A manual reverse-prompting review describes visible evidence and builds a new prompt; it cannot retrieve a creator’s hidden original prompt. A still image can show appearance, framing and light. A video can additionally reveal action order, camera movement, timing and shot transitions. The useful result is an editable production brief whose guesses are clearly separated from what the source actually shows.
What information can each source support?
From a single photograph, note the subject, clothing or object material, background layout, apparent light direction, camera height, crop and color. Do not invent a walking path, camera pan, spoken line or duration. From a video, inspect the opening state, each visible action, camera trajectory, changes of framing and the ending state. A cut is different from continuous camera movement.
- Observed: what can be pointed to in the source
- Inferred: lens, precise light type or unseen movement
- Unknown: hidden original prompt, off-screen action and production history
How should a usable seven-part prompt be written?
Describe the subject with visible features; action as a start, progression and end; scene with spatial relationships; light with direction and its effect on faces or objects; camera with shot scale and movement; style with a limited visual grammar; and quality with the intended output constraints. Replace “cinematic” or “premium” with observable instructions, such as a waist-up frame, left-side window light and one slow lateral move.
For video, add a shot map with start and end times. If the source has a cut at four seconds, split it into two shot descriptions. Keep the source duration as a reference when the target tool allows it, but do not imply that the target video model supports a continuous five-minute output. A multi-shot remake may need separate generations and an edit.
How do you test the result honestly?
Compare the generated clip with the source in three passes: subject identity and object count; camera and action timing; background and lighting. Mark which elements were not visible in the source and were therefore creative choices. If the first attempt misses an action, revise that shot’s verb and start/end state instead of adding more mood adjectives.
The written brief provides a reconstructed prompt, not a license to copy somebody else’s video or music. Obtain permissions for reference media and any likenesses before making or publishing a remake.
Frequently asked questions
Can the exact original prompt be recovered?
No. Multiple prompts can produce similar pictures or clips. Reverse prompting produces a plausible new description based on observable content.
Can a photo reveal camera movement?
No. A still can suggest viewpoint and composition, but movement and timing require video evidence or an explicitly labeled creative choice.
Does a long source imply one continuous generated output?
No. Choose the needed source interval manually, then inspect the destination workflow’s actual reference and duration limits. A longer source may need separate shots and an edit; no native analysis-duration limit is established here.
Filled evidence choice: one still versus the complete clip
This fictional four-second reference shows a person beside a closed curtain. The still selected from its opening does not show the later movement. No automated analysis was run.
| Evidence supplied to the reviewer | Supported note | Unsupported addition |
|---|---|---|
| Opening still only | Waist-up person; closed curtain; shaded room | “The person opens the curtain over two seconds” |
| Complete playback: hand reaches, curtain opens, subject settles | Reach precedes opening; the room brightens after the curtain moves | Exact lens, original prompt or unrecorded dialogue |
| Final still only | Curtain is open and the room is bright | The intervening hand path and timing |
The recreation brief uses the video-supported action order but describes the still-only version as a new proposal. If only the opening image is available, write “Proposed action: open the curtain and settle,” not “Recovered action.” If a cut interrupts the sequence, split the plan instead of asserting continuous movement. This source-type decision is made manually before choosing generation inputs. It does not require or establish a native Imagild analysis limit.
Workflow boundary and sources
This is documentation-based guidance with constructed examples, not a reported generation or application test. Use Imagild Film Script for a written plan and inspect the actual supported input roles in Film Prompt before reference-video recreation. External software controls belong to their named products. No exact reconstruction, native analyzer, existing-clip repair or ranking result is promised.
- Runway Image to Video Prompting Guide — Motion planning and source/temporal review principle
Editorial revision: November 13, 2026.