Hacker News Reader

Comment by samfriedman | Hacker News Reader

Without this technology, if I want to procure a concept-art quality image to promote my video game, I need to pay at least an amateur artist for their time and skill. With this technology, I can simply run some generations and maybe tweak or combine the output in Photoshop. I don't see how AI won't gobble up the bottom X% of the visual art industry almost immediately.

Fine arts, contemporary artists showing in galleries, mediums that can't be printed on photo paper (sculpture, canvas paintings, etc) will be more resistant to displacement. But the massive majority of commodity art & visuals - adverts, billboards, book covers, article images and so on - will definitely be displaced in the very near future.

View on HN

The "de-facto" standard for design and art in public spaces, corporations, websites, flyers, movies, posters, is that which can provably and evidently only be produced by well paid, furnished dedicated professionals. The standard isn't what looks good, it's what is evidently the work of someone or something with sufficient resources to be worth paying attention to. The standard will shift to something that cannot replicated easily by some 10 year old kid with 20 dollars worth of DallE-5 credits.

I can imagine the standard shifts from quality to scale. I.e a company doesn't give a shit that they have nice design, but rather they make something so ugly that noone could mistake it as being something someone commissioned from an AI, and spread it in necessarily low-supply spaces. (i.e. a new social media company's logo is a green poop emoji that they pay every west coast MLB player to get a tattoo of on their forehead).

Either way artists (besides tens of well-known ones) aren't needed too much. "OK SallY-5, make some crap that's unlike all these trending images [image list follows]"

Yes, and: as the space evolves, human artists will figure out what AI art is bad at doing and find new niches.

And then software will "learn," and the arms race will continue.

Right now, I think the winning move for artists is feedback and iteration: "make his head tilt more to the left. make his tie a lighter shade of blue." etc kind of modifications that aren't easily parseable by an AI.

That being said, I'm quite enjoying generating pixel art to stand in for my programmer art before I commission artists.

I think that's something AI will actually be very good at - GLIDE and StyleGAN are better at editing than DALLE is at image generation. But that's a tool artists can use; you just won't have to do the highlights and shadows in your art if you don't want to.

Newer art programs like CSP already have a lot of tools Photoshop didn't, even if it's just 3D models for poses.

You really need to have a closer look at this thing, because it can definitely do all of that, and much more.

Earlier today I saw a video (sorry, lost the link) of an AI editing program (still a prototype). It generates layers of a scene in place, say sky, mountains, rivers, foreground, etc. each of which would be individually manipulated. Next the program seamlessly blends the layers. The host then continued to add objects/subjects and through "in-painting" made all of it blend into a single coherent scene.

This kind of capability is close to being released. This year.

Sounds to me like the Nvidia Canvas App https://youtu.be/mlZYRwJ2oJg

Nope, it was a different one. You'd draw a shape (rectangle, circle) and the attach a prompt to it, after which AI generates it. Next, you can do advanced compositing.

> Right now, I think the winning move for artists is feedback and iteration: "make his head tilt more to the left. make his tie a lighter shade of blue." etc kind of modifications that aren't easily parseable by an AI.

This is literally how you do prompts to AI programs.

> And then software will "learn," and the arms race will continue.

How will the software learn? By having their work feed into the algo. The creative space is massive and fashions and currents change. In order for the algo to catch up it would need a massive dataset and by the time this is satisfied the art would have moved on to something else..

This tool will still remain and will most likely be used by artists too, though just as a tool in the toolbox, not as a complete process. Fotography didn’t destroy painting, it changed it in unforseen ways.

For most works using this as concept art would be deceptive and out right fraud in worse cases. The FCC doesn't pursue it so you'd probably be fine but this is not a replacement for an actual concept art unless you are working on a low quality or low poly game and even then this would be very misrepresentative. This has no real use case outside fraud.

What do you think concept art is? Any picture seems like it'd work as concept art. Even an uncreative illustrator AI would help if the person using it just can't draw.

I'm extremely confused. The situation illustrated was that of an indie game dev using Mid journey or whatever to generate concept art.

Who exactly is being defrauded? Is the dev defrauding themselves? Their future customers? How would anyone even know? It's not like concept art is used for anything external?

Reply on news.ycombinator.com