AI-Generated Images.... Are We Fucked?

[Lemming]

Aside from the issue of the massive amount of energy required to power generative AI, one has to consider the issue of who will control the tools.  Soon as it becomes a sure bet for making money, licensing is going to be stitched up fairly quickly I would imagine.

Yeah, this is the concern. Open source alternatives are universally falling well behind commercial models already, for both image generation (Stable Diffusion has nothing on DALLE-3 in terms of sheer quality) and LLMs (all the ones you can run on your own are awful and will set your PC on fire in the process).

I'm amazed that things like DALLE-3 and character.ai are free to use right now, honestly, and it does feel like we might be in a brief era of exciting publicly-accessible innovation which could quickly come to a close.

That said, progress is continuing slowly in the open source scene. Stable Diffusion is slowly getting better thanks in large part to community efforts, and has the upper hand on the commercial models in that users at least have complete control over what the model is actually doing.

You have way more faith in the creative vision of the average person than me. Unfortunately I think we'll see plenty of worthless slop from the general public way outnumbering the passion projects once a film/tv series is as easy to create as a meme image is today. :(

The closest analogues we have right now are things like indie games, bandcamp/soundcloud music, and online fiction portals - things made by people purely because they want to make them, and distributed either for free or for very cheap. Which, admittedly, are 99% shit, but there's also been a lot of wonderful things from these scenes, and with some of this stuff, there's often a great deal of amusement to be had in just seeing the kind of out-there shit people come up with when they're doing it purely for fun and not to appeal to the mass market, even if a lot of it's pretty dreadful.

[PlanktonSideburns]

And all it took was thousands of years of stuff archived, fed into a network of computers that uses a sizeble chunk of our remaining fossil fuel and look! A slightly chintzy wobbly inexplicably wet looking elf

[touchingcloth]

In my opinion almost all very commercial art, including tv, movies, music, etc, is like that anyway. As far as I'm concerned Taylor Swift and her ilk may as well be algorithmically generated personas with everything from their music, life story, opinions and appearance entirely fake. It looks and feels like that to me now so who cares if it ends up being true?   

You have way more faith in the creative vision of the average person than me. Unfortunately I think we'll see plenty of worthless slop generated by the wider general public way outnumbering the passion projects once a film/tv series is as easy to generate as a meme image is today. :(

*If* a meme is as easy to generate as a meme is today.

Both of these posts do an injustice to people and what they value, I think. Taylor Swift won't be getting replaced any time soon, if ever, though as discussed upthread the "drudge" crafts she works with might be - some background instruments instead of studio musicians maybe, some cleaning up of vocals instead of a human technician. Even Mrs Brown's Boys won't be replacing their human writers and performers any time soon, if ever, with the threat there being to the drudge roles again.

No one is buying FIFA purely to look at the background art or spending thousands on Taylor Swift tickets because they find her indistinguishable from other artists. I think it's a huge leap to assume that computer-generated art will ever appeal to humans in the same way as human-created art does. Venture capitalists want us to assume it's inevitable, same as they wanted us to assume NFTs were all that.

[Astronaut Omens]

I think the line between real creativity and commercial drudge work is much blurrier than people make out, and over-emphasizing that difference undermines the extent to which the existence of commercial art currently makes certain kinds of non-commercial creativity viable. I don't just mean things like session drummers playing on commercial music in the day to support their jazz projects, I mean the fact that for a drum teacher's job to exist, there needs to be a range of possible things for drummers to do, a range of reasons for people to take drum lessons. Maybe the possibility of a job in graphic design means more students study art, which makes the art class that goes on to be important in the development of a great highbrow fine artist possible. You take that work-related element out and a college can't justify running a class for fewer students.
Say someone has a meteoric, brilliant talent that burns out early on, but they manage afterwards to make a living making commercial work,or teaching other people how to make commercial work - its easy to see this kind of move as a compromise, as someone giving in, or giving up- but I'm not sure if these kinds of jobs aren't really something a bit like the existence of pensions in that they actually help the main work to thrive.

     

[ajsmith2]

The closest analogues we have right now are things like indie games, bandcamp/soundcloud music, and online fiction portals - things made by people purely because they want to make them, and distributed either for free or for very cheap. Which, admittedly, are 99% shit, but there's also been a lot of wonderful things from these scenes, and with some of this stuff, there's often a great deal of amusement to be had in just seeing the kind of out-there shit people come up with when they're doing it purely for fun and not to appeal to the mass market, even if a lot of it's pretty dreadful.

I think indie games and music (as they exist just now) aren't that close an analogue to what I was talking about as they still integrally involve first hand human creativity and skill, and the equivalent in film/TV terms would be a self published movie someone filmed and edited on basic equipment and posted to Youtube. I'm talking more about where creating the film/tv becomes a similar process to the creation of an AI image just now, and all that's happening at the human end is the input of a handful of prompts, and out comes a movie. While I'm not saying that creating interesting work that way is impossible, I can't see it's mass adoption by the public as being a net gain, to see the least. It's gonna be 99% shite, and not even interesting shite for the most part.

[The F Bomb]

@Astronaut Omens Good post, as usual.

[ajsmith2]

@Astronaut Omens Good post, as usual.

Aye, seconded.

[Lemming]

I think indie games and music (as they exist just now) aren't that close an analogue to what I was talking about as they still integrally involve first hand human creativity and skill, and the equivalent in film/TV terms would be a self published movie someone filmed and edited on basic equipment and posted to Youtube. I'm talking more about where creating the film/tv becomes a similar process to the creation of an AI image just now, and all that's happening at the human end is the input of a handful of prompts, and out comes a movie. While I'm not saying that creating interesting work that way is impossible, I can't see it's mass adoption by the public as being a net gain, to see the least. It's gonna be 99% shite, and not even interesting shite for the most part.

I wonder if that'll be the most popular application of the technology, though - typing "make a gangster movie" and then waiting an hour while the AI does everything, and just watching whatever it comes up with, will only be briefly fun as a novelty thing in the way that simplistic online AI image and music generators currently are.

Where the technology will shine is in augmenting ideas that people already have; I don't imagine many people will be interested in AI-written scripts no matter how good they get, because the far more appealing option is to use AI to bring your own script to life*. Similarly, just uploading your script and typing "make into movie" and letting the AI do everything is far less appealing than actually manually "directing" each scene by hand, exercising constant control over what the AI is doing, interjecting to frame scenes in the specific way you want, etc. Having the AI do all the acting/voice acting is similarly less appealing than doing all the performances yourself or with your friends and then using AI to modify your voices/appearances into those of the characters in your story.

*GPT4 is already capable of writing fully coherent teleplays and novels, and nobody really seems interested in using it for this purpose beyond the comedic value of suggesting absurd premises and watching GPT struggle to come up with something

Again this is already the case with the image generation and voice acting stuff we have now, things like Elevenlabs and Stable Diffusion are prized for the amount of control they offer over output, and the way in which they're able to work with uploaded human-made material, rather than create things themselves. The big appeal of Stable Diffusion in particular is that it has features that allow the user to insert their own hand-drawn images or photographs and then have the AI make specific and precise adjustments as directed by the user. You can even see it with the large language models, people were gagging for an edit feature in characterai (which has now been implemented) because the text adventures and such are just far more fun when you have greater creative control over them as a human, and the AI is reduced to a supporting role. I'd suggest we're already seeing the ways in which this technology reinforces and aids human creativity rather than replaces or usurps it.

There probably will be a brief era where we're flooded with things that required no human effort beyond a prompt, but IMO the novelty will wear off fairly rapidly as people recognise the possibility of curating the output step-by-step and doing all the fun parts (scriptwriting, acting, directing, etc) themselves, having AI mainly work the practical side of things under close user direction and supervision.

[Dex Sawash]

Semi-related-

Alexa, how long do I cook a turkey breast?

"Simmer for 15-20 minutes on stove"