.png)
2 weeks ago
2
Hey HN!
I'm excited to share Musevise — a web app that helps you instantly identify and understand artworks from images.
- Why I built this When I look at artworks, I often can’t recall the artist’s name or the title. But without the right keywords, it’s difficult to search online. I wanted to solve this problem and make it easier for anyone to explore and appreciate art without needing to remember exact terms.
- What Musevise does Upload any artwork image or paste a URL AI (Claude 3.5 Haiku) analyzes the image and identifies the artist, title, and visual elements Automatically adds Wikipedia context for deeper understanding Works with paintings, sculptures, and various art forms
- How it works Musevise uses a vision-language model (VLM) to convert an image into descriptive text and retrieve related Wikipedia information. One challenge was that Wikipedia entries often vary in how titles or artist names are written (punctuation, translations, aliases, etc.). To handle this, I added a prompt-engineering layer that lets the LLM reformulate the Wikipedia search query dynamically — absorbing inconsistencies before hitting the API. This approach significantly improved matching accuracy without building a custom search index. Another technical challenge: sometimes the VLM fails to return output in the strict JSON format expected by the backend, causing parsing errors. I’m refining prompts and adding fallbacks to make it more robust.
- What's next Currently, Musevise automatically serves Japanese Wikipedia for Japanese browsers, and I’m working on expanding multilingual support. Try it out and let me know your thoughts!
Comments URL: https://news.ycombinator.com/item?id=45644244
Points: 1
# Comments: 0
