My comment is about the conditions that made TFA possible.
The business model explained in the article (do compute-intensive stuff and give it away for free) largely worked for Google, Facebook, and the big platform companies. Stability tried to do the same thing - building amazing tech as a loss leader - without a moat that gets people paying.
Combine that with a supplier that has huge market power (Nvidia) and you have a recipe for an entirely predictable bankruptcy.
It's an interesting question to ask what Midjourney and OpenAI/Anthropic have that Stability and Google doesn't have. The best answer I can give is "good taste" - specifically the ability to take relatively standard model components/architectures and tune the hell out of them so that a lot of people will enjoy using the end product.
Having good taste is a hell of a moat, which is how fashion designers build multi-billion-dollar brands (I am not talking about the old "houses" - those are about the logo, but new brands are different), as well as chefs and visual artists. It's just a very different kind of moat than the one we expect as engineers.
Midjourney's initial success was a combination of elusive marketing and a unique style. They focused on building a large community and took advantage of constant feedback.
> (I am not talking about the old "houses" - those are about the logo, but new brands are different)
Have you looked at Hermès' business model? It has never been about their logo. They'll often hide it from their product.
I agree with you about Hermes and actually the majority of the lines by most of the houses (not the cheap lines), but when you talk about fashion here, a lot of people think about the branding-focused groups like Gucci, Tiffany, or Louis Vuitton.