My initial benchmarking passes were not great, and then they progressively have improved. This little device is surprisingly capable for what it is. Let me show you what I mean. Here are two images. The image of the geometry is from my initial setup and I was dismayed. Was this the extent of what this device could honestly do?
But then I kept at it and got an interesting coffee cup, and that’s what I realized we could go so much further.
In fact, as I continued iterating on the system I found I could start squeezing out even higher quality images. This is surprisingly awesome for me-as you have seen I’m generating a lot of images, and so this can be a material change for me. I kicked off a series of human character generations next. I’ll be writing on that experiment down the way.
Benchmark Variety
While I am not about to try generating video, what I have discovered is that this little system is capable of running coding models and audio models. Here are some results from early benchmarking tests with audio. Are you going to be gobsmacked? Honestly, no. However, it’s a starting point. I suspect even audio will materially improve as I tinker with this thing-it’s just what happens. You do a thing badly, and then you do it well.
This is a SFX that’s not particularly interesting, sounds like noise bouncing. I’ve included a download button if you want to use it. Reminds me of someone using a baseball bat on a metal fence.
The next audio is a cinematic brass hit with a short reverb tail. It’s not quite there, but imagine having free SFX for a vibe coded game. I’m pretty excited about this.
I have some thoughts to push the sound generation more. I’ll write on those results after the next set of human benchmarks wrap up. I also am eyeballing a full audio post. That could be interesting. Just a post with a ton of sound effects, free for you to use in your projects.
Here’s a few art samples of the Jetson. Free is a price that’s hard to beat.
Still a lot more to discover. My imagination stirs with what one might do with a hundred of these stacked up somewhere. Can I steer the device to an always on generative model?
There are times in tech with change so rapid that it feels like this lyric from Sophocles, many years ago. However, there’s a missing blindness to the rise of a sun that takes itself up over the horizon on the next day.
Scours in its running the subaqueous darkness,
Churns the silt black from sea-bottom;
And the windy cliffs road as they take its shock.
Sophocles' Antigone
The Sophoclean image fits the felt experience of tech change: the sea doesn’t politely update; it scours, churns, and strikes the cliffs. AI, platforms, layoffs, tooling shifts, business models collapsing - it often feels less like “innovation” and more like geological violence. Sometimes literal, if you flipped your desk in frustration.
Every major shift produces two hallucinations:
- Apocalypse hallucination: “Everything is over.”
- Salvation hallucination: “Everything is solved.”
Both miss the sun coming up the next morning - institutions still need trust, customers still need outcomes, humans still need meaning, and the messy world still resists clean abstraction. The silt gets churned up, yes. But then it settles into a new seabed.
Emotionally, rapid change feels like Antigone’s sea: dark, violent, impersonal.
Strategically, the winners are usually not the people screaming about the wave. They are the people who ask: what remains true after the wave passes?
In my mind truth is discovered thru action. Hence these posts. If I’m wrong, action will reveal the insight required for the next step. And when I’m right? Well, hill climb that value.