The video stack behind @I_Ship_With_AI

The tools, with the original sources

Install the video skills

My recipe: the rules I use

Every rule below came from a real note I gave on a real video. You can read them here, or download the same thing as a Claude Code skill.

• Voice first, picture second. Write the narration, generate one voice line at a time, and lay the scenes on a 120 BPM beat grid.
• Keep ~180 ms after each voice line. Cutting tighter clips quiet final sounds: “Code” became “Cogue”.
• Caption text from your script, caption timing from Whisper. Whisper mishears names and numbers; your script doesn't.
• Every hit lands on the word that says it: a stamp, a counter reaching its number, a ✓.
• The first frame is the thumbnail. Compose it; never swap in a different poster frame.
• No grain, no jitter, no random numbers at render time. Viewers call it flicker.
• Never show code or file names on screen. Say what happened in plain words.
• Music ducks under the voice, except the closing line. A drop right at the punchline feels wrong.
• One timeline, two formats: 16:9 for X, 9:16 for TikTok and Shorts.
• Check before posting: flicker (frame-to-frame brightness), loudness (about −15 LUFS), and “can you hear every word” through the music.

A plain text file. Put it in ~/.claude/skills/voiced-explainer-video/ and ask Claude Code for a video.

The honest part

The tools do the work, but the first version is never the one I post. My job is saying what's wrong: “where's the voice?”, “why show the code?”, “remove the flicker”, “the music drops at the end”. Plan on a few rounds. AI is the engine. I'm the one driving.