Decide whether you need a reference
Text-to-video starts with a scene description and allows the model to invent the composition. Image-to-video begins with an authorized still whose subject, environment and palette are already set. Use the second route when preserving an existing composition matters more than inventing a new one.
Write a motion brief, not another still prompt
After choosing the source, describe one subject movement, one camera movement and at most one environmental effect. A slow push-in with subtle fabric movement is easier to keep coherent than several simultaneous actions. Avoid asking a short clip to jump through multiple locations.
Budget for iteration
Video generation requires an account and credits. Review the selected model, supported duration and displayed price before submitting. Begin with restrained motion, inspect the result, then increase complexity only when the basic clip holds together.
Inspect across time
A good first frame does not guarantee a good last frame. Check whether the subject’s face, hands, clothing, background geometry and light remain consistent. Review the midpoint and transitions rather than judging a clip by its thumbnail.
Document permission
Use a still you generated or a source you own or are authorized to animate. Do not upload a private person’s photograph without explicit informed consent. Upload-derived outputs remain private.