Original Summary

My wife kept wanting to make her own visual novel, but kept getting stuck because she needed artwork and had to learn how to use some software. Last year I suggested Ren’Py. ChatGPT could help her with it, and she could generate the images there too. When she showed me her first prototype, I got that familiar urge to automate it. In a couple of days, I put together my first prototype, where some simple automation let me consistently turn a short folk tale into a fully working visual novel. And so, this spring, a friend and I decided this idea would work for our first project. We’d never built products before and thought it would be a great learning experience, even though the project would most likely be a commercial failure. So the initial vision was this. Maybe parents want to read fairy tales to their kids, and we could give them a service where they could pick classic tales, customize them around what their child likes, and read them together. Or the child could read them on their own. We started with a landing page to see if anyone wanted this. But since the page didn’t show the product, the $100 we spent on ads brought in one email address. And the landing page itself was pretty mediocre. Then we decided that to get people interested, we needed to show the actual product. But in that case, we might as well build the generation service itself. After spending a few months building and polishing it, we finished and ended up with the current version. Basically, we managed to build what we originally had in mind, but without the focus on parents, because we never quite figured out how to show it to them. For now, we’ve paused development of the service because, of course, we hit a wall with sales. For now, we’ll focus on simpler services and put more effort into sales to get some experience with that. But we learned a lot of lessons and gained valuable experience. 1. You can’t solve a problem you don’t understand. Initially, we wanted to make stories for parents with kids they could read fairy tales to, but we weren’t part of that audience ourselves and didn’t understand their pain points. We kept testing the service on stories that interested us more, didn’t figure out how to target parents with ads, and didn’t settle on a visual style for the novels in advance. 2. You need to think about sales as early as possible because it affects the product itself. For example, you need to let people try the product as early as possible, preferably without signing up. 3. Replit turned out to be a bit pricey. It works for a $20 prototype, but if you plan to keep working with it, you’re better off using other hosting providers with your own LLM subscription. 4. LLMs are very hard to control, and switching models over time is even harder. First, you spend a long time tuning the prompt and context to get consistent results. Then you switch models to get better, cheaper results and spend even more time adapting the prompt to the new model. We still have a few q

  • 情报分类：服务器与云资源
  • 分类依据：内容涉及服务器、云资源或网络线路
  • 信息来源：Reddit · SideProject
  • 发布时间：2026/10/1 06:27:36