I've read this thread with attention, and here are my thoughts on this matter.
1. There is no magical way to produce SS-like ships without using an
ethical workflow that involves some human-made art as the base concept. There aren't any AIs that produce anything like SS ships without something to guide the AI, as I've demonstrated and given full instructions for. Anybody who's used these tools to try and produce something will understand immediately; text prompts alone simply won't get results like this:
2. The way the AIs work, via using LLMs vs. internal matches to model data (i.e., a JSONL descriptor on a picture saying "photograph of an apple", "apple, "photograph", etc. to help the LLM match up with the tokenized visuals in a given model) to achieve
statistical convergence per layer (i.e., for every token used in a prompt, there is some attempt to find areas of the noise that "resemble" said token, and then this process is repeated many, many times to produce a final result) is
not like copying. Nor is it "thought", in the human sense (well, so far as we actually understand how humans think about visual things... which, frankly, we don't).
What's going on is iterative pattern-matching, using random noise as a seed value. If you ask for an apple on a tabletop... it'll look for the "most apple" area in all the pixels of the base image or noise, try and apply some visual tokens for "apple" to said areas, then another pass, and so forth. Typically the entire image is being pattern-matched for anywhere from 40 to 150 iterations, until either the iterator has completed or statistical convergence has resulted in multiple passes that look somewhat, but not entirely, like one another. If you tell an AI to do just 10 iterations, you'll end up with results that aren't terribly great, but as the number of iterations approaches convergence, things get "better", in the sense that the results are more likely to match the prompts.
So it's more akin to
collage than anything else, except, instead of being
one of the early models or the
more-advanced, computer-assisted versions... it's doing it with words and models that attempt to bridge the gap between the words, as tokens, and tokenized bitmaps. It's not copying in any real sense, though; what token within the model is picked for any given area is
random and it's iterated over many times, each time being pushed in new random directions. This is why it often produces strange and nonsensical results with stuff like human hands.
So, in the end, it's merely an algorithm. It's just a fancy filter. It's no different, except in terms of improved outcomes, as greeble filters for 3D models or the kinds of scalable algorithmic helpers we've used for things like heightmap design and other stuff in that vein for years and years now. So, why all the moral panic in this thread?
3. The supposedly-moral problem with these tools is that, to work, they had to be trained on existing works, so that they could pattern-match
correctly, and that they occasionally produce results that clearly resemble (to humans) some of the data they were trained on.
Why is that
fundamentally different than feeding a spell-checker a dictionary, or a grammar-checker many rules of sentence structure, or an engineering tool many different real-world models of failure? At the end of the day, most of the automation we use to do complex tasks in the real world are based on
modeling things that already exist, whether these are simple (words) or very complex (metal shear under load). But it's still just pattern-matching, and pattern-recognition is not exactly a
new computational area. You've all been gaining the benefits of these technologies your entire lives.
Nobody objects when an engineering simulator produces similar results to a real-world use-case, if the situation's similar. It's no different with this technology, really. Sometimes chance makes the available tokens line up with the same source imagery. That's all.
Should the companies building this stuff get sued because sometimes eerily lookalike images come out? Thus far, the answer, legally, is "no". Because when people examine this, and realize that that's more-or-less just luck... the cases have fallen apart.
4. "But it's coming for our jobs this time." If you're one of the creatives in the room... no, it's not.
Like all technology that
actually works, it's allowing people to create value more quickly. That's all. This is just another train where you're either on it or you're looking at a foreshortened career.
The printing press eliminated hundreds of medieval jobs copying texts by hand.
Word-processors cut down on the need for secretarial staff to transcribe hand-written notes.
Spell-checkers and grammar checkers eliminated tens of thousands of jobs as editing became much less time-consuming, transforming entire industries from advertising to education.
Photoshop destroyed traditional photography's economic viability by pushing creators towards, first, digital post-processing, then onto an entirely digital workflow.
Code-checkers gradually went from basic syntax to, "we really can automate much of code review and testing" at larger organizations.
None of this got rid of the people who:
A. Thought the thoughts that needed writing down about business decisions, processes and knowledge worth preserving.
B. Took the photos worth manipulating, improving and using on their own or as a part of larger works.
C. Designed the code that, cleaned up and optimized by increasingly-sophisticated tools, runs most of the world's economic activity.
So, frankly, I
reject this argument. I've been spending my entire life automating things to improve my workflow, because I'm lazy and I want the whole world to be able to afford to be lazy, too. Generations of very smart people figuring out to do it better, faster, cheaper- sometimes even all three- are why we're able to feed, clothe and care for more people than ever before. I get that dumb people think "progress" is just some tech-bro way to say, "you'll own nothing, and you'll love it", etc., but that's short-sighted. When we can create more value with fewer resources,
it's the opposite of zero-sum.
This tech, applied to visual art, really just frees up people to get on with things and make work happen more quickly. For commercial artists, it's more of a boon than a disaster; the next generation of visual artists
will be training their own AIs on their own stylistic approaches to improve their output and quit wasting time on, "this is how I like to light edges" or other small details, except when they wish to (and then they'll push said edits back into the AI models to improve its ability to save them time on their next works).
That will allow artists to focus on the big picture- what is their style, what are they composing, how do they want the viewer to feel, think or approach their work. That's not Big Tech Tyranny, it's freedom.
Yes, that means artists who don't develop a style and are basically just executing stuff that's already around... are hosed. That's
fine; we don't need more mediocre copy-pasta art from humans. Let the AIs do that, because they're good for that. But they don't create new ideas, conceptual approaches, visual styles or methods. For actual artists, who create new things and push boundaries and explore conceptually and teach us about ourselves... this is an almost entirely
good tech. None of this stuff can replace David's continual growth as a painter or designer. The coder versions of these tools can't replace Alex as a game designer. But both of them can be empowered by these tools.
5. At the end of the day,
would a ban be enforceable? How? Let's examine the methods.
"I think that looks like AI art, BAN" seems like a sloppy slope if ever I've seen one.
"I swear you're training on my stuff, even though putting my bitmaps next to yours, it's obvious to anybody with eyes that it's not actually the same at all" doesn't seem terribly tenable, either.
"You've admitted you use it, and we've decided it's evil" Fine, now you have an enforceable standard... but then
nobody will tell you.
This whole mess, so far as I can see, started because I was open, straightforward, and
shared practical techniques, even though I felt too ignorant to contribute meaningful information.
Up until this point... hmm. Character packs using generic AI art? Not a word. Mods using this stuff to make backdrops, etc? AOK, not a whiff of controversy. The hypocrisy of a few of the people in this conversation is mind-blowing, tbh.
Make spaceships, openly share how to do same?
Pandemonium. Panic. It's hammer time! Is that fair? Reasonable? Sane? Do we expect AI Witchhunts from now on, to chase people off these Forums or cast doubt on their work's legitimacy, or can we not judge the works on their own merits?
These tools are
power. Use this opportunity to get powerful and create more Fun for everyone. I've taken time to show you the way.
