solid writeup, the Pinecone music caching trick alone was worth the read

but the $2/video headline is doing some heavy lifting. that's marginal cost once the pipeline exists. doesn't factor in the 40+ hours of debugging apis, handling deprecations, and figuring out why ffmpeg is mangling your audio sync at 3am

and honestly the bigger issue is that panning over still images isn't really "AI video". calling Ken Burns effects a video pipeline is a bit like calling a slideshow a film. the cost comparison only works if you're already selling that to your audience as the same thing