AI Applications in iQIYI Video Advertising: Scene Generation, Video Understanding, and Advertising Placement

The presentation, delivered by iQIYI scientist Liu Qiyue, introduces the use of artificial intelligence in video advertising, focusing on how AI can understand and process video content to create valuable advertising opportunities.

Background : Advertising aims to drive consumer actions, ranging from long‑term brand awareness to short‑term performance goals. Effective ads must match user needs and convey product value within appropriate contexts.

Scene Generation / Recommendation : AI identifies suitable "ad slots" (points) within video scenes, such as product‑related moments, and generates contextual ad creatives (e.g., "creative patches") that align with the storyline, enhancing relevance and conversion potential.

Video Understanding : The system decomposes video into a triple of object‑scene‑action . It recognizes objects (people, pets, items), events (behaviors, higher‑level semantics like weddings), and scenes (locations, ambiance), producing dense multi‑dimensional tags for each clip.

Video Graph : By linking low‑level tags through semantic relationships (synonyms, hierarchy, exclusivity), the platform builds a richer video graph that distinguishes commercial‑value tags from creative‑value tags, improving tag precision and ranking.

Ad Placement Workflow (VideoIn) : The workflow outlines brand, production, and technical requirements for post‑production ad insertion, including space, visibility, and non‑conflict with competitors, and describes how AI automates point selection to meet timeliness and richness demands.

Ad Formats : Various formats are discussed—pre‑, mid‑, and post‑roll patches; floating overlays such as creative patches and pre‑summary stickers; and product placement (pre‑ and post‑production). Each format has different scalability and content‑dependency characteristics.

AI‑Assisted Marketing : AI provides content analysis and creative generation (e.g., image‑to‑video, cover selection, title generation), enabling scalable, context‑aware advertising across video platforms, streaming services, and even offline video‑like environments.

Overall, the talk demonstrates how AI‑driven video analysis and knowledge graph construction empower advertisers to discover high‑value, scene‑aware ad slots, automate creative production, and achieve more effective, scalable video advertising.