画質 高画質

This is pretty interesting! There's a significant difference between optimizing toward a purely text prompt, versus an image of that same text. For example, here is `unified meme theory` based on the image of the text

1/?

3 24

`the moon with the face of an acne-scarred teenager`, a vector illustration according to CLIP + DALL-E Decoder

0 2

`octobus` + `a vector illustration` according to CLIP + DALL-E Decoder

2 versions, 350 iterations vs 40 iterations

0 6

Oh wow, and with a somewhat large decay on the optimizer, it manages to get a solid color background like their samples as well, for "free"!

https://t.co/lG5Pr0R0BE

0 4

Another set of 78 alien planets, with CLIP+VQGAN

This time, just the stills

0 2

We are hiring! PhD positon available in my laboratory. NEU-ChiP project about neuroengineering and machine learning in neuronal networks. More info: https://t.co/XPiyZSa8Kc

23 29

Startup claims new "quantum analog computer" solved the traveling salesman problem for 128 cities https://t.co/M2qnYD6JhO

8 6

Startup claims new "quantum analog computer" solved the traveling salesman problem for 128 cities https://t.co/M2qnYD6JhO

12 9

Artificial intelligence researchers are kidding themselves that human-level performance is within reach, argues one leading thinker https://t.co/WfRbuklOZZ

13 15