New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Any of you messing around with AI images?
It started with me just wanting to make some graphics for our website. Nothing crazy, just some interior datacenter stuff, servers, generic imagery that a tech company may have on their site to replace the older AI images we're using now.
A lot of AI art out there but it's mostly weeb fantasy dumb stuff.
I wanted to create some storyboards and recreate some styles and aesthetics I thoroughly enjoy instead. May have gotten in a bit deep...
The following are from an "Occupational Portraits" series with a sort of mid-70s / Wes Anderson style vibe to them:
But I've been experimenting a bit with some horror / obscure stuff too:
And this:
Comments
With what GPU you render this? Or it's on cloud
Regards
Midjourney is great, I've been playing around with it this evening. So many pictures that can be used for commercial and personal purposes. Costs $10 to start with.> @Calin said:
MidjourneyXL via Clipdrop, but also "uncropping" the images to generate the aspect ratio I want which produces some interesting results since what fills in the empty places within that frame size is unprompted and based on the original image.
Yeah I've done a lot with Midjourney. I use it a lot for LEB article art.
We did a couple articles on it:
How to Use the Missing Midjourney API: Five Ways to Automate Midjourney with Code, Robots, and Even Microsoft Excel
Midjourney 5.1 is Here! What's New PLUS 20 Unique Artist Style Prompts to Make Your Creations Pop!
I was using the free tier while that was possible, but it's not available anymore.
As for the paid version, the $10 plan gives next to nothing in GPU time for the month, I'd spend that in 3-5 days. And the next one which also has unlimited slower requests is too expensive at $30.
https://dezgo.com
Midjourney is awesome.
All of those pictures have a lot of bokeh, did you add that into your prompt?
Edit: Probably my favorite part about Midjourney is that you get full rights.
I didn't specify DoF or bokeh, just 'portrait', which typically would result in a high depth of field / bokeh.
Asked a buddy to send me an idea. He said only, 'enlightenment'. So, this is my idea of what 'enlightenment' may be.
I imagine these tools are not that great for targeting specific copyrighted works or IP? I would love, for example, an image of SpongeBob vs Spider-Man in a Rambo cinematic-esque boxing scene.
Hold my beer.
My attempt(s) 😂
^ LSD in prompt
^ magic mushrooms in prompt
EDIT: OH you said Spongebob vs Spider-Man... Not Rambo. One sec.
@CyberneticTitan here you go:
@SirFoxy what are your prompts?
This is my new take on 'enlightenment':
Prompts are: ((double exposure)) (zoom blur) monk meditating, (wes anderson), (LSD), (laser acid etched), ganesh, abstract, geometric fractal, inner-eye, awakening, astral plane, smoke filled room
My prompts are a bit all over the place. It's less Wes Anderson than the earlier renditions made. Some of those are just general keywords and I'm not certain how they impact it. But stating things like , "geometric fractal" or, "triangle behind man" or whatever help a lot.
Midjourney can do that. But it also shows it has no problem breaching intellectual property. It is amazing what AI can do, but as a copywriter I am not too happy with the way these businesses are working around copyright laws.
Sarah Silverman is suing OpenAI as it can give full summaries of the book. That means OpenAI has added parts of the book to the dataset, which is a breach of copyright.
https://amp.theguardian.com/technology/2023/jul/10/sarah-silverman-sues-openai-meta-copyright-infringement
I fully stand behind these lawsuits. I can imagine as a studio designing SpongeBob or Spider Man, you aren’t too happy spending millions on designing such a character and then see it being copied by a machine. I would say fan art could be an exception. But this is not a fan art. There was no creativity and is has been done fully automatically. No effort from a fan, so no art.
Yup.
https://imgur.io/a/HEMudsP
The full one of the first one was “someone in the forest tripping on LSD” & the second one was “someone in the mall after eating magic mushrooms” — tbh the first one is good enough you could slap it on a poster or shirt and sell it.
I feel like whenever I try to reference specific artists it rarely comes out the way I want it to. Sometimes though.
The following one is “psychedelic man finally achieving enlightenment while tripping on LSD after a very long life”
confused boner
(not really)
Wow these were generated by Midjourney? Might have to give them a whirl.
These cases are definitely grey areas one could argue about for hours, and definitely reduce to philosophical debates. I don't believe that, if you defined some reasonable scale of creativity that these tools would be at the very bottom of that spectrum.
For that OpenAI case specifically I don't see in that article where they are citing that OpenAI trained on their books. If they are saying that ChatGPT summarized their book equates to training on the original source itself I think that's a fairly weak argument (i.e. they could have scraped blogs of other people giving their review and discussion of the books).
@CyberneticTitan Fine, but also OpenAI seems to not have requested the permission of those sources either. The problem is that you are copying these articles into your dataset, which is not the same as reading info as a human.
Yes, we can debate for hours about it. Fact of the matter is that it seems not fully legal. I want to have my work protected. When someone pays a good amount of money for my creativity, it is his right to have an unique article.
Midjourney prompt: "lowendtalk"
Midjourney prompt: "low end talk"
My take:
Imagine I hire an artist and say "I want you to draw me a comic in the style of Robert Crumb". This artist then goes and buys a bunch of Crumb's comics, studies them, practices for weeks, and after a fashion is able to draw very similarly to Crumb. The artist then does the artwork and delivers it to me, and I pay him.
In this scenario, I do not owe Robert Crumb a penny. As long as I don't try to represent it as a Crumb original, I don't see any moral problem here either.
The step Midjourney missed is the "buys a bunch of Crumb's comics" step. They sourced without paying the artist.
There's a bit of novelty in law here, because existing IP law didn't envision this kind of scenario - would it have been OK if MJ bought books and scanned them? checked them out of a library? licensed Getty prints? - but in fact I don't think they did any of those things.
The Gates of Heaven, The Gates Of Hell: Choose your own adventure.
Let’s say your imaginary artist did a Google images search on “R Crumb” (try it) and learned the Crumb style without buying any Crumb works. That’s essentially the MidJourney methodology. How do you feel about paying your artist under these circumstances?
I would bet that 99% of the images if not 100% are actually a copyright violation as they were not authorized the rights holder (Crumb, or whoever he sold publishing rights to).
Actually Google image search returns virtually 100% legal licensed rights holders (mostly entities selling licensed books and merch, like Amazon.) Don’t bet…actually try the search.
Made a second version:
Choose Your Own Adventure: Heaven and Hell edition
I already think copyright is overly protected. It's auto granted, and thanks to big corps you can make 1 really good book/song and your family is set for decades. Even paterns can't match that level and its laws are also whacked.
Incase of that lawsuit, does it mean that I myself can't read the summary/review then tell my friends? Or those blog posts are already illegal? Since logically we can't fully define "a human'', "Conscience" or "Creativity" yet, boil down to the fundamental functions, what difference are we from the machines that we, ourselves, modeled after us?
There is still the user that sends a bunch of prompts to refine to one specific layout/arrangement of one photo, so not full-automatic, and that prompt has to be crafted as well. There is no way you can prove any of those.
The laws there barely keep up with the printed press let alone the digital age though!
As with any law or "moral" there has to be a line somewhere, the problem is there is no way everything fits neatly. "Transformative" for example, why can't one claim that a digital photo of an original art is transformed? The resolution is definitely much smaller than that of all the different molecules that give physical art color. What about cropped in with 90% remaining? 60%? 30%? 10 pixels? 1 pixel? Filtered to one basic colour (~33% remaining), grey scale, silhouette? If it is down to siluthete then there is no way no random thrash pile or one of the crafted shadow art installations is not breached.
With books/articles, what if there is no translation to one specific language and I do that, and in Asia, languages are mostly context based not gramar, even some of the original text are not translated, also the tenses are quite ambiguous, sometimes explanation or added time is needed. What then? Is it equivalent to the original book? Fanfic?
Which one is better?
Midjourney or Stable Diffusion?
Stable Diffusion can be self hosted (= unlimited prompts), while midjourney is cloud hosted, but with prompt limits.
In my opinion, midjourney makes better results.
Lets say that you hired an artist to do artwork in "Picasso" style, and the artist went to a museum and looked at Picasso paintings. No copyright breach of any kind, he just watched the publicly available paintings for inspiration, and then makes artwork in the same style.
This is basically what an AI does, it scrapes publicly available information to learn, or "get inspiration" if you will. AFAIK, no AI will buy source material, they use what is already publicly available (I guess that if you gave an AI your creditcard or paypayl account it probably could buy copyrighted source material, but I doubt anyone wants to do that).
Hence, suing an AI for copyright breach is stupid, it's just an attempt to get some money by someone who knows they will soon be replaced by cheap software.
"But but I was ripping off uneducated local business owners and that's at risk now!!!"
That's my impression of graphic designers today. My college degree is in design, I adapted earlier.
If I study a bunch of art and then make something unique but derivative, no one asks me to pay the artists I studied. This is what literally every artist does. Each artist advances the arts, they don't create things in a vacuum with zero influence. Even if they don't "study" art they ingest the creative work of architects while walking down the street. You can't avoid influence.
But if a computer does the same thing a bunch of people lose their fucking minds. It's clear why. They don't know how to adapt to progress, and it's everyone else's fault. Those are the people that would keep us in the stone age to protect their profits. Half of them have been regurgitating basic design elements for the last 20 years but all of a sudden they feel threatened by repetitive process.
In reality, they should learn to adapt. Learn to be better at writing prompts, take on more jobs for less money that you can complete in less time, and get paid for it. Because the donut shop owner down the street who wants a new logo? Yeah, he can't compete with your practiced and perfected prompt writing. That's what will separate the starving artists from the successful ones in 2024. We have new paintbrushes, expect hard times if you refuse to pick one up.