Variation is an experiment
Ten random generations do not explain what worked. A controlled variation keeps the subject, offer, reference, delivery format and exclusions fixed while changing one creative variable. The result is a useful test rather than a folder of unrelated clips.
Freeze the base prompt
- Approved subject and reference assets.
- One visible action and factual offer.
- Aspect ratio, duration and resolution.
- Brand world, palette and exclusions.
- The frame or outcome that defines success.
The ten-variation matrix
- 1. Direct product or character reveal.
- 2. Problem-first opening.
- 3. Outcome-first opening.
- 4. Macro camera treatment.
- 5. Wider environmental treatment.
- 6. Static camera with subject motion.
- 7. Moving camera with restrained subject motion.
- 8. Bright commercial lighting.
- 9. Dramatic low-key lighting.
- 10. Alternative closing frame with the same core action.
Generate in small rounds
Create two or three first-round variants, reject technical failures and identify the strongest direction. Produce the remaining tests around that evidence. This protects credits and reduces the chance that a broken reference plan is repeated ten times.
Score the same criteria
- Subject and brand accuracy.
- Hook clarity in the opening beat.
- Motion and anatomy quality.
- Offer comprehension without invented claims.
- Ease of editing into the destination format.
- Credits and retries required for an approved result.
Promote the winner into the library
Save the winning prompt as a reusable pattern only after recording which reference, model and settings produced it. Keep the losing lesson as well: knowing that a camera move or lighting setup repeatedly breaks the subject prevents future waste.
