Open Vandivier opened 7 months ago
if we sandbag long enough maybe someone will make a video reader instead of the existing image reader approach
still, image reader is a big win over not having it; lots of videos with no transcript bc they are an image, meme, screenshot, or text wall
all the video to text tools are still using glorified transcription
the low hanging fruit approach here is to take top-performing no-voice videos and write up the takeaways myself
the long term harder but high value approach is to make a tool that is able to take a screenshot as variable cadence (every 1, 2, 3, 10, or 30s...I doubt longer than 30s will be useful. It could be useful to have a list of user-defined time stamps though)
then, snap an image at those time stamps and apply an image-to-text by ai tool
that should really be its own saas tbh and it feels like smth that requires background job operation rather than being a serverless thing, i could be wrong tho
the ladderly custom GPT is built on transcript info
let's add machine vision where it takes images from high value videos and makes useful summaries from them