Accessible Open Source AI Graphics Generation like you've never seen

Oh. Wow. Does this change everything?

[ronin]

Another example of technology racing far ahead of what legal systems are currently capable of guiding and monitoring.

Occasionally, when pushed, I can come up with all sorts of creative visual ideas.

But my achilles' heel when producing visual media is that I'm not a very good artist.

Which is why these days, when not using photos I've taken, I stick to writing my own SVGs almost exclusively rather than trying to produce PNGs etc.

It strikes me that open source technology like Stable Diffusion has the potential to be a phenomenal game-changer.

[arstechnica.com...]

What does everyone else think?

[tangor]

I think it is kinda cool, kinda dumb, kinda stinks, kinda silly.

There's other AI image creators out there as well.

While some results are quite stunning, the vast majority are really meh, colorful, but meh.

Won't replace humans (yet) and the best that can come out of this is to devalue any kind of human generated art or photography. At worse it will create a storm of copyright infringements we've never seen before, or a sameness of result that makes images just a decoration rather than a revelation.

As a geek, however, it is interesting to see and know that a human can program a computer to output images that a human might make, and have absolutely no control over the "creativity" of "human" for copyright purposes.

Just a few thoughts before I really start to think about it. :)

[ronin]

I'm starting to wonder how much of Stable Diffusion is actually not much more than a parlour trick.

The hype suggests that the AI is creating pictures itself.

But it's increasingly looking to me like the AI parses the prompt and then returns pictures it already knows about (which it found on the web) and / or at best merges 2 or more pictures together.

Not that merging two pictures it knows about isn't clever - it is clever - but it's not in the same league as originating artwork, is it?

For example... when given the prompt:

modern apple watches with colorful hd displays in a surrealist style

Stable Diffusion returns a slightly mashed-up modification of Salvador Dali's 1931 painting, "The Persistence of Memory", differentiated from the original by the introduction of three overlays, somewhat resembling contemporary smartwatches.

I'm not persuaded at this point that Stable Diffusion is what it claims to be.

Not yet, at any rate.

[lucy24]

With Stable Diffusion, you may never believe what you see online again

Pics but it still didn't happen.

And then there was
:: quick double-check that this is Foo ::
that John Oliver report [youtube.com] from a couple of weeks ago...

[Sgt_Kickaxe]

I think Google is debating this topic more than anyone. How should they rank infinite numbers of new images created by AI? Every image will be unique and so they can't really be an indicator of spam or duplicate content or even copyright violations... it must be a hot mess on their end.

I just wanted to add that when I run the images from several of these AI sites through Google's image search feature a good number of Google's guesses simply say "dirty". I'm not sure why Google thinks AI images are dirty but suspect Google is noticing variations in straight lines. AI images of a known item tend to distort the edges a bit... and these get labeled dirty.

[ronin]

I think Google is debating this topic more than anyone

Well, that would be poetic justice.

In the 2010s, Google deploys AI to b@stardise search

ie.
You searching: apples, complementary pairs
Google responding: Did you mean apples, complementary pears? Searching for: apples, complementary pears
You: [sigh]

In the 2020s, more sophisticated AI comes along on the indexed-content side of things and Google's AI genuinely has no idea how to handle it.

[tangor]

AI v AI, popcorn on the stove. :(

Image value via machines <