Research protocol
One Image or Four? Measuring the Value of Choice in AI Generation
Compare one- and four-output workflows using time to acceptable choice, confidence, duplicate rate, generation cost, and reviewer workload.
The central idea
More outputs can improve choice, but they also create review cost and near-duplicates. Measure whether four options help people reach an acceptable direction faster and with more confidence than a one-image regenerate loop.
A repeatable workflow
Define acceptable
Write task-specific criteria for a direction worth refining before participants see any outputs.
Assign workflows
Compare one-output sequential generation with four-output batches using matched briefs and generation budgets.
Measure the decision
Record time, generations, acceptable-choice rate, confidence, duplicates, and reviewer fatigue.
Report tradeoffs
Separate faster selection from better final quality and include compute, cost, and attention implications.
Worked example
Participants can solve ad concept, moodboard, and editorial metaphor tasks under each workflow. The test asks whether a four-image set reveals range early or merely presents four similar versions at once.
Review checklist
- Acceptance criteria precede generation
- Workflows use matched budgets
- Decision and quality are separate
- Attention and cost are included
Limitations
- Participant familiarity affects speed
- A small study cannot determine the best count for every task