Step 1 — pick the photo before the template
Choose an image that is sharp, evenly lit and shows one adult subject. Crop closer than feels natural: a face that occupies a third of the frame will hold identity far better than a full-body shot.
Step 2 — open the shortest render first
Find the cheapest available generation and use that for validation. You are not making the final clip; you are checking whether the product reads your photo correctly.
Step 3 — pick the lowest-magnitude template
For a first attempt choose the template with the least motion: idle, portrait or subtle camera push. Low magnitude means few places for the model to fail.
Step 4 — generate and watch twice
Watch once for the overall impression, then again specifically for hands, then again for the background. Three passes catch what a single viewing misses, because attention is selective.
Step 5 — score the four checks
Identity hold, limb integrity, background stability and motion intent. Four passes means keep it. Two or more failures means the input needs to change, not the settings.
Step 6 — change one thing
If you regenerate, change exactly one variable: the crop, the template or the light. Changing several at once destroys the information the attempt produced.
Step 7 — write it down
Source photo, template, verdict. Three fields. By your tenth clip you will be choosing settings from evidence rather than from optimism.