1. The “smoke and fire” rule (make it concrete)

Reject vague grandeur: Do not build visuals around abstract buzzwords like “system,” “architecture,” “underlying logic,” or “empowerment.”
Use lived-in detail: Map ideas to warm, everyday scenes—e.g. multi-agent teamwork as tiny kitchen sprites dividing cake work; single-model limits as one person juggling chores until everything breaks.
Stress texture: Prompts must highlight physical material (flour dust, cream sheen, strawberry color, oven glow) for “food appeal” or satisfying, tactile life moments.

2. The “mute” test

Picture = story: If you mute narration and dialogue, the audience should still follow setup, turn, and payoff. Voiceover annotates the image—images must not become a slide deck for the VO.
Visual loop and contrast:
Problem and solution should read through visual contrast, not explanation-only VO.
Weak: Scene 1 VO “solo is exhausting,” girl sighs; Scene 4 VO “teamwork is easy,” girl smiles—same vague staging.
Strong: Scene 1—girl whisks with one hand and struggles to pour flour with the other → flour explosion (solo pain). Scene 4—friend takes whisking; girl can sift flour calmly (team gain). Action contrast closes the loop.