claude-real-video

Why we included this project

If you've pasted a video link into an LLM and gotten an answer built entirely from the transcript, this tool closes that gap. It processes video locally, pulling one representative frame per scene change instead of sampling at a fixed rate, dropping near-duplicates, and transcribing the audio. What you're left with is a tidy folder of images, per-frame timestamps, and text that any multimodal model can read. That makes it a practical fit for screen recordings, lecture slides, product demos, or meeting videos, where the on-screen content matters as much as the narration. It also works standalone as a plain keyframe extractor, and because everything runs locally, your source video never gets uploaded; you choose which frames or transcript to send to a cloud provider later.