Spaces:

dalle-mini

/

dalle-mini

like 5.43k

Running

I, too, thought of using this to illustrate a fictional work. The lack of control over output either means one edits the photo to exclude the unwanted stuff, or modify the story to include it. You made reference to the fact that there is no persistence in depicted characters. Are you aware of any tools like Craiyon AI, that would allow for persistent characters?

To me it was just seeking for visions. The thing is like an oracle (somewhat like pouring lead or pancake dough), or closer like working with a dream recording device.

I am no AI expert, but I guess a system with persistent imagination (or editability) may need thousands of times more computing capacity because the process would become more recursive than linear. Craiyon AI e.g. not even properly assigns adjectives to depicted objects, so I think there is only a coarse picture recognition that counts in percentage how much of each verb is contained in the picture to prevent more reworking cycles (those perhaps would become infinite if the algorithm is not proven to approach a solution). Verbs at the begin of the text are rated higher than those at the end. So I expect that many things humans interpret in its capability of understanding text may be more like discussing with ELIZA. I remember that a decade ago there were reports that Microsoft planned to create a movie-making software for novices that may have been such an AI system. But I doubt that with nowadays computers it would be affordable enough to be useful for hobbyists.

I still like the way it can be inspiring and might have less human prejudices in combining objects and styles in unfamiliar ways.

I have just adapted the paradigm. Its a planet of mutated shape-shifters. They never look the same because they can't look the same. We humans are just so limited. You're right, saving creations and reusing them will take more processing power. This AI art is exploding. BTW, one artist has now copyrighted a graphic novel using Midjourney. It's AI owning the copyright that has been ruled against, NOT humans using AI to generate images, making mods to those images and then saying 'this is mine'. The artist in this case used text using the Zendaya, an actress, to achieve some consistency between scenes, so I guess that is approach. You're right, the craiyon image has trouble with adjectives and spatial relationships, e.g. above, below, beside, inside, right, left, etc., doesn't seem to matter, it just does its own thing.

Copyright has to be overcome; shortage is crime. And AI itself can not own a copyright (unless one day it will be legally rated a "human"). This is like the invention of automatic accompaniment and drum machines. Music pattern data inside the accompaniment device (that is manmade) can be copyrighted by its manufacturer (for banning resale inside a musical instrument or otherwise competing product), but not the music made by musicians using it. Feeding an AI with the artist's instructions to produce/customize a work of art is the same. Fully automatically generated AI works can not have copyright at all; particularly selling huge quantities of automatically generated NFT "art" needs to be rated an unlawful pseudo-currency solely produced as a tool for veiling tax fraud, drug dealing and terror financing.