The 10-step sequence
Who picks what: you = topic, format, script approval, voice, publish. AI = script draft + character pick, visual plan, pace check, render, audit.
You pick. The AI may suggest 3. Uses the master prompt TOPICS command + the scored topic bank (archers 90, Davy lamp 87, left shoe 84, divers 83, comma 80, ping-pong balls 79, canary 77). Never repeat used topics (backfire, fothering, bridge resonance).
Length: long 60–67s or short 40–46s. Vertical 9:16. Ricks Clips flat cartoon style.
The AI runs the master prompt SCRIPT command: it writes the full script AND picks the characters itself from the character sheets/views, matching the script. You do not pick characters manually. You approve the script (characters included) before continuing.
Pick props per scene from the library. The library starts empty — all props will live there later, with search, categories, and a preview per prop. Tools: list_props, search_props, get_prop.
The AI builds the shot plan from the script, characters, and picked props. One visual per narration line.
Automatic via MCP (check_film_pace + validate_film_plan). The AI fixes failures and re-checks until the plan passes. No approval needed here.
Your own voice, or an ElevenLabs dub.
On your Windows PC with NVIDIA GPU (Claude Code runs the skill there). The MCP is the brain, not the renderer.
The AI checks the finished video (PySceneDetect pacing audit, frame checks). On failure, loop back to step 5 or 8.
You decide where and when to post.