Creative audit API for ad-tech

What creative audit actually has to catch

• Reused creatives across advertisers — The same image or video shows up under different campaign IDs and advertiser accounts, often with a re-encode or watermark.
• Recycled campaign assets — Old creatives quietly resurface in new campaigns with cropped frames, swapped logos, or new audio.
• Brand-safety review at scale — Reviewers triage thousands of placements per day. Without similarity grouping, near-duplicates get reviewed independently and decisions diverge.
• Audio bumpers and music reuse — Branded audio, jingles, and music beds get reused across spots. Audio fingerprinting flags overlap with offset-aligned segments instead of guessing.

The matching primitives

{
  "match": true,
  "confidence": "high",
  "similarity_score": 0.91,
  "processing_time_ms": 1640,
  "media_type": "video",
  "matched_segments": [
    { "source_start": 0.0, "source_end": 14.8, "target_start": 2.3, "target_end": 17.1, "score": 0.93 }
  ]
}

A practical creative-audit pipeline

import requests

API_HOST = "medialayer-image-audio-video-matching-api.p.rapidapi.com"
HEADERS = {
    "x-rapidapi-key": "YOUR_RAPIDAPI_KEY",
    "x-rapidapi-host": API_HOST,
    "Content-Type": "application/json",
}

PATH_BY_TYPE = {
    "image": "/image/match",
    "video": "/video/match",
    "audio": "/audio/match",
}

def audit_creative(media_type: str, placement_url: str, references: list[str]):
    path = PATH_BY_TYPE[media_type]
    best_score = 0.0
    best_ref = None

    for ref_url in references:
        r = requests.post(
            f"https://{API_HOST}{path}",
            json={"source_url": placement_url, "target_url": ref_url},
            headers=HEADERS,
            timeout=60,
        )
        r.raise_for_status()
        data = r.json()
        if data["similarity_score"] > best_score:
            best_score = data["similarity_score"]
            best_ref = ref_url

    if best_score >= 0.95:
        lane = "auto-block"
    elif best_score >= 0.80:
        lane = "review"
    else:
        lane = "pass"

    return {"lane": lane, "best_ref": best_ref, "score": best_score}

Cluster before you review

When a creative library outgrows pairwise

Production checklist

• Server-side keys — x-rapidapi-key never belongs in browser or mobile clients. Call from the audit service.
• Public reference URLs — URL validation rejects private, loopback, and cloud-metadata addresses. Reference creatives need to be publicly reachable, or they need to live on the enterprise direct surface with private deployment.
• Bound concurrency — Bound the per-placement fan-out so a single placement doesn't burst against your RapidAPI plan's per-second limit.
• Use overlap duration on video and audio — Don't act on the boolean match alone — use matched_segments to get total overlap seconds and route based on duration as well as score.