How AffiliateNetwork Reviews Thousands of Creator Posts in Seconds, Not Weeks

AffiliateNetwork connects leading consumer apps and brands with 60,000+ creators producing short-form organic marketing ads, processing thousands of creator videos and millions of views daily.

시각, 오디오, 대사, 모션 데이터를 36개 언어에 걸쳐 통합 처리하는 영상 임베딩 모델. 프로덕션 검색에 바로 사용할 수 있는 단일 512차원 벡터를 반환합니다.

No headings found on page

The Problem

AffiliateNetwork connects leading consumer apps and brands with 60,000+ creators producing short-form organic marketing ads, processing thousands of creator videos and millions of views daily.

The challenge? Verifying creator content meets brand guidelines.

UGC ads aren't movies. They're 10 to 45 seconds of creativity: quick cuts, overlays, voiceovers, memes, flashing logos. Traditional general-purpose video AI pipelines were not built for high-tempo UGC, multimodal overlays, and fast iteration cycles, and simply couldn't keep up.

"As our platform scaled to tens of thousands of posts per day, we realized our existing AI content-review pipeline was not delivering the throughput or video-understanding accuracy we needed. We knew we shouldn't accept the status quo and looked for a better solution."
— Sean Kim, Head of AI & ML at AffiliateNetwork

What Failed

General-Purpose AI Models: frame-by-frame vision + audio transcription

Too slow (minutes vs. seconds)
Missed logos, context, scene changes
Couldn't fully grasp multimodal content
No fast iteration

The issue: Video treated as images + audio, not a unified whole.

Why TwelveLabs

Sean researched TwelveLabs' architecture and found something different: video treated as video, not stacked frames.

Pegasus: Temporal reasoning across videos with low latency, localizing events and actions, enabled the team to analyze creator content for key themes, topics, and trends. Finding specific details, objects, and themes allowed the team to significantly streamline the review process.

"What stood out was that TwelveLabs actually understands video as a unified fusion of media, rather than treating it as separate modalities."

The Solution: AI Post Verifier in 4 Steps

Index once → Marengo creates embeddings in shared vector space
Natural language rules → "Show logo for 2+ seconds" or "must show app in a positive light."
Instant results → Pass/fail, timestamps, explanations
Real-time iteration → Campaign managers refine rules in seconds