedar]

https://iqfy.com/ai-vs-artist-controversy/

"it should be in real artists toolboxes"

what the hell. so were not worthy? we have made no good images? stealing was a problem before ai

its not like ai popped up and suddenly stealing art was invented

I dont think we should be mean, verbally at least. lets win this war with pens, pencils and (almost entirely) prompts!

[dense tapir]

You know that is very childish and sure doesn't help our side at all, ESPECIALLY with all this woke nonsense where words hurt and "hate speech" nonsense. Are these plants to sort of help destroy this? Normally how psyops works especially with how social media works and the dog piling of the NPCs.

[wispy nest]

theres no way in hell the articles thing is real

[dense tapir]

Could be part of the Psyop.

Fact is the govt is scared shitless of this being in our hands for numerous reasons.

iow, let's do a psyop so we can try and shut it down by being the good guys and removing the hate, etc... Not buying it

[gritty cedar]

its a copium piece to make artists not sad and depressed. instead they can be angry at something, much more productive 😁

[wispy nest]

coping so hard you give your enemies two insults against you for free

lol

[wispy nest]

Does anyone recognize this song, please? 1:30 https://www.youtube.com/watch?v=0TB79fG8WyQ

[dense tapir]

It is a remix of a 1960s/early 1970s song.

[modern prism]

google is watching

google ((is watching you)) (((eye))) 1984 (surveillance) 8k very detailed camera tracking (((old poster))) journal ((menacing)) communist (starring) russian lens optic angry red white black

duplicate ugly deformed blurry face building tree people human bricks head colorful product machine smartphone helmet green blue beige pink woman girl skin orange yellow

[vague oasis]

💀 this is literally 80% steven universe artstyle

[wispy nest]

Does anyone know this song, please? 1:30 - 3:00 https://www.youtube.com/watch?v=0TB79fG8WyQ

[split valve]

That did not age well. lol. AI art is now so impressive i prefer to look at the AI art on deviant art than the real art.

And its not stolen, its just influenced by the training data(which was freely available and not stolen).

machine learning created the latent space and its 500+ dimensions, and then convolutional neural networks working to de-noise the random gaussian noise based on my prompt. It will never result in stolen copy and pasted pixel for pixel art.

Because it does not work by stealing art from the training data.

literally thats not how it works

5.8 billion text to image pairs were fed to the model to train it

thats hundreds of terabytes of disk space

yet i can download stable diffusion in about 5GB and run it without the internet. That is a miracle considering the training images were probably all jpg

[wispy nest]

I'd say that words have meaning. If someone believes something is stolen. Then they should talk to the police. Everyone can say whatever they want. The difference is if it makes any sense or not. ;)

[split valve]

well when they complain their art is stolen i would say PROVE IT

[vague oasis]

she throw a tantrum and left the server

[split valve]

ok

well a lot of artists now believe the LIAON-5b dataset was stolen.. and they think AI images are reaching from the training data and copy pasiting.

they are making fools of themselves

[wispy nest]

yeah, it's up to them to prove it, and if they really believed something was stolen. Then I hope they have already reported it to the police, should be the first step and only they can help them. I never understood why they'd not focus on that part. Or I know why, but I don't want to say it ;P

[vague oasis]

this is important, like, how can she explain that that much terabytes of images can fit into a 3-4gb file

its cause theres no images stored there

[split valve]

5.9 billion text to image pairs.

Your computer is lucky if it can hold a few 100 thouasand photos

the judge would have to throw it out of court.

Especially when you prove the AI is working with a random Gaussian noise starting point and diffusing with CNNs whilst reachign into a machine learned 500 dimension latent space of concepts it has learnt

[wispy nest]

And I believe this thing is a civil matter, and those things take such a long time to even get started. Better start earlier than later :/

[split valve]

Artists against AI, have conveniently forgotten that right click copy > Paste, has existed on PCs for 30 years?

[vague oasis]

and as i said, her artstyle is very influenced by steven universe, specially the star pupils...

[split valve]

it does not take an AI to copy a digital artwork

[wispy nest]

In the end, it doesn't matter. All that matter, is what they can prove, and words alone won't do much, I believe. :3

[placid kraken]

i don't know why people think digital art is superior to ai, ai is the future and the future is now

[split valve]

I think judges and governments are going to look at this and think,well this AI is amazing, and world civilization is in crisis because of disease, famine,war, climate, limited resources, corruption etc. Why would we try to stop the fantastical technology, it could be something that could save us in the end.

[wispy nest]

I feel if anything, this will be settled out of court so we won't ever know the solution :/

[placid kraken]

yes i agree! ai art will save the world AND it makes art accessible to everyone

[vague oasis]

the fact that digital art was also bashed in art communities before ai art became popular tells a lot, they hate new things

[placid kraken]

yes this is the truth! the art community is very often toxic and they hate us for being different

[vague oasis]

and photobashing was/is also very taboo in their communities

[placid kraken]

if anything it's digital art that should become restricted not ai!

[split valve]

Snobbish artists with their anotomy sketches all of there walls, that think they are supremely special because they honed a skill.

[sage kindle]

I mean, skills are valuable

it's the gatekeeping that's a problem

[placid kraken]

not digital art made by snobbish digital artists however

[split valve]

They just like to feel like a scholar from 100s of years ago that was so high society they did not need to work a regular job.

So they could do art all day

[vague oasis]

also, artist is an umbrella term: musicians, actors, photographers are also artists

[placid kraken]

i hope in the future music, film and even video games will be able to be created by ai

[wispy nest]

I define artist as someone who believes themselves be one, or that someone else believes they are :3

[placid kraken]

the anti-ai people will be so mad!

[split valve]

heck i play guitar in a covers band.. i could not care less if an AI creates music.. it wont affect me

[placid kraken]

maybe we'll even have whole ai generated bands with sentient ai!

[split valve]

but even if i was a producer or composer, i would be forced to see the bigger picture, which is we have to accept this technological progress, and live with it.. It could free us all from work slavery and poverty in the future

[vague oasis]

distilled sd is gonna be released and it will be game over tbh, they will need to learn how to use ai or they will become obsolete

[carmine spruce]

The music lobby is way too tight compared to the art one, theoretically making a music AI is way easier than an art one but because it's a mine field there there isn't one yet

[vague oasis]

as a musician im not afraid, cause the music industry isnt about music at all, its a whole

[wispy nest]

they're probably just worried about the future and what new skills will be needed to learn

[vague oasis]

people likes musicians not just for their music

[placid kraken]

artists are just little soy drinkers watching them become obsolete as they cry and complain is my dream 😂

[carmine spruce]

most people don't care about the person behind, only those who are passionate about music do

[split valve]

Artists that spend all day making art that is basically a copy of their favorite artists styles anyway,, like Ghibli

or whatever

[vague oasis]

they do care, those are fans

[carmine spruce]

you could argue that it still create a different style and that this is how art have been moving forward tho

[vague oasis]

and theres a lot of potential fans out there

[split valve]

musicians play an instrument because they like doing it ( a lot)... and they perform because it gives them a buzz and people enjoy it

[carmine spruce]

yeah yeah, not denying that, I just meant that fans are a minority when it comes to who consume a media

[placid kraken]

maybe we can even revive some long dead bands with ai if the record label agrees with us?

[split valve]

AI + 3D Holograms

[errant leaf]

Everything is moving forward whether people like it or not-you don't see me harassing people who use their smartphone for photography because it's not 'real photography'

[vague oasis]

i got your point, the average listener is not gonna care about us at all

[placid kraken]

i imagine many nirvana fans would be dying to hear kurt cobain sing again as well

[wispy nest]

I don't like photos because they steal your soul, or so I heard for this one crazy old person! ;P

[split valve]

taken with my cellphone

[vague oasis]

im 100% sure they will do this but in the down low, they wont say it was generated with ai

[carmine spruce]

But to be fair, there is no point in searching for validation from other people, the end goal of AI is to get rid of the idea that any form of art is something you search for, it'll just end up being tailored to perfectly fit your own taste, there will still be some sharing done but it'll be way less once we reach a point in technology can produce quality AI arts en masse for you and you'll just end up searching within what the AI create for you rather than what's available out there as it'll be more fitting for yourself

[placid kraken]

that's true but awful if the ai isn't getting any credit for the work it's doing

[carmine spruce]

(we're very far from that point tho)

[vague oasis]

the music industry is all about money, "unreleased michael jackson songs" will sell better than "ai generated mj"

[placid kraken]

true true

[wispy nest]

One thing I keep thinking about. I wonder why people seem so worried is because they have noticed a decline in commissions?

[split valve]

i think the prompts should be encrypted and hidden so that the Anti-AI artists cant shout "hey thats my style you stole it" they will be in doubt, and wondering if it really is their style or just a mix of thosuands of artists styles learnt by the AI

[carmine spruce]

bigger artists are fine but smaller artist did get less from what I've hear of the small artists group I used to follow

[placid kraken]

because they're mad something superior exists ig?

[wispy nest]

I hope it's not just hate, sounds so bad if that's what they have inside of them. :(

[placid kraken]

i have seen many ai pieces that i would consider infinitely better than digital art tbh

[carmine spruce]

I mean you can't really blame artists for being mad at it, they spend years developing the skill, look at how people are already hardcore defending this tool when it's only a few month old, so increase that by 10 fold and you get how it feel for artists right now, it's just a natural human reaction

but it's pointless to be mad anyway since the thing is here now

[placid kraken]

why do they not think of the training required to code the ai?

[wispy nest]

yeah, the same thing happens each time a new technology emerges which removes a job from someone. Sometimes that job changes, sometimes it gets redundant. But I don't think people will stop making art, they will just change a little! :3

[split valve]

if Aritsts want praise and validation they should go out on the streets and do their art, like buskers.

See if their skill is even worth anything afterall

[vague oasis]

they can definitely use their skills in their favour, sd is not good generating specific details or simple small details yet

[carmine spruce]

for sure, right now it's not fully developed so there is still place to grow, it'll get better with time but yeah there will still be artists left in the end

[placid kraken]

another weird thing they complain about is hands, but they also often complain about how bad they are at drawing hands??

hypocrisy is a weird thing

[analog adder]

not to mention the theft when a large portion are using pirated software

[placid kraken]

very true

[analog adder]

I've tried to bring this up but they start twisting my words like I'm defending Adobe, I'm not, use software you can afford its not hard

[wispy nest]

There's a saying that don't argue with fools, they'll just bring you down to their level and win with experience ;P

[analog adder]

its mind boggeling so yeah I didn't go further, its a waste of time

[wispy nest]

just look at me, I always win, but that's because I never listen to what others have to say. T-That count, right? :P

[analog adder]

lol

[wispy nest]

no, wait, that was a typo. I meant understand instead of listen!

[placid kraken]

anyway here's this piece on batman i made, sorry if it's a bit choppy i'm kinda new to this ai art thing

[wispy nest]

oh, no! isn't that a...a—a copyrighted character!?! ;P
everyone flee! :P

[errant leaf]

honestly I'm not too annoyed about people being upset because of copyright or that they'll lose commission work- that much is understandable. what really gets under my skin is people not taking the effort to look into it and fact check the garbage they're very confidently shouting to us; they seem to think AI art is basically automated photoshop or something, and every SD user is by default a degenerate art stealing tech bro

[fiery reef]

Chill evening

[quiet knot]

黙って金をとれ！

lmao

[sage kindle]

DOCCCE WHOO POGE WO+O lol

[rain gazelle]

Looking for a 3200x900 image for dual-screen wallpaper, new years eve themed.

Haven't made anything great so far.

Any takers for the challenge? 🙂

Here's one that just completed and it's awful. 😦

[gritty cedar]

try more steps and ask for it to be "a photo of : your prompt" it will be more realistic and detailed

[gritty cedar]

also pick a time of day and be specific, if you let it wander and have a city with nighttime lighting and a sky that is sunset lighting it wont appeal to the eye as much

[rain gazelle]

eh I gave up 🙂

I wanted detailed, but also flat enough to be a wallpaper.

Ultimately couldn't visualize it, so I just generated some simple stuff.

[pallid ruin]

Mmmm have you tried dividing it into 3 images? @rain gazelle and then merge them again?

[rain gazelle]

I did not. I tried various upscaling. But frankly I couldn't get anything I wanted at the lower res, so it was futile. 🙂

[pallid ruin]

I´ll try 🤗

[pallid ruin]

I divided it in 3 images and used img2img

Maybe it works better with more images at a higher resolution, but I only have 3gb vram 🥺 😇

[rain gazelle]

right on, ill scope it out

[pallid ruin]

Wait

Is it hassand1.4 + anythingv3? @wintry karma

I think have them I´ll try to merge them

Yeah but at what %?

🤷

Forget it dude, my English is not the best but I try to communicate in the best way I can.

[hard wagon]

New Double Exposure embedding for SD 2.x
https://huggingface.co/joachimsallstrom/Double-Exposure-Embedding

[gritty cedar]

very wow

[onyx whale]

I just thought it was interesting to share that the dreambot added a stock photo watermark to the image, I havent seen that before

[worn wyvern]

generated with midjourney, upscaled by esrgan

[wild seal]

awww

[dense tapir]

A new 16 vector embedding. Grab it, and read more, here - https://huggingface.co/GeneralAwareness/Ppgra

[wispy nest]

If you recognize the song let me know please. 1:30 - https://www.youtube.com/watch?v=0TB79fG8WyQ

[fervent rune]

. ,lmnliysuufpi

]

[wispy nest]

if anyone recognize the song 1:30 let me know pls https://www.youtube.com/watch?v=0TB79fG8WyQ

[wispy nest]

trying to make a Rage Comic model

[errant leaf]

that is terrifying

[wispy nest]

yeah it tends to be stupid funny or stupid scary, depending

"Rage comic stupid mario"

Y U NO santa

[errant leaf]

that one is actually suprisingly good

[wispy nest]

I used full caption so I can call on some specific meme

let's tr me gusto, I named it liking it

liking it santa

that is in the scary territory

true story Santa

[errant leaf]

imagine the horrors if we had this tech 10 years ago

[wispy nest]

imagine the world in 2 years

[cyan snow]

Yeah man loab will end us all

[split valve]

https://mpost.io/midjourney-and-dall-e-artist-styles-dump-with-examples-130-famous-ai-painting-techniques/

grab any row of text to paste into your prompt to get style

[split valve]

Garfield cartoon style example

Dreamworks style

Synthwave

[split valve]

Narrative realism

i might try keeping the same seed and do a full style reference

build a bit of a library of styles

would be great if there was a drop down menu that was customizable in the prompt settings so i could just select any style and it puts the artists in the prompt

[lucid socket]

auto's ui has that if you're running locally

[split valve]

i've not dived into any of the local stuff because of my crap gpu.

normally i upgrade every 3 yrs or so.. But Nvidia decided to go down the root of screw the average customer and make them pay. So im still waiting for a good upgrade from my GTX card

[lyric rapids]

this absolutely killed me LMAO

the prompt is "A broken android in a post-apocalyptic world"

[modern prism]

nice one what model did you use

[pale pagoda]

YOUNG MAGIC ROYALS

[neon radish]

any1 knows whats the technique to make an image like this? the base was an owl

[hasty nova]

So I'm going to try to make another model after many failed attempts (I have had one successful one) and I am going to use dreambooth I think. If I were to do captioning, how would I do that? I am going to use thefastbens's colab (https://colab.research.google.com/github/TheLastBen/fast-stable-diffusion/blob/main/fast-DreamBooth.ipynb)
And I don't think it takes in text files to my knowledge. I am thinking it's going to be a decently sized dataset and I would like to caption things but I don't know how really. My last attempt at a well captioned model turned out horribly and I just named the files with the captioning. Any tips?

[modern prism]

Nice how did you get such highquality pictures

[pale pagoda]

Upscaling, 🤭

[pale pagoda]

The ones I just posted are at 2045-2550 resolution,

[modern prism]

what upscale algo did you use

ldsr ?

I upscaled mine and it's not that great quality

the base img probably was already great

[pale pagoda]

Erm tbh I use stable diffusion and then I use midjourneys upscalers

[modern prism]

ah ok lol

[pale pagoda]

If you want I can post original and then UPSCALED

[modern prism]

yeah I'm curious to see

[pale pagoda]

Okay

[pale pagoda]

Top is 512-768 and bottom one is2048-3072

Tbh the only difference is the bottom is sharper

[modern prism]

true ty

the base image must have been quite good already

I never managed to get that high quality looking results with SD

[pale pagoda]

You can , also if you ever come across redshift diffusion that a good one aswell

[modern prism]

I'll try thanks

[wintry sundial]

amazing. game devs. concept artists and 3d modellers can all cope and seethe. (speaking as a 3d modeller myself)

[errant leaf]

Concept artists and 3d modellers should be ecstatic, lol- their work and speciality just became a lot easier/more productive

[frigid briar]

very much this.

it would be one thing if there was a limited amount of art to compete over making, but society has an infinite appetite for consumption of exactly 2 things: healthcare and art

[errant leaf]

Exactly, it just means there's more to consume, and that average quality will go up + higher quality will be more appreciated

[split valve]

1.5

[modern prism]

great one, I'm curious what did you ask in the prompt to get this ? I don't need the exact prompt just want to know what did you ask for... illuminated pillar in a cave ?

don't know how I'd go about this

[sage kindle]

You can get a lot by prompting using artist names, and because the filename for that image is 2242512506_Fantastic_Planet_Jeff_Koons_Raphael_Lacoste_Justin_Gerard_Leiji_Matsumoto_____people_character_perso.png it's almost certain that @split valve did exactly that

[split valve]

Somehow i have lost the input image that i gave it as a starting point. It only needed a 50 - 60% likeness

[modern prism]

thanks, I never use artist names

[split valve]

results in this kind of remix with that prompt

[modern prism]

maybe I should start

beautiful

[sage kindle]

same, it's an ethical line I'd prefer not to cross

[split valve]

https://mpost.io/midjourney-and-dall-e-artist-styles-dump-with-examples-130-famous-ai-painting-techniques/

go here for ideas on giving it a style

[modern prism]

thanks @split valve

and @sage kindle too ofc

[split valve]

i mean yea there is using an artists name but you could always just type in the art style

it would still have learnt it

[sage kindle]

LAION isn't captioned all that well, so artist-names often go deeper into style than art-style-names

I'd love to see more engagement on getting end-users to help add tags to images now that there's a sizeable audience interested in diffusers

[split valve]

dumb question but cant they use that image to text ai to automate the process?

[sage kindle]

sure, but a) that's essentially circular logic and b) humans can give much deeper captions

[split valve]

so it would not really learn anything new if they did that

[sage kindle]

Words like luminous or happy or such could be much stronger as prompts if humans were doing more prompting

[split valve]

it needs humans

[sage kindle]

Yep!

The risk is that captions get subjective; we'd probably need community moderation (e.g. so anti-diffusion folks can't pollute the data with nonsense) and perhaps weight the prompts tags for cases of human disagreement

There's a lot of overlap here with e.g. the sorts of things that Wikimedia does (hi, been a Wikipedia admin since 2007) so it's reasonably familiar as a problem

[split valve]

Pitty about the anti-AI folk, they just don't understand whats happening and are attacking the wrong people.

[sage kindle]

I sympathize with them, but they lose a lot of my sympathy when it's gatekeep-y or uninformed or just plain malicious

[split valve]

It could be so powerful if the dataset was captioned really well.. The concepts the AI would learn

but its so much work

[sage kindle]

hence dividing it up into a Wikipedia-like problem

[split valve]

award people with the promise of money and they will do anything.

look at the sites like turbo squid, cg trader, shutter stock, etc etc

[sage kindle]

Coming from the Wikipedia side of things, actively keeping money out of the system can be interesting if there's a good mission

[split valve]

tiny promise of money and they will work their socks of by the millions

[sage kindle]

Money has the unfortunate side-effect of complicating things and introducing people who are in it for the money over the mission

[split valve]

There is that

[sage kindle]

e.g. if money was offered for "tagging images better", we might see people doing poor-quality mass-tagging for the sake of earning more money from whatever bounty was offered; quality over quantity

[split valve]

could give them free credits to spend on things like dreams tudio

[sage kindle]

not a bad idea

I'd probably do something like a raffle with a logarithmic weighting for more contributions, e.g. 1 tag gets you one entry into a (monthly?) raffle, 10 tags gets you two, 100 three, and so on

some reward for contributing, easy entry, but diminishing returns

[split valve]

sounds good that way its not guaranteed money so less likely to be exploited

[sage kindle]

Yeah, plus you've got to expect a Pareto sort of distribution where 20% of people contribute 80% of tags and vice versa, so you want to make sure that the 80% of people feel like it's worth it to contribute even a dozen tags