Temporal Frame Deconstruction: How 0.5s Video Splitting Uncovers Hidden Meta ROAS Drivers

In modern performance advertising across Meta (Facebook & Instagram) and TikTok, media buyers are accustomed to looking at aggregate asset-level metrics: blended Click-Through Rate (CTR), Cost Per Click (CPC), and Purchase ROAS. However, treating a 30-second video creative as a single, indivisible asset creates a severe analytical blind spot.

A video ad is not a monolithic signal; it is a rapid sequence of discrete audiovisual stimuli occurring across time. When an ad fails or scales, the underlying cause is rarely the entire video—it is almost always localized to a specific scene transition, an unreadable text overlay, or a visual stagnation window that prompts user scroll-away.

AdHelix implements an automated temporal segmentation pipeline that breaks down raw video streams (1080x1920, 9:16 vertical video) into precise 0.5-second frame intervals. Using OpenCV frame extraction combined with multimodal vision models (Google Gemini 2.0 Flash and Meta SAM 2.1), each slice is extracted and tagged across the Universal 8-Pillar Creative Taxonomy.

For each 0.5-second frame, the pipeline computes visual density metrics (bounding box coverage percentage, text overlay area), color temperature (Kelvin scale), optical motion vectors, and speaker gaze coordinates.

Viewer attention does not decay linearly. Grounded in cognitive psychology and empirical eye-tracking research, AdHelix categorizes each video creative into three chronological structural phases:

• Phase 1A: Orienting Hook (0.0s – 1.5s): Governed by Yarbus & Haxby facial gaze processing and the human orienting reflex. At this millisecond threshold, the viewer decides whether to stop scrolling based on pattern interruption, direct eye contact, or high-contrast visual framing.

• Phase 1B: Cognitive Hook (1.5s – 3.0s): The viewer evaluates the initial value proposition or problem claim. This determines the official Meta Hook Rate (3-second video plays divided by impressions).

• Phase 2: Body & Micro-Hook Pacing (3.0s – End-3s): Attention naturally decays exponentially. To maintain retention, the video must introduce micro-hook re-engagement cuts every Delta-t < 2.5 seconds (Teixeira, Harvard Business School).

• Phase 3: CTA & Closing (Last 3.0s): The outbound conversion proposition, value justification, and brand end-card.

Rather than relying solely on Meta's discrete 25%, 50%, 75%, and 100% video completion checkpoints, AdHelix applies the Cox Proportional Hazards Model (Time-to-Event Survival Analysis). By modeling drop-off as a continuous survival curve, statistical power is scaled by 30x across the total duration watched.

This allows performance teams to isolate the exact frame timestamp where hazard rates spike—pinpointing weak cuts, confusing copy transitions, or visual fatigue.