Turn a reference shot into an AI video prompt

Start with what the reference actually shows. Then make the creative choices for your new shot. A video analyzer can draft the first part; it cannot decide your story, guarantee an exact recreation, or see every frame.

Separate observation from invention

“Two characters in a field” identifies subjects. A production prompt also says what moves, how the camera behaves, where the action ends, and which details matter. A shot report gives you a first pass at those observations, but you should check the original around each timestamp.

Do not guess a precise lens focal length from a sampled frame. When a camera appears to move, describe the visible effect, such as “a subtle push in,” rather than claiming the original used a dolly or a digital zoom. Brief actions can fall between sampled frames.

A real sample, followed by a new creative prompt

Our published eight-second sample opens with an animated orange cat and a small duck in a grassy field beneath a blue sky. The report describes a slight zoom around 00:00–00:01. That is a reported observation from processed footage, not a generated video prompt.

Now retain the useful direction: a readable two-subject composition and restrained camera push, but choose original characters and setting:

The first description reports the reference. The second is a brief for a different scene. Keep those artifacts separate when the reference includes someone else’s characters, artwork, brand, or product claim.

Copyable video-to-prompt worksheet

• Subject: Who or what is visible? Avoid guessing identity from one frame.
• Action: What single, continuous action should happen?
• Frame: What shot size, viewpoint, placement, and aspect ratio do you want?
• Motion: How do the subject, camera, and background move?
• Look: Which lighting, palette, and setting details are genuinely visible?
• End state: Where should the subject be when the shot finishes?

Begin with the essential physical action and add detail only when it helps. Google’s Veo guide recommends clear subject, action, style, camera, composition, and ambiance; Runway recommends simple starting prompts and iteration.

Adapt the same observation to your generator

Text-to-video

Describe the whole new scene: subject, setting, one action, camera behavior, and look. Set 9:16 or 16:9 in the generator controls where available. Test a clear motion before adding dialogue, on-screen text, or several scene changes.

Image-to-video

The starting image already supplies much of the appearance and composition. Concentrate the text on motion, camera behavior, and what changes over time. This follows Runway’s image-to-video guidance. A video analysis does not create or license a starting image for you.

Several shots

Write one prompt per intentional beat, then assemble the clips in an editor. Track character appearance, wardrobe, props, travel direction, lighting, and aspect ratio across generations. Do not force a long montage into one short generation prompt.

How to use AdScanVideo for this workflow

1. Choose an original or authorized clip up to 10 minutes and 200 MB for the daily free or single credit workflow. Upload an MP4, MOV, WebM or MKV file; the analyzer does not accept video links.
2. Select Shot breakdown and run the analysis. Read the AI Video Prompt sections when present. Current processing analyzes video and available audio; it does not guarantee every edit, brief action or camera move is identified.
3. Check the original around each timestamp. Correct missed actions, on-screen text, camera interpretations, and dialogue.
4. Rewrite one selected shot for your own subject and chosen generator. Generate a short test, then adjust one prompt element at a time.

For production planning, see the shot-list guide. For another prompt example, see the video-to-prompt workflow. If a link fails, use the file-upload troubleshooting guide.

Find the actual prompt, then verify it

In a completed Shot breakdown report, look for the AI Video Prompt attached to the shot you want. Keep that generated text separate from your own rewrite. If a report omits the section, build a manual brief from checked observations and label it as manual rather than presenting it as product output.

1. Read the associated visual description and timestamp range.
2. Replay that range in your original file, including a little footage before and after it.
3. Check the subject, direction of travel and action ending. A pose in one frame does not prove the entire movement.
4. Compare camera movement against the background. A moving subject does not necessarily mean the camera tracks it.
5. Keep useful observations, correct errors and replace character or brand details for your own scene.

The public cat sample above is an observation example followed by an editorial adaptation. It is not evidence that the adapted prompt generated a matching video. Analysis also cannot recover the original creator's hidden prompt or supply reuse permission for their footage or soundtrack.

Free prompt revision worksheet

Download the manual prompt revision CSV. It separates the source observation, report excerpt, creative change and generation review. The supplied row uses the existing cat sample; generated result and pass criteria remain unmeasured. Add your own test results after watching them.

• Source evidence: a timestamp, visible detail and checked or uncertain status.
• Prompt version: your subject, one action, camera instruction and intended end state.
• One change: the specific instruction changed from the prior version.
• Result check: what the new clip actually shows, including failures.

This is an editable CSV for manual review, not a downloadable product report or an automatic similarity score. The free product report is readable on screen. Built in Copy, Word and PDF for an eligible free report unlock with one credit without rerunning the analysis; paid analyses include those actions.

When a generated clip misses the brief

Decide what failed before adding more adjectives. If the action is wrong, reduce the brief to one movement and a clear end state. If the camera is wrong, distinguish subject motion from camera motion and try a simpler camera instruction. If appearance drifts, use an authorized starting image where your chosen generator supports it and keep the subject description consistent.

Text instructions and reference controls differ between generators. Google's video prompt guide distinguishes camera, subject and scene elements. Use the controls available in your tool, then judge the resulting clip against your brief. A report alone does not guarantee the next generation will match.

Common questions

Can AI turn a video into a prompt?

It can draft descriptions from selected visual moments. Review the footage, then translate the description into an instruction for your new creative goal and generator.

Will it capture every shot?

No. Fast edits, brief actions and camera movement may be missed or misinterpreted. Replay the original and use one focused moment for your first test.

Can I use one prompt for Veo and Runway?

Reuse the observed shot and core idea, but adapt the inputs. Image-to-video workflows often need a starting image and motion-focused wording, while text-to-video needs the scene described in the prompt.

Does the CSV unlock product exports?

No. The worksheet is a separate free manual resource. Built in Copy, Word and PDF for an eligible free report cost one credit to unlock without rerunning AI.

Sources and example

• AdScanVideo: real shot-breakdown sample
• Google AI for Developers: Veo prompt guide
• Runway: Gen-4 video prompting guide
• Runway: image-to-video prompting guide