1. Start with a source you can actually use
Use footage you recorded, commissioned, licensed for derivative work, or verified as public domain. A public social URL is evidence that a clip exists; it is not proof that you can download, edit, or publish it.
Keep the original media file when possible. Platforms recompress video, crop frames, and may remove metadata that helps analysis.
2. Choose a shot the model can preserve
Begin with one subject, a visible body, limited cuts, and a camera move that is easy to read. Occlusion, hand-to-object contact, reflections, and fast rotations are useful benchmark details, but each one raises the failure rate.
A three-to-fifteen-second shot is a better first test than a fully edited Short. You are testing identity replacement and continuity before testing an entire story.
3. Match the character reference to the source
Pose, framing, and clothing silhouette matter. If the source shows a seated three-quarter view, a clean three-quarter character reference usually gives the model less geometry to invent than a front-facing headshot.
Use fictional or authorized subjects. Public figures, political impersonation, minors, and deceptive identity uses should be rejected before provider spend begins.
4. Save the route and quote before generation
A production workflow should record the provider route, requested duration, resolution, product credits, maximum provider spend, and the exact inputs approved for the job. That turns a generation into an auditable attempt instead of an unexplained charge.
ClipRecast quotes the route before a job is queued. The quote is saved with the job so a later upstream price change does not rewrite the history of what you approved.
5. Review continuity, not only the face
Inspect hands, object contact, feet, hair edges, clothing boundaries, camera motion, and the first and last frames. A convincing face with a disappearing bottle is still a failed creative.
Keep technical failures separate from creative preference. That distinction is what makes automatic credit returns and provider disputes defensible.
6. Change one variable in the next test
Once the motion is stable, vary one thing: character, wardrobe, visual style, voice, or language. Small test matrices make it possible to learn why one output performed better instead of merely generating more files.
The goal is not to produce the largest batch. It is to reach the next publishable test with a known cost and a reason for the variation.
This guide describes a product workflow and general information; it is not legal advice.