Relative depth is a visual cue
Kivo uses Video Depth Anything Small to estimate relative depth from clips up to 30 seconds. Its grayscale output describes estimated near and far regions. It is not calibrated distance data, a 3D scene, or a guarantee that a generative model will obey the original camera motion.
Separate appearance from motion
Use an image to describe the setting or character, then add the depth video as a spatial and motion reference. Explain each role in the prompt. A depth map suppresses most color and texture information, which can reduce competing appearance cues from the original footage. It still contains silhouettes, scene layout and occlusions, so it does not remove every influence or anonymize a person. The generation provider applies its own input and output rules. Process only footage you have permission to use.
Convert for free in your project
Open a video in the asset library, select Convert to depth video, and choose Video Depth. Clips up to 30 seconds are processed asynchronously and saved as a new asset without spending credits. Keep the source: the depth result is another reference, not a replacement for the original clip.
Choose a clip that answers one motion question
A short continuous shot is easier to interpret than a montage. Trim to the camera move or action you need before uploading; the conversion accepts clips up to 30 seconds, not an entire film. Keep the horizon and subject readable. Fast cuts, motion blur, transparent objects and reflections can produce ambiguous depth. If the resulting map flickers or loses the subject, choose a cleaner source rather than expecting a prompt to repair the estimate.
A prompt that separates appearance and motion
Upload your appearance image and the converted depth clip to the same project, then choose both through @ in the prompt editor. The names below are examples: select your actual assets instead of pasting an unbound name. Kivo converts selected references to the numbered material labels required by the video model. Start with one simple move and use a Draft to inspect the result.
Use @Product image for the object design, materials and colors. Use @Depth move only as a reference for scene spacing and a slow camera push-in. A small ceramic lantern rests on a dark wooden table at dusk. Keep one lantern visible throughout. Soft warm light from inside the lantern, quiet background, no additional objects or on-screen text.
What is free, and what still needs review?
The depth conversion itself costs zero Kivo credits and requires a signed-in project. Generating a new video from the reference is a separate paid task with its own quote. Conversion runs asynchronously; keep the task and return to check its status rather than submitting duplicates. Compare silhouettes, floor contact and camera direction in the depth map, then check those same points in the generated Draft. Free conversion does not mean instant processing or guaranteed motion transfer.