Malaysia-based royalty-free stock media platform 123RF has unveiled a generative AI-powered video comprehension capability at AWS re:Invent 2025.
The service leverages Amazon Nova models to analyse actual visual content in videos and images, surpassing traditional keyword-based searches for greater accuracy. Early tests on five million videos doubled descriptor precision and enhanced search relevance for its 12.4 million global users.
The AI system identifies specific elements such as green bags in searches, ignoring irrelevant tags such as green backgrounds. It also detects trademarked logos and branded items in subtle details, automatically categorising them for proper licensing like commercial use or royalty-free. Ecommerce customers benefit from faster pairing of visuals with products, boosting daily listings.
123RF reduced content review time by 92 percent for more than three million monthly image uploads, previously handled by up to 40 human reviewers processing 3,000 images daily.
Marketing teams now launch campaigns 35 percent faster. Customers find assets 90 percent quicker across 15 languages, handling cultural nuances beyond simple translation.
Built on Amazon Bedrock with Anthropic’s Claude 3 Haiku for nuanced translation and Amazon Nova Pro/Lite for content understanding, the platform flags copyright risks autonomously.
“Our AWS-powered AI technology now ‘sees’ images the way humans do. This breakthrough helps our customers find exactly what they’re looking for faster, eliminates duplicate content, and identifies potential copyright issues with remarkable accuracy – all because the AI understands the image itself, not just the words describing it,” said Phoebe Liew, CTO of 123RF.
“123RF demonstrates how AI transforms creative workflows – tasks that once took designers days now take minutes, enabling Malaysian companies to compete on a global stage. This has set a new benchmark for what’s possible beyond the stock media industry,” said Hussein Mohd. Ali, Country Manager of Malaysia at AWS.
Entelechy Asia 