AI Art: It is criminal to not post your prompt

The Something Awful Forums > Main > General Bullshit > AI Art: It is criminal to not post your prompt

Rutibex: Sep 9, 2001; by Fluffdaddy

KinkyJohn posted:

Yeap pretty much all of the stock image corps are screwed. But to be fair, gently caress them. They have hosed over photographers and artists for the longest time, thanks for paying me less than 10 cents royalties per image used

They even crowdsource their image tagging by making you meticulously tag and name everthing yourself.

Which is why it would be the sweetest twist of fate if the "crowd" sources back their tags to build a stock image model that will put them out of business

its pretty obvious from the outputs that Dall-e 2 was trained on a lot of stock photos. it has a way of framing objects and people with weird lighting that you only see in a stock photo shoot.

midjourney feels like it was trained mostly on classical art

# ? Sep 25, 2022 11:56

Adbot: ADBOT LOVES YOU

# ? May 29, 2024 18:22

Rinkles: Oct 24, 2010; What I'm getting at is...
Do you feel the same way?

And SD reproduces watermarks (I think DALLE2 might on occasion too).

# ? Sep 25, 2022 12:03

Rutibex: Sep 9, 2001; by Fluffdaddy

I've found a neat AI that takes your simple prompts and spices them up for Midjourney. its gives you a bunch of options so you can pick the most weird or interesting ones

https://huggingface.co/spaces/doevent/prompt-generator

"portal to another world" becomes
"portal to another world inspired by M�bius and Buckminster Fuller:: intricate complexity, rule of thirds, face by Artgerm:: perfect symmetrical face::2 �ar 9:16"

"atomic cowboy" becomes
"atomic cowboy, beard, large nose, happy, portrait, headshot by Robert McGinnis, Craig Mullins"

# ? Sep 25, 2022 12:08

Moongrave: Jun 19, 2004; Finally Living Rent Free

extremely haunted image

# ? Sep 25, 2022 12:24

Pitdragon: Jan 20, 2004; Just another lurker

I did "a mad max *famous person*, manga scan, artwork by kentaro miura, high resolution" and got some decent results with stable diffusion:

Richard Ayoade

Benedict Cumberbatch

Andy Samberg

Selena Gomez

Jeff Bridges

Idris Elba

Tom Hanks

Natalie Portman

Christopher Walken

It also gave me some uh, more interesting results with certain celebrities:

Ron Perlman

Rosie O'Donnell

Danny Devito

# ? Sep 25, 2022 12:48

Rutibex: Sep 9, 2001; by Fluffdaddy

Pitdragon posted:

Rosie O'Donnell

mutie!

# ? Sep 25, 2022 13:54

lunar detritus: May 6, 2009

I was hit with this captcha and the images look very AI generated

# ? Sep 25, 2022 14:53

mcbexx: Jul 4, 2004; British dentistry is
not on trial here!

Am I understanding this correctly, k_euler(_a) is the only of the current sampling methods which will not be able to reproduce a previous prompt when providing the exact seed number?

# ? Sep 25, 2022 15:45

WhiteHowler: Apr 3, 2001; I'M HUGE!

mcbexx posted:

Am I understanding this correctly, k_euler(_a) is the only of the current sampling methods which will not be able to reproduce a previous prompt when providing the exact seed number?

Euler-a will exactly reproduce a previous prompt's output as long as you keep all input values the same, including the number of sampling steps. I don't exactly understand the logic, but I believe the "-a" variants use previous steps' output to inform the next step, so modifying the number of steps by even one can produce a drastically different result.

Other sampling modes tend to converge on an extremely similar result (both with themselves and with each other) above a certain threshold of sampling steps. This is why if you use the same seed for DDIM, the output at 40 steps should look almost identical to the output at 50 steps. Which should look very similar to the output from LMS at 40/50 steps.

If you do the same with Euler-a or DPM2-a, you'll usually get drastically different results at 40 and 50 steps.

There are several examples of this on Reddit. This post is good -- it shows the same prompt/seed/CFG values across the different sampling methods, and their output at different step counts. Note how Euler-a and DPM2-a have completely different results, both from the other methods and from themselves at different step counts.

WhiteHowler fucked around with this message at 16:19 on Sep 25, 2022

# ? Sep 25, 2022 15:59

Sedgr: Sep 16, 2007; Neat!

Liked the way these turned out. Two subjects was a surprise its usually so focused on one. Reflection on the third from last is pretty neat as well.

# ? Sep 25, 2022 16:33

EVIL Gibson: Mar 23, 2001; Internet of Things is just someone else's computer that people can't help attaching cameras and door locks to!; Switchblade Switcharoo

Objective Action posted:

The trick is that the Photo GANs aren't temporally stable so you get shimmer and artifacting from that. You really have to use a recurrent model built for sequential data or video. The other catch there is that most models right now are built by starting with good video and crapping it up so they have ground truth.

This is why Topaz Video Enhance works better than its normal image upscale app because the model it uses trains with real frames from videos and learns what normal video looks like.

Like it knows VHS will not give you the crispest HD because a lot of stuff is lost for the format.

Photogans will try its hardest to make everything look super crisp which will give you the shimmering edges and very sharpest clouds while everything looks like mud.

# ? Sep 25, 2022 16:49

Rutibex: Sep 9, 2001; by Fluffdaddy

lunar detritus posted:

I was hit with this captcha and the images look very AI generated

thank you for your training tags

# ? Sep 25, 2022 17:17

Comfy Fleece Sweater: Apr 2, 2013; You see, but you do not observe.

KwegiboHB posted:

This should be right up your alley then.
Link for large picture. https://imgur.com/a/yIUnxEl

I can't wait to feed this thing every single Heavy Metal comic.

Awesome

I posted this earlier - For Apple M1 Mac owners:
DiffusionBee https://diffusionbee.com/ is by far the easiest 1-click-install I've found (yet). If you want to take your first AI baby-steps , don't know anything about programming or python etc, and don't have a PC with a good GPU, DiffusionBee is for you. Will it continue to be developed or is it just a proof of concept? Who knows! It seems very polished and nice for a pre-alpha type of app. Made by some former Microsoftie (https://twitter.com/divamgupta)

But also, I just spotted InvokeAI, which looks way more full-featured, but requires more fiddling around. It has "only" 2.5k stars but development seems to be moving forward very quickly.
(I've not tested this one, so don't blame me if it sends all your credit card numbers to a nigerian prince. seems legit tho)

https://github.com/invoke-ai/InvokeAI/tree/development#readme

So come on apple dorks, get in on the action

Further edit:
A comparison between samplers: https://www.reddit.com/r/StableDiffusion/comments/xmwcrx/a_comparison_between_8_samplers_for_5_different/

TLDR: K_DPM_2_A and K_EULER_A are the "creativity/variability" samplers. I've been using EulerA and it's basically the weirdness knob. The others are not useless tho, you can get a result much faster/in less steps, depending on your goal.

quote:

Remember
Results converge as steps (-s) are increased (except for K_DPM_2_A and K_EULER_A). Often at ≥ -s100, but may require ≥ -s700).
Producing a batch of candidate images at low (-s8 to -s30) step counts can save you hours of computation.
K_HEUN and K_DPM_2 converge in less steps (but are slower).
K_DPM_2_A and K_EULER_A incorporate a lot of creativity/variability.

Suggestions
For most use cases, K_LMS, K_HEUN and K_DPM_2 are the best choices (the latter 2 run 0.5x as quick, but tend to converge 2x as quick as K_LMS). At very low steps (≤ -s8), K_HEUN and K_DPM_2 are not recommended. Use K_LMS instead.
For variability, use K_EULER_A (runs 2x as quick as K_DPM_2_A).

Comfy Fleece Sweater fucked around with this message at 18:28 on Sep 25, 2022

# ? Sep 25, 2022 17:54

an actual frog: Mar 1, 2007; HEH, HEH, HEH!

Comfy Fleece Sweater posted:

I posted this earlier - For Apple M1 Mac owners:
DiffusionBee https://diffusionbee.com/ is by far the easiest 1-click-install I've found (yet). If you want to take your first AI baby-steps , don't know anything about programming or python etc, and don't have a PC with a good GPU, DiffusionBee is for you. Will it continue to be developed or is it just a proof of concept? Who knows! It seems very polished and nice for a pre-alpha type of app. Made by some former Microsoftie (https://twitter.com/divamgupta)

Yeah, if you have a mac and are curious to try AI image generation with zero hassle don't sleep on this. It's very basic compared to the automatic1111 suite, but you literally just install it like any other mac app, launch and start typing.

# ? Sep 25, 2022 18:47

Longpig Bard: Dec 29, 2004

Objective Action posted:

The trick is that the Photo GANs aren't temporally stable so you get shimmer and artifacting from that. You really have to use a recurrent model built for sequential data or video. The other catch there is that most models right now are built by starting with good video and crapping it up so they have ground truth.

The problem being they, well, suck at crapping up video realistically. They almost universally just down-rez it and maybe, if they are feeling frisky, blur it a little bit. Real SD video is usually covered with all sorts of compression artifacts so even state of the art video upscalers usually eat poo poo if you don't do some work ahead of time to clean up the source as much as you can.

I took your video and ran this an ESRGAN 1x_NMKD_dejpeg_Jaywreck3-Lite_320k pass (cleans up blocking and JPEG like artifacts, one of the better general purpose one of these I've found even for non-JPEG). Ran PaddleGAN's implementation of PPMS-VSR on the now much cleaner input to do the super-resolution and frame blending. Ran it through a downscale in Handbrake back down to 720p to get it back down to web postable size. Then went over to FFMPEG and muxed the original audio back onto the new video file. Finally I ran one last pass and remux with ESRGAN using 1x_Sayajin_DeJPEG_300k_G to color correct it.

Even after all that the results are better but still only mediocre. Most noticeably on the shots where lots of small text is visible. Even in the original video those fine details are destroyed and the AI just doesn't have enough info to properly restore it so it still smudgy.

https://i.imgur.com/Otu2lmk.mp4

Imgurs compression makes it look a little worse but its not much better than that raw unfortunately. Arguably at this resolution its kind of a moot point and just some back belocking/bebanding/CAS sharpening in Handbrake would give as good or better results.

Edit: Also Imgur seems to have ignored the aspect ratio settings on that upload but I'm too lazy to fix it rn so you get stretched rear end widescreen I guess.

Edit2: OK I lied it was bothering me so I fixed it.

Tried it in Topaz Video Enhance AI - Using the "Artemis Low Quality" model. It increased the sharpness overall.
https://i.imgur.com/DcXus3B.mp4

# ? Sep 25, 2022 19:41

Comfy Fleece Sweater: Apr 2, 2013; You see, but you do not observe.

haaay guuuuys

sorry I forgot the exact prompt, but it was a leftover from "goony wizard"

"Goony wiz" gave me this guy, of course

# ? Sep 25, 2022 19:58

Sedgr: Sep 16, 2007; Neat!

They're ~~good~~ weird Drags Brent.

Maybe thats enough dragons for the moment.

# ? Sep 25, 2022 20:03

Chronojam: Feb 20, 2006; This is me on vacation in Amsterdam
Never be afraid of being yourself!

Rutibex posted:

its pretty obvious from the outputs that Dall-e 2 was trained on a lot of stock photos. it has a way of framing objects and people with weird lighting that you only see in a stock photo shoot.

midjourney feels like it was trained mostly on classical art

You can get the lighting and framing however you want, though. If you want studio lighting, ask. If you want lighting that's filtering through cracks in the ceiling with dust motes in the air on a cold day, ask for that instead.

# ? Sep 25, 2022 21:08

Sedgr: Sep 16, 2007; Neat!

It comes up with some real cool stuff sometimes.

# ? Sep 25, 2022 21:44

KwegiboHB: Feb 2, 2004; nonconformist art brut
Negative prompt: amenable, compliant, docile, law-abiding, lawful, legal, legitimate, obedient, orderly, submissive, tractable
Steps: 32, Sampler: DPM++ 2M Karras, CFG scale: 11, Seed: 520244594, Size: 512x512, Model hash: 99fd5c4b6f, Model: seekArtMEGA_mega20

KinkyJohn posted:

I guess what I'm asking is, is there a single word for deformed hands in any language?

Put "Fleshpile" in your negative prompts.

Decidedly NSFW link: :gonk:

https://haveibeentrained.com/?search_text=fleshpile :nms:

https://haveibeentrained.com/ is a pretty good site to find out that "maleficent" the adjective is drowned out by "Maleficent" the antangonist.

# ? Sep 25, 2022 21:45

Rutibex: Sep 9, 2001; by Fluffdaddy

KwegiboHB posted:

Put "Fleshpile" in your negative prompts.

Decidedly NSFW link: https://haveibeentrained.com/?search_text=fleshpile

https://haveibeentrained.com/ is a pretty good site to find out that "maleficent" the adjective is drowned out by "Maleficent" the antangonist.

lol wtf why are they putting a bunch of sicko deviant art into the AI training. don't search for "simpsons"

# ? Sep 25, 2022 21:57

EVIL Gibson: Mar 23, 2001; Internet of Things is just someone else's computer that people can't help attaching cameras and door locks to!; Switchblade Switcharoo

unzin posted:

Tried it in Topaz Video Enhance AI - Using the "Artemis Low Quality" model. It increased the sharpness overall.
https://i.imgur.com/DcXus3B.mp4

Do the four up comparison and play with all of the settings including "grain".

For some reason, putting more/less grain into VHS source footage helps a lot.

Just try every model they have and even the ones that don't make sense like the one for buildings.

quote:

Ron Perlman

Rosie O'Donnell

Danny Devito

These three make it fit Berserk even better.

Don't know if you are aware, but there a lot of demons trying to fit into the human form with various degrees of success.

You first posted the pretty Griffiths, and then you have Irvine level.

Also Danny Devito looks like it was pulling references from his role in the Spawn movie

EVIL Gibson fucked around with this message at 22:46 on Sep 25, 2022

# ? Sep 25, 2022 22:36

Snowy: Oct 6, 2010; A man whose blood
Is very snow-broth;
One who never feels
The wanton stings and
Motions of the sense

KwegiboHB posted:

https://haveibeentrained.com/ is a pretty good site to find out that "maleficent" the adjective is drowned out by "Maleficent" the antangonist.

That explains why the graffiti looks so bad

# ? Sep 25, 2022 22:55

Sedgr: Sep 16, 2007; Neat!

More machine city at sunset... I like that the second one has the SUN inside the city. I prefer to think this machine city has harnessed fusion power.

# ? Sep 25, 2022 22:59

AARD VARKMAN: May 17, 1993

a stone monument in the shape of a fat arkansas farmer by bruce brenneise, brad kunkle, and peter mohrbacher, intricate, 3d render, neosurrealism. digital concept art, pixel art, rendered in octane, trending on cgsociety, trending on artstation, extreme detail

negative (thanks to whoever posted this so i could steal it :ssh:

)
(((deformed))), [blurry], bad anatomy, disfigured, poorly drawn face, mutation, mutated, (extra_limb), (ugly), (poorly drawn hands), messy drawing, penis, nose, eyes, lips, eyelashes, text

# ? Sep 25, 2022 23:03

Tunicate: May 15, 2012

Snowy posted:

That explains why the graffiti looks so bad

hmm, negative weighting CAPTCHA seems to improve my images noticeably

# ? Sep 25, 2022 23:09

Moongrave: Jun 19, 2004; Finally Living Rent Free

neat

# ? Sep 25, 2022 23:09

Snowy: Oct 6, 2010; A man whose blood
Is very snow-broth;
One who never feels
The wanton stings and
Motions of the sense

AARD VARKMAN posted:

negative (thanks to whoever posted this so i could steal it )
(((deformed))), [blurry], bad anatomy, disfigured, poorly drawn face, mutation, mutated, (extra_limb), (ugly), (poorly drawn hands), messy drawing, penis, nose, eyes, lips, eyelashes, text

I may have missed this but is there really a need to include penis?

I haven�t encountered any unwanted penises :wink:

# ? Sep 25, 2022 23:09

Rahu: Feb 14, 2009; let me just check my figures real quick here; Grimey Drawer

Rutibex posted:

lol wtf why are they putting a bunch of sicko deviant art into the AI training. don't search for "simpsons"

The training set includes 5.8 billion random images from the internet. Which honestly seems low, there have to be way more pictures on the internet than that!

# ? Sep 25, 2022 23:28

mobby_6kl: Aug 9, 2009; by Fluffdaddy

Rahu posted:

The training set includes 5.8 billion random images from the internet. Which honestly seems low, there have to be way more pictures on the internet than that!

And you'd certainly expect to see way more penises in the results

# ? Sep 25, 2022 23:30

Moongrave: Jun 19, 2004; Finally Living Rent Free

Snowy posted:

I may have missed this but is there really a need to include penis?

I haven�t encountered any unwanted penises

When it happens you�ll wish you�d listened

# ? Sep 25, 2022 23:52

TheWorldsaStage: Sep 10, 2020

Anyone have an issue with the Automatic repo inpainter in that it drifts a little above the mouse, so anything at the bottom of a pic is impossible to paint?

# ? Sep 25, 2022 23:54

Objective Action: Jun 10, 2007

TheWorldsaStage posted:

Anyone have an issue with the Automatic repo inpainter in that it drifts a little above the mouse, so anything at the bottom of a pic is impossible to paint?

Yeah, you can wiggle it up there with enough patience but its buggy as gently caress at the edges and tends to just eat the mask sometimes when you switch tabs. I've switched to using their upload mask function and just doing the mask in paint. Extra steps but at least I don't get a mask eaten anymore.

# ? Sep 26, 2022 00:07

Rutibex: Sep 9, 2001; by Fluffdaddy

"the logo of the cows guild"

# ? Sep 26, 2022 00:09

surc: Aug 17, 2004

I'm just starting to get in and play around with what works and doesn't writing a prompt for stable-diffusion, but dang that concepts-library is a cool resource.

"colorful 2d illustration spaceship trailer parked in the desert under a moon with a bar set up out front and people drinking in the style of <moebius>" (using the moebius style from https://huggingface.co/sd-concepts-library)