Twenty video-generation prompts for product ads, social clips, cinematic b-roll, and animation loops, plus a shot-list validator and a clip QA rubric. Each prompt gives your assistant a role, steps, a strict JSON output, a workflow ending at the generate button, a test plan (good input, broken input, verdict flips). 5 sections: 1. Product ads (6): hero loop ad, 3-shot feature callout, unboxing with colorway lock, before/after that refuses unsourced claims, app demo, aspect variant matrix. 2. Social clips (5): hook-first short (1.5-second hook cap), caption explainer locked 1:1 to shots, trend adapter that rejects protected references, sourced-quotes-only testimonial, loop-on-the-cut teaser. 3. Cinematic b-roll (4): establishing sequence on one palette and light, macro pack with plain-background rule, camera-move library, atmosphere shots with negative space for text. 4. Animation loops (3): logo stinger that composites your real mark (no model garble), kinetic type with no generated glyphs, ambient background that stays quiet. 5. QA and spec (2): a validator catching timing overflows, missing negative prompts, and bad aspects before you spend credits, plus a rubric that regenerates shots, not clips. The prompts produce shot plans and per-shot generation strings, not videos. Paste each gen_prompt into Seedance, Veo, Krea, Sora, Runway, Pika, or any text-to-video tool. 10 of the 20 ran during packaging: 1 real code run (the validator, 3 breaks, 3 flips), 7 spec lints with negative tests, 2 keyframe analogs, 10 flagged SCENARIO_ONLY: our pipeline holds no video API. Every prompt carries a negative test. Works in ChatGPT, Claude, Gemini, or any assistant. Spreadsheet copy included. License: personal and commercial use. No resale or redistribution of the pack itself.