Use a visual summary when you want a similar idea, and a timestamped shot plan when timing and action order matter. A single attractive paragraph is often enough to explore a look, but it is not a reliable specification for reproducing a sequence.
Video analysis produces an interpretation, not the original hidden prompt. It cannot guarantee identical motion, identity, sound or editing when passed to another model.
Choose the right output mode
Similar-result brief: summarize the subject, action, scene, lighting, camera movement, visual style and quality requirements. Allow the duration and staging to change.
Timing-matched plan: record the source duration, split it into meaningful action beats or cuts, and state each segment's start, progression and end. Use verified media metadata for duration and frame rate when available; do not infer exact numbers from appearance.
Imagild's reverse-prompt workflow can be framed around these two goals. Review the analysis before using it as a generation brief.
Example: an eight-second coffee sequence
The following is an original hypothetical sequence, not an analysis of a supplied video.
| Time | Visible action | Camera and continuity |
|---|---|---|
| 0–2 s | The subject reaches toward the cup with the right hand | Waist-up framing; camera remains still |
| 2–4 s | Fingers close around the handle and lift the cup | Keep the same cup, hand and light direction |
| 4–6 s | The cup pauses near the face; eyes turn toward the window | A subtle push-in begins; no cut |
| 6–8 s | The cup lowers slightly and the subject holds a quiet expression | Settle into a stable closing composition |
This table states the order and continuity anchors. It does not imply frame-accurate control is available from every video generator.
Turn the plan into a concise generation instruction
Use the reference image for the seated adult, dark green shirt, pale room and ceramic cup. Over an eight-second continuous shot, the subject reaches with the right hand, lifts the cup, pauses near the face while looking toward the window, then lowers it slightly and settles. Begin with a static waist-up composition and introduce a slow, subtle push-in during the pause. Preserve the cup's shape, the subject's identity and the window-light direction. Keep hand contact and motion physically natural. No extra cuts or newly appearing objects.
Runway's image-to-video guidance distinguishes the visual information supplied by an image from motion instructions in the text. That is useful prompt structure, not evidence that its exact controls or timing behavior apply to Imagild's provider.
Check seven elements without inventing facts
Subject, action, scene, light, camera, style and quality should each contain observable or explicitly requested details. Put uncertain details in a separate note. Do not infer race, health, private identity or exact production equipment from a video.
Separate actual source properties from desired output properties. “The source is 24 fps” and “export at 24 fps” are different claims.
Frequently asked questions
Must the recreation have the same duration? Only when the user's goal requires it. Timing-matched analysis should preserve the source timeline; a similar-result brief can target another duration.
Can a long plan be generated in one request? A script's length does not establish the provider's single-clip capability. Verify current API limits and plan multiple shots when required.
Why does a precise prompt still differ? The model interprets timing and motion, while references and supported controls vary. Inspect each result and revise specific failures rather than assuming more words guarantee accuracy.