How I Solved My Own User Problem by Rapid Prototyping with AI
I've been using a genAI tool called ComfyUI for some months now, and one particular part of my workflow was problematic... 😰
[Article originally written in April, 2025] [Update: July 28, 2025]
The generated images all come with metadata, including the positive and negative prompts we write to explain to the AI how the image should look. This metadata is very important for a good workflow. Going back to some cool idea from the past, doing different iterations and having an easy way to visualize past metadata is essential.
I tried some free image gallery manager apps, but no luck. With more than 2k images, most apps started loading very slowly and didn’t offer a way to view the images’ metadata. So I set out to solve my own problem. I decided to create a web application that would work as an image gallery and show me the metadata I needed.
I opened ChatGPT and started chatting with it, explained the project, where it should run, what features I wanted and the technology to use, then asked for a plan. The plan looked solid, so I started vibe-coding with it. Even with manual copy/paste to files and terminal commands, the process went smoothly and the backend and frontend started running quickly. Meanwhile, I read somewhere - probably on Reddit - that GPT 4.1 was free for a limited time through Windsurf Editor, so I installed it and started using it in the Windsurf environment. Having an AI integrated directly into the code editor is a game-changer compared to the chat-and-copy workflow.
It took me just a few hours spread over 4 days to go from an idea to a working webapp, making my ComfyUI work much smoother.
I have a backlog of improvements: I’m fixing bugs here and there and adding new features during rainy weekends. But honestly, this was all about answering the question to myself: “Can I do it? How should I do it?” And if it solves my problem, then even better, which it did! Five years ago I would have been stuck relying on whatever applications existed in the market, but now I can just open an AI Code Editor and build exactly what I need.
The project is called GalleryFlow and it’s on GitHub if you want to check it out, give it a try or even contribute, it would make me very happy! ☺️