🔥
Mar 29 (edited) • +I BUIILD FOR YOU
Curiosity killed the,..... ignorance? Mikimbri Mikinos Mardukis.
Curiosity killed the cat, .... meh, i guess it had to happen at least once. But what happens most times is that curiosity kills the ignorance.
Ever wondered how AI Generates Images, like the actual how... from Static to Real?
Think about an old TV with no signal, its pure static, totally random noise.
AI image generation works in a similar way.
This is how an Old TV Works (The Static Part)
When an old analog TV turn on or loses its signal, you see static, Thats the fuzzy, crackling snow on the screen.
What's actually happening is the TV's antenna is picking up random electromagnetic noise from the environment, with no organized signal to decode.
((Quick fun fact -- That static on your old TV? That's not just random noise — some of that is actual radiation left over from the Big Bang. The birth of the universe, showing up on your screen.))
Every pixel on the screen is essentially random black, white, gray with no pattern.
When a real broadcast signal comes in, it carries organized information, color, shape, movement.
The TV decodes that signal and the static resolves into a clear picture.
So a tv works like:
No signal = random noise = static
Organized signal = structured information = clear image
The TV itself doesn't create the image, it just receives and decodes what's being sent to it.
-------
AI does something wild with that same idea.
Instead of receiving noise and finding a signal in it, AI starts with noise on purpose and then guides it, step by step, into an image.
Imagine taking that Big Bang static and instead of just watching it, you could whisper instructions to it and it slowly organizes itself into whatever you describe.
Bam, and there's your fav show!
The first diffusion model was invented at Stanford in 2015 by Jascha Sohl-Dickstein.
Then in 2020 everything changed. Ho et al. dropped the paper "Denoising Diffusion Probabilistic Models" (DDPMs)
The name "diffusion" itself comes from the fact that the model starts with a high-entropy image. That means pure noise with no structure, like the static on the tv.
And, then gradually diffuses the entropy away, making the image more and more structured and realistic.
You know how the universe moves toward chaos, Hot goes to cold, glass breaks if it falls, things fall apart, order breaks down. That's entropy.
It's the natural direction of everything.
Diffusion models run it in reverse.
They start with pure chaos, max entropy, total noise, and then learn to walk it backwards, step by step, pulling structure out of disorder until something intentional appears.
The universe spent 13 billion years...
((Fun Fact : Zetsumetsu Mikimbri is 13.8 billion years old, the same exact age of our universe))
going from that Big Bang static toward complexity.
These models learned to do the opposite, collapsing chaos into creation in seconds.
And that idea sat in a Stanford lab for 5 years before anyone figured out how to unleash it.
This is Zetsumetsu 5th year. I wonder what will what mark we will leave on the year.
6
5 comments
Curiosity killed the,..... ignorance? Mikimbri Mikinos Mardukis.
