In one Claude Code session I went from a GitHub repo I did not know to an 80-second explainer video narrated in my cloned voice. HyperFrames renders the video from HTML, and Qwen3-TTS clones my voice locally from 20 seconds of a conference talk.