![\"\"](\"https:\/\/crepal.ai\/blog\/wp-content\/uploads\/2025\/12\/image-186-1024x579.png\")
<\/figure>\n<\/figure>\n\n\n\nHere’s what consistently worked:<\/p>\n\n\n\n
- \n
- Beat sheet with constraints: A simple 5\u20137 beat list with time stamps and one constraint per beat.<\/li>\n<\/ul>\n\n\n\n
Example (from 12\/28 at 10:42 AM):<\/p>\n\n\n\n
0\u20133s: Cold hook (question). Must show phone screen.<\/p>\n\n\n\n
4\u201310s: Problem in one sentence. No jargon.<\/p>\n\n\n\n
11\u201320s: Demo 1 action. Include close-up.<\/p>\n\n\n\n
21\u201330s: Demo 2 action. Add pop-up text.<\/p>\n\n\n\n
31\u201345s: Payoff + CTA. Keep it human.<\/p>\n\n\n\n
Why it works: Models latch onto structure and constraints without getting lost in prose.<\/p>\n\n\n\n
- \n
- Two-column notes (idea vs. visual): Left column = point: right column = what we should see\/hear.<\/li>\n<\/ul>\n\n\n\n
When I fed this to Claude, it gave me crisp shot lists that translated well to Descript and CapCut.<\/p>\n\n\n\n
- \n
- “Job to be done” one-liner up top: One sentence like, “Help solo marketers explain privacy-safe analytics in under a minute.” It kept outputs on message, especially with Gemini.<\/li>\n\n\n\n
- A short style reference: “Tone: Vox-style annotation, quick cuts, no stocky b-roll.” Links help, but even a sentence works.<\/li>\n<\/ul>\n\n\n\n
What didn’t help:<\/p>\n\n\n\n
- \n
- Long narrative paragraphs. The models pulled random details and invented transitions that felt\u2026 mushy.<\/li>\n\n\n\n
- Vague goals like “inspire” or “go viral.” You’ll get bland hooks. Use concrete viewer outcomes instead.<\/li>\n<\/ul>\n\n\n\n
If you hate templates, keep it minimal: a one-line goal, a 5-beat list, and one style note. That combo gave me the best signal-to-noise across tools.<\/p>\n\n\n\n
Understanding How AI Interprets Brainstorming Video Ideas<\/h2>\n\n\n\n
I wanted to see why some prompts clicked and others fizzled. So I compared responses side-by-side.<\/p>\n\n\n\n
How AI Reads Context, Intent, and Structure<\/h3>\n\n\n\n
- \n
- Context: Models map your topic to known patterns. If you say “founder story,” they default to hero’s journey beats. Give one counterexample (“No origin story, start at tension”) to steer them.<\/li>\n\n\n\n
- Intent: Words like “teach,” “demo,” or “convert” change pacing and CTA placement. When I used “reduce anxiety,” Claude slowed the opening and added reassurance lines. Useful for UX\/product explainer vibes.<\/li>\n\n\n\n
- Structure: Lists with timestamps generated more accurate hooks and shot lengths than paragraphs. OpenAI’s models, in particular, honored timing better when I used 0\u20133s, 4\u201310s blocks.<\/li>\n<\/ul>\n\n\n\n
A quick way to calibrate outputs:<\/p>\n\n\n\n
- \n
- Include a negative rule: “No stock phrases like \u2018in today’s world.'”<\/li>\n\n\n\n
- Add a measurable goal: “Viewer understands X in 45s.”<\/li>\n\n\n\n
- Provide a failure test: “If this script could be read over generic b-roll, try again.”<\/li>\n<\/ul>\n\n\n\n
Docs if you want to go deeper:<\/p>\n\n\n\n
- \n
- OpenAI prompting guide<\/a><\/li>\n\n\n\n
- Anthropic prompt tips<\/a><\/li>\n\n\n\n
- Google\/DeepMind best practices<\/a>.<\/li>\n<\/ul>\n\n\n\n
They echo what I saw: structure + constraints beat poetic paragraphs.<\/p>\n\n\n\n
Brainstorm Video AI Workflow: From Notes to Finished Concept<\/h2>\n\n\n\n
\n ![\"\"](\"data:image\/svg+xml;base64,PHN2ZyB3aWR0aD0iMSIgaGVpZ2h0PSIxIiB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciPjwvc3ZnPg==\")
<\/figure>\n<\/figure>\n\n\n\nHere’s the flow I used across three projects this week. It’s simple enough to reuse, but specific enough to avoid generic soup.<\/p>\n\n\n\n
- \n
- Draft the notes (8\u201310 minutes)<\/li>\n<\/ol>\n\n\n\n
- \n
- Make the beat sheet with timestamps and one job-to-be-done line.<\/li>\n\n\n\n
- Add two style signals: pacing and shot feel (e.g., handheld vs. locked-off).<\/li>\n<\/ul>\n\n\n\n
- \n
- Prompt for ideas (5 minutes)<\/li>\n<\/ol>\n\n\n\n
- \n
- Ask for 5\u20137 angles, not 50. Quantity kills judgment. I also request one surprising angle that breaks format.<\/li>\n<\/ul>\n\n\n\n
- \n
- Pick and deepen one angle (10\u201315 minutes)<\/li>\n<\/ol>\n\n\n\n
- \n
- I ask for a 45-second script, a visual plan per beat, and alt hooks.<\/li>\n\n\n\n
- I ban clich\u00e9s: “No \u2018game-changer,’ \u2018unlock,’ or \u2018in today’s\u2026'”<\/li>\n<\/ul>\n\n\n\n
- \n
- Sanity check with a real test<\/li>\n<\/ol>\n\n\n\n
- \n
- Read it aloud. If you stumble, viewers will too.<\/li>\n\n\n\n
- Drop lines into a timeline to check actual duration.<\/li>\n<\/ul>\n\n\n\n
- \n
- Move to a video-ready brief<\/li>\n<\/ol>\n\n\n\n
- \n
- Turn the best script into a concise brief with deliverables.<\/li>\n<\/ul>\n\n\n\n
- Move to a video-ready brief<\/li>\n<\/ol>\n\n\n\n
- Sanity check with a real test<\/li>\n<\/ol>\n\n\n\n
- Pick and deepen one angle (10\u201315 minutes)<\/li>\n<\/ol>\n\n\n\n
- Ask for 5\u20137 angles, not 50. Quantity kills judgment. I also request one surprising angle that breaks format.<\/li>\n<\/ul>\n\n\n\n
- Prompt for ideas (5 minutes)<\/li>\n<\/ol>\n\n\n\n
- Anthropic prompt tips<\/a><\/li>\n\n\n\n
- OpenAI prompting guide<\/a><\/li>\n\n\n\n
- Two-column notes (idea vs. visual): Left column = point: right column = what we should see\/hear.<\/li>\n<\/ul>\n\n\n\n
![\"\"](\"https:\/\/crepal.ai\/blog\/wp-content\/uploads\/2025\/12\/image-187-1024x559.png\")
<\/figure>\n<\/figure>\n\n\n\n![\"\"](\"https:\/\/crepal.ai\/blog\/wp-content\/uploads\/2025\/12\/image-188-1024x638.png\")
<\/figure>\n<\/figure>\n\n\n\n![\"\"](\"https:\/\/crepal.ai\/blog\/wp-content\/uploads\/2025\/12\/image-189-1024x465.png\")
<\/figure>\n<\/figure>\n\n\n\n