The question
Can an AI agent make an anime lyric music video that holds up to a studio standard, and does it do better working alone or leading a team of sub-agents? I ran it as a three-phase experiment on my own songs (released as Rezephyr), with Yiyi, my Claude Opus 5.5 agent in Hermes Agent.
What was made
- P1, solo agent: Like a Prism, a colour GPU-shader lyric MV.
- P2, agent team: five storyboards for Lost Galaxy, 147 pages in five styles forced at least 80 % apart.
- P3, agent team: the INK SPIRAL MV for Rezephyr - Lost Galaxy (feat. Hatsune Miku), a strictly black-and-white manga video with 97 shots and hand-built lettering, plus a portable skills repository.
What I measured
- Drum sync: the team MV locks frame changes to the kick ×7.43 above random (solo ×2.86).
- Style: 0 of 241 sampled frames contain any colour.
- Cost: 266.82 M tokens across the lead and 24 sub-agent runs, $237.18 at API list price, read from the agent's own usage database.
- Time: the team phases spent most of their wall-clock time waiting for the plan's usage window to reset.
What I learned
Pre-production beat headcount: craft research, a written doctrine, storyboards and data contracts did most of the work; sub-agents made that pipeline fit inside context and quota limits. The full write-up, all charts, the 22-page PDF report and credits for all 58 artworks are on the research page.