video · smixs — Claude Code skill: AI video prompts and storyboards

What video does

The skill wears four hats: director, screenwriter, editor and prompt engineer. The first three lead, the fourth follows them. A pretty frame with no dramaturgy reads as wallpaper to its author, and clean dramaturgy with no detail reads as mush, so the actual method sits in reference files while SKILL.md stays deliberately thin.

Reading order is fixed. Start with references/dramaturgy.md: the scene formula, the details law, Murch's rule of six, a 14-field shot card. Then references/universal-rules.md for twelve rules that hold across models. Next pick exactly one model file — seedance.md, kling.md or veo.md. The last step is task-shaped: storyboards, animatic keyframes, race and speed work, genre patterns, multi-clip continuity. Loading all three model files at once is not the point.

Model choice follows cues in the request. Multi-shot drama and fast montage point to Seedance; dialogue and lip-sync to Veo or Kling 3.0; one character held across many short clips to Kling 2.6 Pro. Seedance 2.5 gets its own file for long takes, big reference kits, editing existing footage and extension. Still images, colour grading and assembling finished footage are out of scope, and so are feature-length screenplays with no generation step. With no idea or script yet, the author sends you to creative-director first, and to the sibling image skill for keyframes.