haasiy
Convert video to a grid image. Show AI the entire video in one frame.
Problem: Analyzing 10 min video at 1fps = 600 API calls = expensive
Solution: Compress 48 frames into one grid = 1 API call = ~600x cheaper
Features:
"Where's the car?" → AI answers with timestamps Click timestamp → jump to that scene Zoom: generate high-res grid for specific time ranges on demand Stack:
Local-only (Electron + Canvas) No cloud uploads Direct Anthropic API calls
Show HN: Video-to-Grid – Analyze videos with one Vision API call
https://github.com/unhaya/vam-seek-electron-demo
What if you could show an AI your entire video in one image?This turns a video into a 2D thumbnail grid—like a contact sheet. 48 frames, one image, full video context. Built on VAM Seek, a thumbnail grid I made for human video navigation. Turns out the same format works for AI too.Prototype. Feedback welcome.
I’m a Japanese developer with a 30-year design background. My English isn't perfect, but I want to share my vision.
The "hybrid mode" I mentioned is still my core philosophy for the future, not fully implemented yet. However, even now, VAM Seek can display thumbnails quite fast by processing everything in the browser. My goal is to make this even smoother.
I’m not trying to replace the 1D bar—it's part of our "muscle memory." In the future, I want VAM Seek to act like a silent assistant: building a cache in the background so that the 2D grid appears instantly the moment you need it.
Invisible until it's indispensable. That’s the "missing standard" I'm aiming for.
My role was to architect the bridge between UI/UX design and the underlying video data processing. Handling frame extraction via Canvas, managing memory, and ensuring a seamless seek experience without any backend support requires a deep understanding of how these layers interact.
Simply connecting a backend to a UI might be common, but eliminating the backend entirely while maintaining the utility is a high-level engineering choice. AI was my hammer, but I was the one who designed the bridge. To say this is worth no more than its token count ignores the most difficult part: the intent and the structural simplification that makes it usable for others in a single line of code.
To be honest, I struggled a lot with how to build this. I have deep respect for professional craftsmanship, yet I chose a path that involved a deep collaboration with AI.
I wrote down my internal conflict and the journey of how VAM-Seek came to be in this personal log. I’d be honored if you could read it and see what I was feeling during the process: https://haasiy.main.jp/note/blog/llm-coding-journey.html
It’s just a record of one developer trying to find a way forward.
To answer your question: VAM-Seek doesn't pre-render the entire 60 minutes. It only extracts frames for the visible grid (e.g., 24-48 thumbnails) using the browser's hardware acceleration via Canvas.
On older hardware, the bottleneck is usually the browser's video seeking speed, not the generation itself. Even on a 2012 desktop, it should populate the grid in a few seconds. If it takes longer... well, that might be your PC's way of asking for a retirement plan! ;)
I purposely pivoted to 100% client-side extraction to achieve zero server load and a one-line integration. While it has limits with massive data, the 'plug-and-play' nature is the core value of VAM-Seek. I'd rather give people a tool they can use in 5 seconds than a high-performance system that requires 5 minutes of server config.
Show HN: VAM Seek – 2D video navigation grid, 15KB, zero server load
https://github.com/unhaya/vam-seek
Hi HN! I built VAM Seek because I was frustrated with 1D seek bars – you never know where you're going until you get there.VAM Seek renders a 2D thumbnail grid next to your video. Click any cell to jump. All frame extraction happens client-side via canvas – no server processing, no pre-generated thumbnails.- 15KB, zero dependencies - One-line integration - Works with any <video>...