AI Deep End

 After Googling on November 14th, 2021, I found a few new online "text-to-image" generators:


"Giacometti sculptures on the beach"


Hypnogram worked pretty well for text "Giacometti sculptures on the beach":



Kinda-English ruDALL-E gave more realistic results (like the original DALL-E) than the VQGAN+CLIP programs, but was written by a Russian.  So the English text may not have translated well.  Note that every time you hit "Submit" this program delivers a different image:




The text in the left "Giacometti sculptures on the beach" translated into "Скульптуры Джакометти на пляже" on the right before delivering an image.  So I just submitted "Скульптуры Джакометти на пляже" on the right and generated better images:



I enlarged my favorite output with waifu2x:

Enlarged




Random Word Generator

Just to get completely creative and lose all control whatsoever, I inserted four random Russian words, created in the Random Russian Word Generator, and generated images that were completely unexpected:






Other "text-to-image" programs online:

Dreamlines 2.0 -- A very early "text-to-image" kinda thing that I rediscovered on my blog, from perhaps a decade ago

Random word/sentence generators:

One day we might get even more sophisticated and have text-to-movie (GPT4?).  AI will slice up movies and recombine them, the same way VQGAN+CLIP does for images.  Everyone might be watching their own personal movies on AI-Netflix, rather than sharing a common blockbuster experience.

I Googled text-to-movie and got Xtranormal (see sample YouTubes).  Maybe we are halfway there, inputting a GPT-3 script.


But I haven't see an Xtranormal video with an AI script.  Perhaps one could make one, and integrate it into a bigger movie...a movie filmed on their new iPhone.

No comments:

Post a Comment

Note: only a member of this blog may post a comment.