Do not count generations; measure cost per usable second

01Measure cost per usable second

Credits, seconds, resolution and subscriptions cannot be compared directly across tools. Track usable-shot rate instead: when ten clips are generated, how many enter the finished video? A cheap attempt is costly if only one result works.

02Keep a simple cost ledger

For every generation record tool, model, duration, resolution, credits, wait, usability and failure reason. After several rounds you can see whether the weakness is the prompt, reference, model or shot design. Include direct subscription and API costs, local hardware and storage, waiting and editing time, plus privacy, rights and correction risk.

03Make one three-shot, fifteen-second project

Start with an ordinary object you may film: a cup, book or desk lamp. Avoid real people, lip sync and a complex plot. Make three five-second shots without dialogue: a wide establishing push-in, a close detail with one small change, then a medium ending frame that leaves room for captions.

04Validate cheaply before upgrading

Use low resolution, short duration or a fast model to test framing and motion. Once one prompt works, upgrade quality or upscale it; do not render every first-round candidate at maximum specification. Write three storyboard lines, keep one visual reference, make a few candidates per shot, record failure, then edit, add sound and validate on a phone.

05Name failures and define done

Do not write only ‘bad result’. Classify identity drift, wrong action, wrong camera, physics error, visual defect, audio error or factual error, then change the most likely cause once. The project is done only when all three shots connect, the object holds together, sound and captions are clear, rights and facts are checkable, and the export plays on its target platform.

Next: check rights, likeness and disclosure before publishing.

Continue to rights and publishing