What's this about?
An experiment gave Claude Fable 5 and GPT-5.6 Sol each a $100 budget and full creative control to direct a music video end-to-end — writing the shot list, prompting video generators, and editing the result — then compared the two AI "directors."
What exactly is an AI "directing" a video?
The language model doesn't generate video itself; it acts as the orchestrator — reading the lyrics, planning scenes, writing prompts for a video-generation model, evaluating clips, and assembling the edit. It's the same agent pattern as coding tools, applied to creative production: the model is the director, other AI tools are the crew.
What's the community saying?
HN 146 points, 90 comments — mostly about creative failure modes
- Hilarious literalism — the videos illustrate lyrics word-for-word ("Michelle Pfeiffer, that white gold" → blonde woman in gold top); one commenter: the absurd literalism "perfectly epitomized AI slop in the first 30 seconds."
- Defense of literal — counter-thread citing Weird Al, "Riptide," and Land Down Under: literal music videos are a valid, fun genre.
- What's missing is subtext — great videos work on the *theme*, with story arc and reveal, not the words; current models don't do subtext.
- Scaffolding would help — leaning into animation styles and encoding "what makes a good music video" into the harness would likely fix much of it.