In 2017, I was having an existential crisis around my gender identity. I was in a very dark place. I discovered FaceApp which helped me build the confidence to transition.
Shortly after that, I figured out that the app would see faces that weren't there in some pictures. At the time, I called them "dirt faces," because they would most commonly show up in mud puddles or lawns.
Eventually, I figured out how to bend it to exploit intentionally, and my fascination with neural bending was born.
In addition to taking a billion transition selfies, I used the app every day, trying to trick it into fleshing out my drawings, using it as a generative tool for making what I considered to be a radical new form of art that I believed only I knew about.
Characters and depth targets
That's when I discovered the vtuber thing. People making animated character models with game engines like Unity and Unreal. The tool there was not an AI tool. Well, mostly not an AI tool (AI is in everything).
And I started subverting and repurposing Vroid Studio to make more interesting art that was outside the use case.
Vroid studio reminded me a lot of my experiments with Poser back in 2007, so it was pretty easy to get into, and I would use my AI outputs from FaceApp as a manual reference. I considered this "ai art," and started using the AI art tag about this time.
Simultaneously, I had developed a theory I called "depth targeting" based on what I thought I saw FaceApp and artbreeder doing.
AI researchers told me I was wrong and/or crazy, but my theory of depth targeting, and the fact that these systems develop a sort of 3d cache was later vindicated in 2023. That is how the system works.
I used the theory of depth targeting to develop what I called "noise painting" mostly in its current form. It hasn't changed much. It takes less iteration now, but it's still the same theory. The machines are just better.
This expanded when I discovered Reallusion's Headshot for Character Creator 3, which used the same base PyTorch libraries at the time that FaceApp did! You know what that means... everything I had learned up until that point was suddenly applicable to the 3d animation project! And I was SO excited.
My results, preliminary as they were, were absolutely fucking mind blowing. This was cool! I really felt like I was on the cutting edge, and there's no arguing that. I was.
The backlash and the discovery
Anyway, I kept going like this until that one day in August of 2022 when the Zapata video dropped and all the nazi catgirls showed up trying to censor anyone who was making any kind of art that involved any kind of AI. They weren't really educated on the topic, but they had the smugness to make up for it.
Over three months, I received over 250,000 messages, including racism, transphobia, and death threats, and antisemitism. All around and about my use of "diffusion models."
I had no idea what a diffusion model was. But when I checked it out, I realized something super interesting.
See, it turns out, that when you generate a face with a diffusion model, the depth targets are easier for generative systems like Headshot to use than regular photos, which meant more accurate renders.
And this is true, even if you take a regular photo, and run it through the machine, without making any meaningful changes. I was on fucking cloud 9.
But the outrage sparked something in me that I had only felt a couple of times before, and never that intensely. A feeling that can only be described as "Dadaist Disgust."
It was a moment of metamorphosis that made me into what I am today.
From control to chaos
Around the same time, and for unrelated reasons, I had started pushing harder on my noise paintings. Making them wilder, crazier, more chaotic. Up until that point, noise painting had been a way to control the controllable.
But the revelation came when I realized that you could also use them to make the machine hallucinate in amazing breathtaking ways, creating entirely new, and sometimes impossible aesthetics that could be created no other way.
AI's have no frame of reference for the real world. They guess at things, based on what little they know. But humans do, and that's a problem, because your own biases affect and limit what you yourself are capable of creating.
AI's can create images that your hands cannot. Your brain won't let you do some of this stuff. You can do things with geometry, and perspective, and anamorphic layers in seconds that would take you years of personal development and brain hacking to even contemplate doing as a human artist.
It's so fucking magical. There's nothing that even comes close.
It sucked me in.
A new playground in sound
When Chirp hit, I realized quickly that I could do the same thing with sound. And that opened up an entirely new playground that was so strange and unfamiliar and magical. Very satisfying creatively.
And the rest is history.