Developer Open-Sources chalkpress to Convert Video Lectures into PDF Handouts

On 2026-11-12, a developer posted on V2EX announcing the open-source skill chalkpress, hosted in the GitHub repository luolin1024/chalkpress-skill. As described in the post, the project is not positioned as another video-to-text tool but as a way to compress a recorded course into a single booklet: each page carries one keyframe image paired with the subtitles from that time window, and the text layer is searchable and copyable, laid out as a landscape A4 document.

Usage works through an agent. The user sends a single instruction, together with the video link, to any command-capable client such as ChatGPT or Claude Code. The agent then runs the whole pipeline on its own: installing dependencies, running ffmpeg scene detection to grab keyframes, transcribing locally with Whisper, and then proofreading the subtitles. The post gives an example of the proofreading correcting homophone errors, turning a misheard term into the proper one, and notes that 241 terms from a software qualification exam glossary are built in for checking. Finally the material is typeset into a landscape A4 PDF, with an optional study note that rates exam points with star ratings placed as the first page.

On the design side, the post stresses that the pipeline itself is deterministic Python with zero LLM calls and no outbound requests whatsoever. The work that does depend on an LLM, such as subtitle proofing and note writing, is left entirely to the user's own agent, which performs it with its own intelligence. That is why no API key is required at any point. So far the only publicly available information comes from this single post; the materials do not include any follow-up developments, user feedback, or additional reporting from other sources.