Claude Code can ship a music video ad without ever opening a video editor. Here's how: 1. Idea Brief (Claude Code) Start by giving Claude a concept + a similar video in the style you like as a reference and have it save notes in the idea-brief.md. Claude can watch videos using the /watch skill from claude-video. This is super helpful to teach Claude what you're going for. 2. Generate Music (Suno) You can use Eleven Labs or Suno — but imo, Suno wins hands down. The music is just way better from what I've seen. Then asked Claude to pick a good 30-second window from the song. 3. Lock the script TO the song Listen to the song and create a storyboard – the song's beat grid is the spine of our storyboard. Write to it, don't fight it. Claude can listen to the song and design the storyboard and scene grid around it. 4. Lock the product as an anchor (CC uses Nano Banana) In this case, I generated one yarn AJ1 shoe and one yarn Nike box, then I threaded both as references into every keyframe. It's best to lock the props before the scenes otherwise it is much more likely to drift. 5. Then also lock the character references as an anchor (CC uses Nano Banana) First, generate character references. If characters look off, regenerate before moving to motion. It's much cheaper to catch issues here. These character refs are threaded into every scene keyframe. 6. Keyframes → Clips (CC uses Nano Banana → Seedance + Kling) I had 14 keyframes. Each animated via Seedance 2.0 or Kling 3.0 depending on the motion. Lacing intercuts go to Kling. Aerials and pull-backs go to Seedance. For the yarn→photoreal morph: generate the yarn start_image FROM the photoreal end_image so they're pixel-aligned. Only the texture differs. Then split the morph into THREE Seedance clips (yarn → 30% → 70% → photoreal) instead of one big interpolation. Smoother, and any bad stage is replaceable on its own. You may need to give feedback and iterate a few times here. 7. Stitch + Review (CC uses ffmpeg + Whisper) Then /watch the master cut and delegate the review to the agent. It catches more things and keeps me out of the loop as much as possible. 8. Turn everything into skills The first time I made a video like this, it took 3 hours. The second time, it took 1 hour. The third time, it took 30 minutes. All the learnings from the human-in-the-loop process should get encoded into skills / code. If all of the above sounds like a lot, it is and it isn't. It's a lot the first time, but if you start encoding everything into skills for Claude, it get much easier. You become the Creative Director, Claude does everything else.(Shiv)