Hacker News Reader

Comment by evanjrowley | Hacker News Reader

IMO the most realistic generated psychedelic images were made by Google's Deep Dream 10 years ago. It seems like it's been mostly forgotten about after the advent of DALL-E, Midjourney, etc.

View on HN

Agreed, it made some seriously wild stuff. Assuming you’re talking about this: https://research.google/blog/inceptionism-going-deeper-into-...

Here’s an example I generated in 2015:

https://www.wandereurope.com/wp-content/uploads/2015/07/drea...

Source image: https://www.wandereurope.com/wp-content/uploads/2015/07/mars...

Eh a lot of that ‘deep dreamy’ stuff ends up looking so same-same - I don’t need to look at endless various of ‘dog faces and insect body parts and glittering gemstones’ you know? I want to see actual weird stuff, bizarre genometry, a sense of space that’s mind-bending. Maximalism is not the be all end all of trippy visuals.

I believe this music video (Röyksopp - Impossible) was created with something along those lines:

https://www.youtube.com/watch?v=ljnxZypkjn0

Check this one out https://www.youtube.com/watch?v=PGuBpKqtHdY

And this https://www.youtube.com/watch?v=Dd3ZQOzyquo&t=385s

The AI generated ones are getting really good nowadays.

Deforum stuff is cool, but also a bit of a flash in the pan because of the disconnected nature of the generated frames.

Pure video generation is getting better, but like all generative AI it lacks imagination and cannot really create new things, making it less qualified for truly mindblowing psychedelic visuals in its current state.

It would be cool to have something like the first one, except generated in realtime and you get to "steer".

If it works by zooming in slightly each frame and then asking the AI to refine the image to better match the prompt or whatever, we could steer through the world by changing the centre of zoom.

If generating frames is fast enough it could feel like flying a spaceship through a psychedelic environment.

I understand your point. There's "trippy visuals" and then there's "that's what I saw when the DMT hit" - My comment was really leaning towards the latter.

Eh, it's close but misses the mark if we're specifically trying to match psychedelic visuals. If it were more fractally and geometric sure, but the dogs and fish and shit just isn't it.

This is pretty bang on for a high dose - https://www.instagram.com/share/reel/BALFNVBS20

Or this: https://www.instagram.com/share/reel/BBumGiQdP6

Seems closer to representing the experience, but probably less interesting for enhancing the experience.

made me wonder if the brain works in a similar way converting the incoming eye information into what the brain "sees" - it's just pattern matching based on what has been seen before and showing the best fit - psychadelics disrupt that.

I kind of miss the early AI days, where the AI felt closer to my subconscious than me.

Reply on news.ycombinator.com