Media libraries automation: AI-powered tagging and duplicate detection

GetDevDone built an AI-powered automated layer that tags every incoming asset, catches near-duplicates before they waste budget, and flags the right risks before anything enters a live campaign. The creative team keeps full control. 

Our client, a performance marketing agency, manages paid media for eCommerce, D2C, and retail brands across Meta, Google, and TikTok, supported by high-volume creative production of images and short-form videos.

Business challenge

With 300+ assets produced monthly across 15 client accounts, the agency’s creative output had outgrown its library. Governance hadn’t kept pace with volume, and the compounding effect was starting to hurt operations:

• Tags were inconsistent at best, skipped entirely under deadline pressure, making the library progressively harder to search.
• Near-identical assets were uploaded repeatedly with no duplicate detection, wasting production spend.
• Rights and stock restriction checks were unreliable, leaving the team one compliance issue away from a serious client problem.

The agency needed a solution that would enforce order on the way in without slowing down production or removing human judgment from the decisions that mattered.

Solution delivered

GetDevDone built an AI-powered review layer that sits between the upload step and the media library. Every incoming asset passes through an automated pipeline — tagged, checked for duplicates, and reviewed for rights risks — before it reaches the campaign workflow.

Key features

AI auto-tagging. Each creative is automatically tagged on arrival (labels, on-image text, logos, colors, and scene type), normalized to the agency’s existing naming conventions so results plug directly into campaign briefs.

Near-duplicate detection. Perceptual hashing catches re-uploads that have been cropped or reformatted. A configurable threshold lets the team set where “likely duplicate” ends and “worth a second look” begins.

Reverse-image checks. The assets are checked against the web at upload, surfacing stock restriction violations, competitor usage, or rights issues before they reach a live campaign.

Quarantine queue. Any asset triggering a flag is held until a reviewer clears it. Nothing moves forward without a human decision, built in as a hard requirement from day one.

Searchable library. Auto-generated metadata powers instant search and filtering by tag, color, scene type, upload date, and status.

Admin controls. The team has full visibility into every automated decision, with adjustable thresholds and a full audit log for accountability.

Technologies & tools

• Node.js / TypeScript (Cloud Functions for Firebase): event-driven backend
• Google Cloud Vision API: labels, logos, SafeSearch, web detection
• Cloud Storage: asset ingestion
• Firestore: metadata, tag index, and duplicate registry
• Perceptual hashing library (pHash / blockhash): near-duplicate detection
• React (admin UI): tag review, thresholds, quarantine queue

Business outcomes

Delivered in 6 weeks on budget, the solution integrated smoothly into the agency’s existing stack with full audit logs. The library went from a growing source of operational risk to a searchable, governed, compliant hub with mandatory human oversight, no auto-rejections, and enforced review.

Production efficiency

Around 85% of auto-generated tags are accepted without edits, recovering hours of manual tagging every week and redirecting the team’s attention to key creative work that moves campaigns forward.

Protected production spend 

Duplicate detection catches near-identical assets before they enter campaign workflows, reducing the budget drain of re-uploaded creatives and keeping storage clean.

Asset discoverability that scales with the team 

Instant search by tag, color, scene type, and status replaced the habit of tracking down whoever uploaded an asset. The knowledge lives in the system, making it a universal source of truth that the whole team works from.

Rights risks resolved before they reach clients 

Every upload is checked against the web before it gets near a live campaign, surfacing stock restrictions, license over-reuse, and competitor appearances at the point they’re easiest to fix.

Consistent review standards

Every flagged asset clears a mandatory human decision before moving forward. The quarantine queue holds that standard firm, whether the team is under pressure or not.