Imagild field note · Reviewed September 26, 2026
Can AI reverse a prompt from a photo or video?
A reverse-prompt tool 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.
Imagild’s Reverse Prompt workbench accepts JPEG, PNG or WebP images up to 15 MB and MP4, MOV or WebM videos up to 50 MB and five minutes. It asks you to sign in before analysis. Uploaded media is sent to the model provider for that analysis; only use material you have permission to submit.
- 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 tool 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.
Questions before you start
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 the video limit mean a five-minute video can be generated in one shot?
No. Five minutes is the input-analysis limit for the reverse-prompt tool; generation length depends on the separate video workflow.
Prepared from the live product workflow and reviewed by the Imagild product team. Visual examples are labeled by provenance. Feature availability may change; see the editorial methodology and current tool before purchase.