Glitches are afoot at Playground AI this morning. Images not uploading, time-outs, unknown error crashes and logging me out of Pro while I’m creating, only able to get back by refreshing the page and losing my creations.
I predict new filters and/or DALL-E3 showing up soon.
Researchers at University of Chicago claim to have developed a way for artists to “poison” their images so AI software trying to train on them have their embedding info messed up.
Previously, they created “Glaze” which I don’t think ever saw much use since it made the original images look kind of shitty. This looks like the next step in the fight.
Adobe Illustrator recently added generative fill text-to-vector capability. As you know, vectors (as opposed to bitmaps) are infinitely scalable and easily editable.
A few vectors I generated:
It’s still in beta, and not quite ready for prime time professional work (e.g. logo design), but I see the potential. You can set the level of detail, choose type parameters (scene, subject, pattern, icon), and even use an imported image as a reference.
Adobe’s also starting to add AI to After Effects and expanding Firefly. Can’t wait till that matures. This video highlights some of the new stuff they’re working on.
If it’s not noticeable to the human eye, then an algorithm can learn to not notice it as well.
I asked for Minoan, but got Classical Greek. Nice picture, anyway.
Roman gladiators:
I asked Bing for a visual joke. Bing apparently only knows one joke. And it’s a little bit raunchy. (Scroll for 4 images)
Mildly interesting. I asked for a “splooting Boston Terrier” but got rejected on the prompt. Then I explained what splooting was and it generated the image. Apparently, it autocorrected to ‘shooting’ and rejected that. But I got an image of a shooting Boston Terrier anyway…
OK. last one. Bing rejected my prompt that was something like “A boston terrier as slurms mackenzie partying with girls in bikinis.” After it got rejected, I asked it to suggest a new prompt and it did! Thus, successfully generating some girls in bikinis.
Sorry for the Bing links. Too lazy to host elsewhere.
I doubt Dall E 3. At some point they removed DE2 as even a paid option. (And originally ot used to be a free one.)
This reminds me of something I made a few days ago. It popped into my head that I’ve never tried the mask of Agamemnon in any AI before. I tried to think of how to work it into a prompt, and after a few moments decided not to bother trying to describe how someone is using it and just let the mask itself do something. So I tried “the mask of Agamemnon windsurfing”. From the one set of images created it appears that DE3 doesn’t have an accurate idea of the mask of Agamemnon or of windsurfing, but that what it bluffed with is pretty awesome.
DE3 also understands the style of storyboards. Can’t tell a coherent story with them, though.
Trying to create fables in Bing/Dall-E 3. I couldn’t get the grapes high enough that the fox couldn’t reach them, but the images probably wouldn’t have been very interesting anyway.
Nick Valentine in an alternate universe where the Institute had to deal with a dying Earth rather than a radioactive one.
I was very impressed at how well DE3 understood and depicted this.
A terrified frog dangling from the bottom of a quadcopter by his tongue in the air above a pond with bullrushes and lillypads.
I’ve mentioned before how various AIs had little clue how to handle Gizmo and Alf (and E.T.). I have found that DE3 knows them pretty well. (With occasional inconsistent distortions. And E.T. tends to have long legs.)
Inspired by the above post, I created the following images in Bing:
Pretty dang good, actually
I just tried making that scene in 3D. In three of them Homer is a giant, but I like the 3rd image.
