Creative video analysis

The Creative Analysis Lab breaks down short-form video ads and organic content frame by frame to diagnose retention bottlenecks, evaluate visual movement, and predict conversion performance.

How creative analysis works

Upload Video (.mp4 / URL) ──> AI Frame Extraction ──> Speech & Audio Sync ──> Retention Curve Diagnostics

1. Upload or link: Upload a video file (.mp4, .mov, .webm up to 250MB) or paste a public video URL (TikTok, Instagram Reels, YouTube Shorts).
2. AI multimodal evaluation: Computer vision and audio pipeline extracts frames every 0.5 seconds, transcribes speech, and maps audio energy levels.
3. Diagnostic report: Receive a comprehensive score breakdown and specific second-by-second improvement recommendations.

Evaluation breakdown

1. Retention curve simulation

Maps simulated audience attention from 0:00 to video completion:

• 0–3 seconds (The hook): Percentage of viewers captured. Scores above 70% indicate top-quartile performance.
• 3–15 seconds (The setup): Identifies where pacing drags or visual monotony causes drop-off.
• 15s–end (The climax and CTA): Evaluates whether the call-to-action is clear, timed correctly, and engaging.

2. Five-point radar score

Each video receives an overall 0–100 score across five dimensions:

Dimension	What is evaluated	Optimal range
Hook strength	Visual contrast, facial expression, bold opening statement	80–100
Pacing velocity	Cut frequency (scene changes every 1.5–2.5s), speech cadence	75–95
Visual engagement	Dynamic text on screen, b-roll quality, zoom transitions	70–90
Audio clarity	Speech intelligibility vs background music volume balance	80–100
CTA effectiveness	Single, clear next step with visual and vocal reinforcement	85–100

3. Frame-by-frame breakdown

Inspect every individual scene:

• Exact timestamp with extracted thumbnail.
• Detected visual elements and emotional expressions.
• Speech transcription synchronized with on-screen captions.
• Actionable critique (for example, "Scene 3 is 4.2 seconds long with static camera – consider adding dynamic punch-in zoom").

Batch iteration comparison

Upload up to 4 variations of the same video (for example, testing 4 different hook openings for a TikTok ad) to compare predicted retention curves and choose the winning variant before launching ad spend.