Original vector input and staging comparison only No AI inference
Run build-staging-qwen-study.ts with Bun and Sharp into a new output directory
ffmpeg -framerate 24 -i staging-frames/frame-%03d.png -c:v libx264 -crf 20 -pix_fmt yuv420p -movflags +faststart staging-read-study.mp4
ffmpeg -framerate 24 -i chunk-frames/frame-%03d.png -vf scale=1280:832 -c:v libx264 -crf 20 -pix_fmt yuv420p -movflags +faststart qwen-chunk-source.mp4