AI Agent Hub
Back to skills
✍️

Ad Creative Fission

Content Creation Updated 2026.08.29

Paste the following prompt into your AI chat to install this skill:

Please follow https://skillhub.cn/install/skillhub.md to install @user_49271746/ad-creative-fission into your AI assistant.

About this skill

Problem Solved

In short-video ad campaigns, creative performance metrics (CTR, CVR, ROI) are often disconnected from the underlying creative content. Teams typically learn which assets perform well but lack attribution for which variables drive conversion, forcing subsequent production to rely on trial and error. This tool bridges that gap by linking ad performance data with multimodal video understanding to identify the creative structures that actually drive conversions and batch-generate actionable script variations.

How It Works

  • Automated Data Retrieval: Fetches asset video URLs and performance metrics via the tencent-ads-assistant API, eliminating the need for manual Excel/CSV exports.
  • Multimodal Copy Extraction: Automatically downloads videos, extracts frames, and uses vision models to parse fluent copy, hook types, and audience targeting into a structured videos_multimodal_fixed.csv.
  • Attribution & Fission: Treats assets as variable combinations (emotional hooks, pacing, visual anchors, scenes, characters) and cross-references them with CTR×CVR data to isolate Top/Bottom combinations. It then uses high-converting "parent" assets to batch-generate P0/P1/P2 script variations across four dimensions: hook angles, audience variants, benefit restructuring, and text-card slogans.
  • Dynamic Video Analysis: Uses opencv to extract programmatic features like scene transitions, motion amplitude, and color profiles. Paired with temporal frame extraction and multimodal segment validation, it reconstructs the visual narrative chain ("how it was shot and why it works") and renders a fully static HTML report.

Scope & Limitations

This tool focuses strictly on creative analysis and script generation; it does not execute ad placements or modify platform data. Attribution conclusions require sufficient sample sizes, and low-volume scenarios must be explicitly flagged as "insufficient sample." Findings reflect correlations and empirical patterns rather than direct causality. HTML reports use static rendering to avoid security restrictions under the file:// protocol.

Use Cases

  • Education or local-life ad teams need to attribute which hooks, audiences, and benefits drive CTR/CVR and spend.
  • Creative optimizers need to batch-generate P0/P1/P2 script variants from high-converting parent assets for production.
  • Ad analysts need to compare conversion performance across outdoor live-action, white-background, text-card, and multi-scene edits.
  • Campaign reviewers need to identify high-click, low-conversion assets and determine whether they attracted non-target users.

Best For

  • Education, e-commerce, or local-life ad operations staff reviewing short-video campaign assets weekly.
  • Creative planners or content strategists turning high-converting assets into shootable scripts.
  • Ad analysts managing creative A/B testing, placement, and rollout schedules.
  • Team leads using real spend data to maintain a shared creative label library and workflow.