<aside> 🎬

How I made the La La Land goodbye scene from the reel, with the exact prompts I used.

No AI video tools at all. I generated 10 still images, and everything that moves — mouths, blinks, city lights, camera, motion graphics, captions — was built in code by Claude.

</aside>

The whole flow

1. Prepare 2 character reference images
2. Generate 4 scene images
3. Generate 6 more images with only the expression changed (mouth open · tears · smile)
4. Make the dialogue voices (TTS)
5. Give Claude the images + voices and ask for a motion-graphics video
6. Wrap it as a 9:16 reel

Before you start

1. Where do I run it?

• The video was made in Claude Code (Opus 5.5) — desktop app Code tab or terminal.
• You need Python and ffmpeg. If you don't have them, ask Claude: "check if they're installed and install them if not."

2. Where did the images come from?

• Nano Banana in Google Flow (image mode, 16:9, one image at a time). Image generation cost me no credits (that's on my account — Flow access can differ by Google plan).
• If Flow doesn't work for you, the Gemini app also generates with Nano Banana for free (daily limits apply).
• Any image tool that accepts reference images works.

3. It won't come out identical