Bilibili Analyzer

Takes a Bilibili video URL, downloads it, extracts frames using FFmpeg, and spins up parallel AI tasks to analyze the screenshots in batches. The workflow generates structured markdown documentation that reorganizes the content by topic rather than timeline, which is the key difference from basic transcription. You get tutorials with actual code snippets pulled from the frames, complete with source attribution like "from frame_0025". Built-in frame deduplication using SSIM similarity checks to avoid analyzing hundreds of near-identical screenshots. The quality checklist is unusually strict about image-text correspondence, requiring every screenshot to match its surrounding context. Works well for programming tutorials where you want the setup steps and code blocks without rewatching a 40-minute video.