Test an AI video workflow in ten planned attempts

Choose one available model and record its version, supported inputs and quote. Use sources you are allowed to process. Define pass conditions for action, identity, shape, framing and any audio requirement before viewing outputs.

Keep each original result, even if it fails. Record the full charge and the seconds you can actually use. Do not replace an inconvenient failure with an unrecorded extra attempt.

Keep the plan relevant to the project. The following ten checks deliberately separate difficult requirements rather than hiding them in one overloaded prompt.

• 1. A simple static composition with one main subject.
• 2. The same subject performing one slow action.
• 3. A restrained camera move around a clear scene.
• 4. A supported image-to-video shot with a clear source.
• 5. A face or identity-continuity check, where relevant.
• 6. Hands interacting with one visible object.
• 7. A product-shape or clothing-detail check.
• 8. A faster action or an occlusion typical of the project.
• 9. A required audio or dialogue check, if supported; otherwise repeat the most important visual task.
• 10. A second attempt at a representative task to inspect variation.

Watch the beginning, middle and end and listen to any required soundtrack. Note where a defect appears and whether an edit can remove it without losing the scene’s purpose. Count a partial usable section honestly.

Avoid awarding a single overall score that hides a fatal error. A product shot with a changed label may fail even if the camera movement is attractive.

Summarise which tasks worked in these ten attempts, what failed and the actual cost of accepted footage. Keep the conclusion limited to the tested version, settings and scenes.

Use the findings to improve the source or choose another compatible workflow. Stop at the agreed budget; further tests need a new plan rather than an expanding invisible comparison.