Skip to main content
Every completed analysis includes a Recreation section — scene-by-scene, AI-ready prompts. These are the bridge between “understanding a video” and “building your version.” Open the Recreation section of your analysis to see them.

​
What each scene entry contains

{
  "scene_number": 3,
  "timestamp_start": "0:12",
  "timestamp_end": "0:17",
  "visual_prompt": "Close-up of hands on a laptop keyboard, papers scattered around, shallow depth of field, muted warm office lighting, slight motion blur. Shot: static with slow push-in. Style: candid, UGC-adjacent realism.",
  "animation_prompt": "Camera: static with subtle push-in. Motion: ambient — fingers shift slightly, paper edge flutters. Duration: 5s. Style: naturalistic, no dramatic movement.",
  "style_anchor": "UGC-realist, warm desaturated, tight-frame intimacy"
}
visual_prompt — describes the frame for an image model. Includes composition, lighting, subject framing, and visual style in generative terms. Drop into Midjourney, DALL·E, Adobe Firefly, or any image model. animation_prompt — extends the visual prompt with camera direction and motion type for animation tools. Designed for Kling I2V or SeedAnce. style_anchor — a short fingerprint of the visual language. Keep this consistent across scenes to maintain cohesion in your remade video.

​
Where to use them

ToolWhat to pasteNotes
Midjourneyvisual_promptAdd --ar 16:9 --style raw for video-oriented output
DALL·E / ChatGPTvisual_promptWorks well for realistic styles
Adobe Fireflyvisual_promptGood for brand-safe stock-style outputs
Kling I2VImage from above + animation_prompt5–10s clips; use V3 for realistic, Omni for stylised
SeedAnceImage from above + animation_promptBetter for longer motion arcs
A production pipelineThe full recipeHand it off to handle image, animation, and assembly automatically

​
Production workflow

​
Option A — Manual (scene by scene)

  1. Pick the 3–5 scenes that carry the most visual weight (hook, proof moment, CTA)
  2. Generate a keyframe image for each using the visual_prompt
  3. Animate each image using the animation_prompt in Kling or SeedAnce
  4. Assemble the clips with your VO in your editor

​
Option B — Hand it to a production pipeline

Export the full recipe and pass it to a connected production pipeline. It receives the entire analysis and runs image generation, animation, and assembly automatically — no scene-by-scene work on your side.

​
Tips

You don’t need all the scenes. Most videos have 15–25 scenes. Pick the ones that carry visual weight — hook scene, the proof moment, and the CTA close. 5–8 scenes is usually enough for a 30–60 second video. Keep the style anchor consistent. If the original was UGC-realist, warm desaturated, use that fingerprint across all your generated scenes. Visual style drift across cuts reads as low-effort production. Adapt, don’t copy. The prompts describe the type of shot, not the exact content. “Close-up of hands on a keyboard” is a shot type — replace the laptop with whatever your product interface looks like. For B2B ads: The prompts often lean into tight talking-head + product-screen combinations. Those are the two workhorses of B2B video. If the recreation prompts confirm that pattern, you have your shot list.