Sooo... I haven't been online for a month. At least not streaming or posting videos. Which seems weird as I just had a really fun stream in November, when several regular viewers tuned in just to see me 💖

Yet again, I've gotten sucked in the vortex of a new side project.

And this time I was sooo sure it could make me money! I was putting a lot of hours to make it work.

The project

Like the best side project, this one was born from my own needs. I wanted to be able to change outfits for my VTube model without having to pay a designer to create it from scratch.

Animated outfits (= textures) are not like regular images though... you can't simply generate them with current AI models.

So I figured out a process to generate a new outfit using existing AI models, then (through a batch process) "reverse-modify" the image to fit the original texture.

The idea seemed simple... at first I estimated an effort of a few days. (Isn't that always the case with a new project?)

But as I went along, it got more and more complicated. I learned a ton about 2D texture animation and how Live2D works, including its javascript SDKs.

After a week I realized I wouldn't be able to offer this as a self-service app... there were too many technical 'gotchas' and limitations to understand.

No big deal, right? I could just offer a custom service to generate new outfits for others, and handle the (still rough-looking) interface myself.

Getting ready to sell

By the time I had the process workable, I had lost weeks, and my enthusiasm had considerably diminished.

Especially once I realized that

• There was no way to guarantee that users would be happy with the result (outfits are highly subjective). And even a single negative review can kill a nascent custom service business.
• Generating new outfits this way would still take 30-60 minutes of manual work (at least the first one for a new model). Subsequent models could then be generated in seconds, but often required a few minutes of final touch-ups. Add all the potential communication with customers pre- and post-sale, I could not offer outfits for less than $10, which is not competitive when some Live2D designer artists in places like Indonesia or Pakistan can offer manually created outfits for less than that
• Some artist sell premade Live2D models with TOS that prohibits editing the texture. Which seems crazy to me... like buying a sofa and the seller prohibits you from putting on a cover to personalize the look. But I get it: artists sometimes sell these models cheap in order to get the "free advertising" when users stream with their model and credit them. This approach gets ruined when users drastically change the model.

I did end up posting the service, but haven't made a single sale yet 😔. Which is just as well.

The 'tism Tax

I probably have to write this month off as another one claimed by what I call the "autism tax." People who are on the spectrum (like me, with high functioning autism) can get "sucked in" by a particular interest or pursuing a particular goal - getting obsessed with making something work, tinkering, analyzing, comparing, exploring.

That is perfect for IT, where you need incredible stick-to-it-iveness to figure out stuff... from a new language, to a new javascript framework, to why your code doesn't work and debugging for hours. (That's why I believe many passionate coders are on the spectrum; maybe even the majority; even if they are undiagnosed.)

The downside is though... we often don't abandon or "change horses" quickly enough, when initial goals turn out not practical.

For every Mark Zuckerberg who writes software to track campus girls' relationship status there are millions of equally-autistic nerds who write projects that get them nowhere.

Lessons learned (?)

Back to this particular project ... The moment I realized I could not turn this software into a fully automated self-serve app I should've ended development, or at least sat back and thought hard if I really wanted to offer an AI-driven custom Live2D outfit modification service to a creative community that is largely anti-AI in the first place.

But my stubborn self had too much fun figuring out all the possibilities and wanted to play around with the SDK a bit more. Plus it was so fun just to see what outfits AI would generate, and then "see them in action" on models, animated by my own body movements. (Yes I'm a typical woman in this regard. Trying on various new outfits - even just virtually - is just so much fun.)

Sigh.

Looking forward

Well, at least it's only been four weeks, and I'm ready to move on to other projects now. And I can still use what I've written for my own VTubing activities.

But it's still a bit depressing, to go from high-flying hopes of having a successful app/business to realizing it won't make me any money, most likely. 😢

Let's pour one out for all fellow homies that worked on passion projects in 2024 that did not work out. 🍺 At least we learned a lot and had fun, right?

One of the things I like most about myself is that I can find as much fun (or more) just being on the computer all day 🖥️, as a "normie" does flying off to some expensive trip in the tropics 🏖️🌴 And those first two weeks when I thought I was writing software that would revolutionize the VTuber scene were... "Chef's kiss" 😂