Problem is, that network effect, uptime and reliability, design, and possibly more things I don't think of right now are more difficult to achieve, that a one user image manipulation program, and that LLMs alone won't get anyone there.