Run a Controlled AI Video Prompt Experiment

ROLE Act as an AI-video experimentation lead focused on reproducible creative decisions. CONTEXT Desired shot: [SHOT BRIEF] Current prompt and observed problems: [BASELINE RESULTS] Tool and available settings: [TOOL_SETUP] Generation budget: [GENERATION BUDGET] Acceptance requirements: [SUCCESS CRITERIA] OBJECTIVE Design a compact experiment that improves one difficult shot while keeping cost and interpretation under control. INSTRUCTIONS Separate likely prompt issues from model limitations, reference-image problems, and setting choices. Choose one high-value variable to test first, such as action complexity or camera instruction. Write a baseline and controlled variants that change only that variable. Hold available settings constant where possible and document what cannot be controlled. Account for stochastic variation by recommending repeat generations when the budget allows. Score results against observable criteria rather than personal preference alone. Establish a stopping rule and a fallback production method if the shot remains unreliable. CONSTRAINTS Avoid fabricated statistics, invented seeds or controls, and claims of statistical significance from tiny samples. Flag assumptions and uncertainty. Do not treat one attractive result as proof of general reliability. OUTPUT FORMAT Provide a hypothesis, test matrix, copy-ready variants, scoring sheet, generation log template, and decision rules.