Register a SA Forums Account here!
JOINING THE SA FORUMS WILL REMOVE THIS BIG AD, THE ANNOYING UNDERLINED ADS, AND STUPID INTERSTITIAL ADS!!!

You can: log in, read the tech support FAQ, or request your lost password. This dumb message (and those ads) will appear on every screen until you register! Get rid of this crap by registering your own SA Forums Account and joining roughly 150,000 Goons, for the one-time price of $9.95! We charge money because it costs us money per month for bills, and since we don't believe in showing ads to our users, we try to make the money back through forum registrations.
 
  • Post
  • Reply
Sedgr
Sep 16, 2007

Neat!

A little sexy, a little hexy, a little rexy.






Adbot
ADBOT LOVES YOU

Moongrave
Jun 19, 2004

Finally Living Rent Free
gonna work on my spookum prompts now then....





gonna be a fun month!

surc
Aug 17, 2004



an actual frog
Mar 1, 2007


HEH, HEH, HEH!
I was trying to generate sick robots with cool, moody backgrounds of like sparks 'n' poo poo.

SD gave me this :krad: hotrod skull instead


And this steampunk robo-satyr,. The filament lamps are a neat touch




Also, shoutout to the birb-style concept

Rutibex
Sep 9, 2001

by Fluffdaddy

mcbexx posted:

I feel "artists" who huff and puff over it don't really understand what it has to offer, especially to people who are not gifted with the manual skillset, but now have a tool to express their imagination.

I think the artists who complain most are the ones who were "cheating" to begin with. tracing subjects, using fancy art filters and expensive image tools to make nice stuff really easily.

the barrier to entry on lovely mass produced e-art has bottomed out

pixaal
Jan 8, 2004

All ice cream is now for all beings, no matter how many legs.


Sometimes I like to see where it's going and take a journey in img2img I liked this skywhale playing in the clouds and upscaled her.

Rutibex
Sep 9, 2001

by Fluffdaddy
randomly generated magic cards! each portion is generated by a differernt random process. First I make a spell name with Seventh Sanctum

https://www.seventhsanctum.com/generate.php?Genname=spell

Then I feed the spell name into midjourney to make an art. The I pick the best art for a card midjourney spits out and combine it with BNF magic card generator. Then I turn it into a magic card using magic set editor. Pretty fast process I could easily make a whole set in an afternoon :v:

https://www.toothycat.net/wiki/bnf.pl?page=AlexChurchill/MagicCardGenerator
https://magicseteditor.sourceforge.net/








Elotana
Dec 12, 2003

and i'm putting it all on the goddamn expense account
Those are all awesome except for the one with the despised Default Stylize Girl

Prolonged Panorama
Dec 21, 2007
Holy hookrat Sally smoking crack in the alley!



mcbexx posted:

I feel "artists" who huff and puff over it don't really understand what it has to offer, especially to people who are not gifted with the manual skillset, but now have a tool to express their imagination.

The "manual skillset" is there for anyone that practices enough, though. Regardless of medium. It's not a "gift" any more than being a good cook or a competent programmer is. You just have to work at it.

And I wonder whose imagination is being expressed, exactly. The images that come out of these AIs are all almost instantly recognizable as such, despite the huge range in subject matter and surface level "style." Even when it's very skilled digital artists employing them, and they tend to use them best. There's some easy forensic clues like faces being very samey (especially young female faces) and hands being off, or an overabundance of high frequency detail, or the surreal melty-blendy way things tend to flow together. Those will all get fixed in time I'm sure, and "real" artists can simply fix those things manually anyway.

But there's a deeper commonality too, something hard to articulate. A lack of focus, beyond a singular subject? A lack of "story" that we're used to seeing in pictures of such surface-level complexity (which usually arises from multiple, interacting/connected focal points)? Maybe "intentionless" is a good word for it. And it's not surprising, since these are just random noise that's been evolved toward a grab bag of key words. The output can be cool, complex, organic, surprising, but in very singular sort of way. There's a common texture or feeling to all of it, a conceptual mushiness, underneath the maximalism. It all looks very "finished," very polished, but some fundamentals are lacking or absent. Enough that it's always a little jarring to see, because a human artist can't get that good at the details without being even better at the more basic stuff. A sort of uncanny valley for art, maybe.

They're fun to play with and will devour the low end of the market. A great tool and a new medium. Instant feedback from an alien collaborator. Limitless inspiration. Endless bland goop. A real artist's best friend. A real artist's worst enemy. Beneath the notice of a real artist; a category error. I don't know. But I don't blame anyone that feels threatened/insulted/excited/inspired.

Rutibex
Sep 9, 2001

by Fluffdaddy

Elotana posted:

Those are all awesome except for the one with the despised Default Stylize Girl

thanks! the random creature ones are turning out even better. Midjourney is better at weird creatures than abstract concepts
https://www.seventhsanctum.com/generate.php?Genname=jokemonster




K3nT3n
Oct 8, 2015
If you see me posting my own tweets, please report me so I will be probated for a month.
https://twitter.com/KenWolfTran/status/1574956949233287168?s=20&t=xSrW-NktUj6Qb1Mu36v7Ag

This is part of my contribution to A.I. kind...

Big Scary Owl
Oct 1, 2014

by Fluffdaddy
"the end of the human race"




"brazil in the post apocalypse, by Hieronymus Bosch"


Made using Midjourney

lunar detritus
May 6, 2009


Prolonged Panorama posted:

. A sort of uncanny valley for art, maybe.

Until someone makes a blind study comparing human art and AI-generated art, both types of sample carefully selected instead of simply taking whatever first comes up after writing a prompt, this just feels like coping.

Bottom Liner
Feb 15, 2006


a specific vein of lasagna

lunar detritus posted:

Until someone makes a blind study comparing human art and AI-generated art, both types of sample carefully selected instead of simply taking whatever first comes up after writing a prompt, this just feels like coping.

the stuff shared here is mostly exactly what they said though

TheWorldsaStage
Sep 10, 2020

Prolonged Panorama posted:

The "manual skillset" is there for anyone that practices enough, though. Regardless of medium. It's not a "gift" any more than being a good cook or a competent programmer is. You just have to work at it.

Few things here. I could never be a programmer. I am math and code stupid. It is not something I am capable of I am in my 30s and I can admit this about myself. The analogy doesn't exactly work.

Time and money are two things in very short supply for many people. Affording the years time to practice and buy materials may not be feasible to a lot of people. Being able to use ai art to express themselves shouldn't be looked down on, or viewed lesser as.

Also to your point of basically calling ai art mcdonalds for art, you can kind of see similarities in some things but certainly not all. When you bring in gimp and photoshop into it you can't distinguish classical digital art to ai art. That's not to mention how young the tech is and how quick it's growing.

My question to you, is I get an outline and basis for what I want to do and I fine tune things for hours from color, lighting, composition and a myriad of things. Do you not see that as creative or artistic?

E: I haven't been able to do photography in years, got too expensive to travel. I really like having ai art as an outlet and I guess I just don't see it as the effortless blob of manufacture you hint at.

TheWorldsaStage fucked around with this message at 05:22 on Sep 28, 2022

Mischievous Mink
May 29, 2012








Necronomicon as a prompt adjective is fun.

Big Scary Owl
Oct 1, 2014

by Fluffdaddy
"depression", Midjourney:



This really looks authentic imo, and it also feels strangely familiar too.

Mozi
Apr 4, 2004

Forms change so fast
Time is moving past
Memory is smoke
Gonna get wider when I die
Nap Ghost
chris changle celebrating the holiday he is famous for, euphoric portrait --testp


Frob Grooks official portrait, 4k high quality photograph --testp


Same prompt


Frob Grooks official portrait, by rembrandt and raymond swanland, volumetric and perfect lighting, 4k, 8k, hd --testp


Geeble Phobox official portrait, by rembrandt and raymond swanland, volumetric and perfect lighting, 4k, 8k, hd --testp


Gub Chibbibibi official portrait, by rembrandt and raymond swanland, volumetric and perfect lighting, 4k, 8k, hd --testp


Sharkasaurus official portrait, by rembrandt and raymond swanland, volumetric and perfect lighting, 4k, 8k, hd --testp --upbeta

Prolonged Panorama
Dec 21, 2007
Holy hookrat Sally smoking crack in the alley!



TheWorldsaStage posted:

Few things here. I could never be a programmer. I am math and code stupid. It is not something I am capable of I am in my 30s and I can admit this about myself. The analogy doesn't exactly work.

Time and money are two things in very short supply for many people. Affording the years time to practice and buy materials may not be feasible to a lot of people. Being able to use ai art to express themselves shouldn't be looked down on, or viewed lesser as.

Also to your point of basically calling ai art mcdonalds for art, you can kind of see similarities in some things but certainly not all. When you bring in gimp and photoshop into it you can't distinguish classical digital art to ai art. That's not to mention how young the tech is and how quick it's growing.

My question to you, is I get an outline and basis for what I want to do and I fine tune things for hours from color, lighting, composition and a myriad of things. Do you not see that as creative or artistic?

E: I haven't been able to do photography in years, got too expensive to travel. I really like having ai art as an outlet and I guess I just don't see it as the effortless blob of manufacture you hint at.

I should have said [insert skill here] - sure, everyone has different tendencies and inclinations, but most things are learnable, and the skill ceiling is pretty high. People aren't born knowing how to code or cook or draw or paint or play piano or dress well or lead a sales call or ride a bike or drive a car or play Call of Duty or touch type... or write prompts for image AIs! It's all just practice.

I wasn't saying I look down on AI art, just that unless it's been modified pretty heavily (and skillfully) it tends to be quickly identifiable. So while there is a big element of "people expressing themselves" (which is good!!!), the tools themselves are imposing a strong sort of look/feel/style/vibe as well - it isn't pure expression, or even really anywhere close to it. All tools suffer this in their infancy, we'll see how AI tools grow, in tandem with users getting more sophisticated and skilled. It's an interesting time.

Of course taking output from an AI and further refining it toward your vision is creative/artistic. I'm doing it myself! It's exciting/inspiring/threatening, as a digital artist at work and at home. I don't have a verdict or judgement on this stuff. If I did it wouldn't matter.

Wheany
Mar 17, 2006

Spinyahahahahahahahahahahahaha!

Doctor Rope

Yes, YES

Jonny Nox
Apr 26, 2008




oh, poo poo. bagpipes guy is going to give us Gub Chibbibibi bagpipes. Genuinely can't wait.


edit: also, Gub Chibbibibi is now the thread mascot.

mobby_6kl
Aug 9, 2009

by Fluffdaddy
Interesting article on using SD as an image compression mechanism. At a glance it seems like a pretty obvious extension of text inversion /embeddings tbh but they did a good job developing some methods and testing and evaluating everything.



https://arstechnica.com/information-technology/2022/09/better-than-jpeg-researcher-discovers-that-stable-diffusion-can-compress-images/

surc
Aug 17, 2004

Prolonged Panorama posted:

I should have said [insert skill here] - sure, everyone has different tendencies and inclinations, but most things are learnable, and the skill ceiling is pretty high. People aren't born knowing how to code or cook or draw or paint or play piano or dress well or lead a sales call or ride a bike or drive a car or play Call of Duty or touch type... or write prompts for image AIs! It's all just practice.

I wasn't saying I look down on AI art, just that unless it's been modified pretty heavily (and skillfully) it tends to be quickly identifiable. So while there is a big element of "people expressing themselves" (which is good!!!), the tools themselves are imposing a strong sort of look/feel/style/vibe as well - it isn't pure expression, or even really anywhere close to it. All tools suffer this in their infancy, we'll see how AI tools grow, in tandem with users getting more sophisticated and skilled. It's an interesting time.

Of course taking output from an AI and further refining it toward your vision is creative/artistic. I'm doing it myself! It's exciting/inspiring/threatening, as a digital artist at work and at home. I don't have a verdict or judgement on this stuff. If I did it wouldn't matter.

I mean you could literally say that "the tools used to do [art] impose a strong sort of look/feel/style/vibe" and that would be correct, it's just that most of them have been around long enough that it's not thought about, and people look for the differences in the work rather than focusing on "oh yeah this was DEFINITELY painted with oil paint" or w/e (e: and also generally it was made by somebody who put time into that skill if you're looking at it so it's not just globs of paint). I have absolutely seen art that doesn't look like other AI art at all and has a focus and all the things you're mentioning, and when they've shared prompts, their prompts are like 15 times as long as mine and have massive amounts of them playing with emphasis on elements and specific synonyms for a concept to do that. Most people are just still doing the equivalent of stick-figures, but with ai art. The artistic skill used is the prompt writing.

I know you're not really disagreeing with any of that about AI-gen-art, just seems weird in where the fault is being assigned. I think the issues with ai-art variety atm are the prompt-writers not being skilled at it, more than it not being able to deliver

surc fucked around with this message at 15:16 on Sep 28, 2022

Snowy
Oct 6, 2010

A man whose blood
Is very snow-broth;
One who never feels
The wanton stings and
Motions of the sense



Jonny Nox posted:

oh, poo poo. bagpipes guy is going to give us Gub Chibbibibi bagpipes. Genuinely can't wait.


edit: also, Gub Chibbibibi is now the thread mascot.

Wellll… I tried my best. All my Gubs came out as portraits of him later in life. Then I tried “as a child”, “young Gub Chibbibibi” “as a young boy” and got so many unsafe content errors I was afraid I was about to get in trouble. I really need to install some of this stuff rather than just using Huggingface.

















That last one is the one result I got from trying to make him young :v:

TheWorldsaStage
Sep 10, 2020

A Scottish Leprechaun!

E: I didn't know there was a Scottish equivalent to a Leprechaun and now I'm going down a rabbit hole

Snowy
Oct 6, 2010

A man whose blood
Is very snow-broth;
One who never feels
The wanton stings and
Motions of the sense



TheWorldsaStage posted:

A Scottish Leprechaun!

E: I didn't know there was a Scottish equivalent to a Leprechaun and now I'm going down a rabbit hole

I can’t wait :)

Also I tried NightCafe starting with the original Gub image and got this



That’s just my first try but I really need to get out of bed at some point today

Comfy Fleece Sweater
Apr 2, 2013

You see, but you do not observe.

Mozi posted:

Sharkasaurus official portrait, by rembrandt and raymond swanland, volumetric and perfect lighting, 4k, 8k, hd --testp --upbeta


Oh I just did a Shark movie too!



exclusive image still from the movie

Sedgr
Sep 16, 2007

Neat!

Demon trouble?








Y'all need...



Chello De Don
Nov 12, 2006

and now i do
When I'm writing my prompt to Midjourney, does applying a weight to a phrase like this:

isometric map::1.75 only apply the weight modifier to the word "map", or will it take into account the full phrase "isometric map"? If not, can applying parentheses around the phrase fix it, like so:

((isometric map))::1.75, <rest of the prompt>

Tunicate
May 15, 2012

Chello De Don posted:

When I'm writing my prompt to Midjourney, does applying a weight to a phrase like this:

isometric map::1.75 only apply the weight modifier to the word "map", or will it take into account the full phrase "isometric map"? If not, can applying parentheses around the phrase fix it, like so:

((isometric map))::1.75, <rest of the prompt>

applies to everything before it, also segments it as a new prompt

so

firefighter ::2
is different than

fire::2 fighter::2

Chello De Don
Nov 12, 2006

and now i do

Tunicate posted:

applies to everything before it, also segments it as a new prompt

so

firefighter ::2
is different than

fire::2 fighter::2

Ah okay that makes sense. Thanks Tunicate!

mcbexx
Jul 4, 2004

British dentistry is
not on trial here!



Training your own models in the cloud for as little as $0.30 per hour (that's about one prompt "term", depending on complexity).

Get yourself, your mom, dad, girlfriend/boyfriend, dog, archnemesis into your images.

Mind: boggled.

https://www.youtube.com/watch?v=7m__xadX0z0

Elotana
Dec 12, 2003

and i'm putting it all on the goddamn expense account
People asked about this at Midjourney office hours today but apparently the training and storage doesn't scale well across millions of users all wanting their own banks of S*, so they're instead focusing on improving image prompts (which are basically an ad hoc version of this)

I'm interested to see what those can do once they're not squashed to 256x256

Snowy
Oct 6, 2010

A man whose blood
Is very snow-broth;
One who never feels
The wanton stings and
Motions of the sense



Doggles posted:

"a corgi wearing an astronaut suit, dragoncon, 35mm, f/2.8, flash fired, full length"


:kimchi:

Bagpiping corgi wearing an astronaut suit, dragoncon, 35mm, f/2.8, flash fired, full length





Uhhhhhh

Siguy
Sep 15, 2010

10.0 10.0 10.0 10.0 10.0
I finally got Dreambooth running on a rented server and was able to train a new model on pictures of my daughter (sorry, not sharing pictures of a 5-year-old on Something Awful). It's kind of a horrible, clunky, error-laden process and the final results were only at the "mostly looks like her" level rather than the "perfect resemblance" level but I still really enjoyed it. I'd gladly destroy the earth with CO2 emissions by using rented server after rented server to pump out millions of pictures of my kids as space vikings and whatnot.

Elotana
Dec 12, 2003

and i'm putting it all on the goddamn expense account
Yeah I pretty much bought private mode on Midjourney because my nephews and nieces wanted themselves as characters and I didn't want their headshot image prompts public

lunar detritus
May 6, 2009


Siguy posted:

I finally got Dreambooth running on a rented server and was able to train a new model on pictures of my daughter (sorry, not sharing pictures of a 5-year-old on Something Awful). It's kind of a horrible, clunky, error-laden process and the final results were only at the "mostly looks like her" level rather than the "perfect resemblance" level but I still really enjoyed it. I'd gladly destroy the earth with CO2 emissions by using rented server after rented server to pump out millions of pictures of my kids as space vikings and whatnot.

What server did you use? I'm looking at lambda labs, runpod and vast.ai but I'm not sure which is the best value/price choice

mobby_6kl
Aug 9, 2009

by Fluffdaddy

Elotana posted:

Yeah I pretty much bought private mode on Midjourney because my nephews and nieces wanted themselves as characters and I didn't want their headshot image prompts public
So how does the training work with Midjourney then? You just pay them to train it on some images for you?

I tried getting SD trained, the process was pretty straightforward but I ran out of memory on my machine. Could probably use AWS or even my work cloud but didn't get to it. Haven't touched Midjourney since SD became available.

WhiteHowler
Apr 3, 2001

I'M HUGE!

Siguy posted:

I finally got Dreambooth running on a rented server and was able to train a new model on pictures of my daughter (sorry, not sharing pictures of a 5-year-old on Something Awful). It's kind of a horrible, clunky, error-laden process and the final results were only at the "mostly looks like her" level rather than the "perfect resemblance" level but I still really enjoyed it. I'd gladly destroy the earth with CO2 emissions by using rented server after rented server to pump out millions of pictures of my kids as space vikings and whatnot.

How many images did you train it with? Some people have had success with fewer than 20 training images, but the recommendation I read said 150-200 images with several different angles and contexts gives the best results for training it on a person's likeness.

WhiteHowler fucked around with this message at 23:54 on Sep 28, 2022

Adbot
ADBOT LOVES YOU

Tunicate
May 15, 2012

The corridor crew guys did a video with that

  • 1
  • 2
  • 3
  • 4
  • 5
  • Post
  • Reply