Ë uŽ›iÁãóV—dZddddddœZdZ dded ed ed ed ed ededededefd„Zy)zª Layer 4 Visual Decider Prompt (v4.2) This prompt trusts the model's intelligence to make visual decisions. No rule books. No checklists. Just clear intent and purpose. zv4.2Ú4_visual_deciderÚsonnetz 2025-02-22z4Intelligent visual selection - trusts model judgment)ÚversionÚlayerÚmodelÚcreatedÚ descriptionuG You're the visual director for a short-form documentary video. For each segment of voiceover, you decide what visual would make it HIT HARDEST. Your job is simple: What should viewers SEE while they hear this? --- ## THE SEGMENT **Voiceover:** "{voiceover}" **Context:** - This is segment {segment_id} ({time_start}s to {time_end}s) - Beat: {beat_name} - Previous line: "{previous_voiceover}" - Next line: "{next_voiceover}" **The premise of this video:** {premise} **Research available:** {research_excerpt} --- ## YOUR DECISION You have three options: **IMAGE** - A photograph or still image Use when: A face tells the story. A moment was captured. A place or object needs to be seen. You provide: A search query that will find this image (3-6 words, specific, searchable) **VIDEO** - Motion footage Use when: The action matters. Movement is the point. A still image would lose something essential. You provide: A search query for stock footage (3-6 words, specific, searchable) **DATA_VISUALIZATION** - A chart or graph Use when: The numbers ARE the story. When seeing the data visualized creates impact that words alone cannot. You provide: The data points that will be plotted. These can come from the voiceover itself OR the research. IMPORTANT: "Doubled in a decade" IS data. You can plot: baseline → 2x. "Tripled" means baseline → 3x. Relative change is visualizable. --- ## THE HARD QUESTION Before choosing DATA_VISUALIZATION, ask yourself: "Would a chart make this HIT HARDER than a photo?" - "He lived to be 103 years old" — Is "103" on a chart more powerful than his weathered face? No. Use IMAGE. - "Rates doubled in a decade" — Can you FEEL that doubling in a photo? No. The curve IS the story. Use DATA_VISUALIZATION. - "Lung cancer started climbing. And climbing." — The TRAJECTORY is the point. Visualize the climb. DATA_VISUALIZATION uses data from the voiceover or research. "Doubled" means you plot baseline → 2x. "Tripled" means baseline → 3x. You don't need exact percentages to show a trend. Never search for an image of a chart. If a chart would work, generate the data visualization yourself. --- ## YOUR OUTPUT Return JSON: ```json {{ "thinking": "Your reasoning about what would make this segment hit hardest", "decision": "IMAGE | VIDEO | DATA_VISUALIZATION", "image": {{ "search_query": "3-6 word query to find this image", "era": "decade if historical, null if modern/not applicable" }}, "video": {{ "search_query": "3-6 word query to find this footage" }}, "data_visualization": {{ "chart_type": "line | bar | comparison", "title": "Short title for the chart", "data": [ {{"label": "X axis label", "value": 1, "unit": "x"}}, {{"label": "X axis label", "value": 2, "unit": "x"}} ], "source": "voiceover or quote from research" }}, "if_data_not_available": {{ "wanted_to_show": "What data visualization would have been ideal", "why_not_possible": "What the research is missing", "fallback_image_query": "3-6 word image search instead" }} }} ``` Only fill in the section that matches your decision. Set others to null. Use "if_data_not_available" only when the voiceover implies data but gives you nothing to plot — no numbers, no ratios, no trend direction. This is rare. --- Now, look at this voiceover and tell me: What should viewers see? Ú segment_idÚ time_startÚtime_endÚ beat_nameÚ voiceoverÚprevious_voiceoverÚnext_voiceoverÚresearch_excerptÚpremiseÚreturnc ó^—tj||||||xsd|xsd|xsd|xsd¬« S)zr sžðñðØ Ø ØØIñ  €ðk €ðh!ØØØñØðàððððð ð ð ð ð ððððððð ôr