Opening scene of Split Seek
After you drag in a folder
The results from your search
A moment playing that was found by the AI

Inspiration

I like to travel a lot and when I travel I build my camera to film my trip. I often come back from my trips with a lot of videos and wanting to make an edit/vlog about it, but spend hours scrolling through clips checking which video has a specific moment. I decided to build a tool that can search for me instead.

What it does

Split Seek uses Jev AI by typesafe.ai to search through your videos to find moments in your video. Drag in a folder, let it load, then search for moments.

How we built it

I built it by using Jev AI and Rust to ensure the results of your search come as quickly as possible. I knew going into this project that I want to use Jev AI by typesafe.ai due to it's fast inference speed and cost-efficient token usage. Tauri was used as the frame work. I used ffmpeg to help pull frames out of clips

Challenges we ran into

The videos I work with are my personal clips, which are shot in LOG (grey and lacking saturation), this made the detection difficult for the AI, so I had the program add a little saturation and a basic LUT to the clip to help the AI with detection.

Accomplishments that we're proud of

I am happy that I have a working app that I will genuinely use when editing videos from my travels. It becomes tedious to find clips so this app will help a lot.

What we learned

I feel as through I improve my work flow when it comes to building with AI. I usually follow this type of pattern when building, but this hackathon rounded out my edges. I felt like I learned to slow down a little bit more when building with AI. It is so tempting to just blaze through changes without checking the result.

What's next for Split Seek

I want to refine the search feature. Right now the app takes a frame every 2 seconds for the AI to analyze and the cut off for results is 60%. I want to test on more clips to see what the best benchmarks are.