this doesn't explain how it can synthesize images that have never seen before. I'm pretty sure a space satellite donkey with feather wings was not in the training data.
2026-05-14 13:03:28
81
Johnny :
It’s a collage machine.
2026-05-13 18:04:48
260
ཐི༏ཋྀ :
Stop using AI, all programs, tools and etc
2026-05-15 10:56:30
44
GYVAN :
ChatGPT when I ask it for an original image
2026-05-14 04:58:08
83
irongoat012 :
the way ai generates slop is more interesting than the slop itself
2026-05-16 01:59:07
2
prick676767 :
Who cares, you get an image
2026-05-13 18:54:12
12
A Bussy So Sloppy 🍑💦 :
why doesn’t the same prompt create the exact same image?
2026-05-14 22:07:15
2
John Smith :
Most new models nowadays are flow matching based, not diffusions anymore.
2026-05-15 10:11:50
10
3,543 others :
I've had the hardest time getting AI to create a 16x16 matrix array Pixel image. you'd think this would be much easier to do, but I found it so difficult to get good results.
2026-05-14 02:02:11
2
Now We Build :
Reversing random noise into a high-fidelity asset is just pure mathematics. The fact that numbers translate into raw creativity is wild.
2026-05-22 13:18:47
1
Felipe :
and that latent space is thousands of dimensions deep
2026-05-14 05:49:13
0
Kevin :
PENGUINS MENTIONED OMG!!!!!
2026-05-16 21:44:02
2
rain :
our way is simpler😭
2026-05-15 22:25:58
1
Sagheer :
I use stable diffusion on my rtx
2026-05-14 21:51:12
1
smc :
they do see it as we do, as we see it as you just described. pixels (wavelengths/photons) which are processed in a computer (cpu) and translated into words
2026-05-14 22:44:45
1
derty :
and there is flow matching to reduce the inference time of normal diffusion models
2026-05-13 22:56:38
1
Hervé Lalimace :
Awesome explanation !
2026-05-14 06:43:40
1
Now We Build :
We went from teaching machines how to count to teaching them how to hallucinate patterns out of static noise. The engineering behind latent space math is wilder than the actual art it outputs.
2026-05-21 06:56:18
1
🎃 Goose 🎃 :
as much as I do dislike AI "art" it's really impressive that we managed to advance so far in such a short time
2026-05-14 06:19:17
7
onetwomiku :
latent diffusion slooowly being replaced by pixel space diffusion (again). would be nice to have video about it to send to friends )
2026-05-14 19:31:57
0
12345aeio.u :
2026-06-27 11:51:29
0
ruben :) :
wdym impossible to visualise?
2026-05-16 09:03:30
0
Slime :
this was not a good explanation, and I'm not sure how much of this is even accurate
2026-07-17 20:47:26
0
Orenz :
2026-05-31 09:53:45
0
✨️✨️✨️✨️✨️✨️✨️✨️✨️ :
cool
2026-05-13 18:12:56
0
To see more videos from user @futurewithfawzi, please go to the Tikwm
homepage.