Malaysia's 123rf launches AI video-search tool on AWS

123RF, a Malaysia-based stock media platform, has launched a generative AI-powered video comprehension service on Amazon Web Services (AWS) to enhance search accuracy for its global users.

The new feature analyzes the visual content of videos and images.

Early tests on 5 million videos doubled the accuracy of video descriptors and improved search relevance.

123RF processes over 3 million image uploads monthly and uses AI to automatically detect trademarked logos, categorize licensing, and flag branded content.

The AI technology can identify visual details and match similar images.

123RF reported a 92% reduction in content review time, and said customers now find creative assets in 90% less time.

The platform used Amazon Bedrock and Nova models for these AI capabilities.

🔗 Source: 123RF

🧠 Food for thought

Implications, context, and why it matters.

Why 123RF’s “doubled accuracy” and “90% time savings” claims lack independent validation

The article claims doubled descriptor accuracy and 90% faster asset discovery. It gives no benchmarks or metric definitions and no checks with Shutterstock, Adobe Stock, or Getty Images (large stock media marketplaces) 1.
There is no third party validation or product proof of how 123RF’s multimodal (text-plus-visual) search stacks up against rivals. The claims read like marketing and give buyers no reason to switch.
Independent stock photo reviewers look at image quality, search, and user experience 2. None of the sources reviewed assess 123RF’s AI features, so buyers cannot tell if it offers an edge over current options.

Inference optimization services for Bedrock workloads present a Financial Operations (FinOps) opportunity

123RF is raising expectations for multimodal video comprehension with Amazon Bedrock and Amazon Nova models. Media and digital asset management platforms will add similar AI features, which drives demand for services.
Tech operators plus cloud consultants can offer FinOps with inference optimization (reducing the cost and latency of running models). Amazon Bedrock pricing varies by throughput mode (on-demand vs. provisioned throughput), batch processing (grouping requests), and prompt caching (reusing prior responses) 3. Batch mode can cut spend by up to 50 percent, while caching can reach 90 percent savings 3.
Teams can build skills in rightsizing model selection (choosing the smallest capable model) and refine token usage (the billable units of text or data models process). They can add intelligent prompt routing (automatically sending simpler requests to cheaper models), which can cut costs by 30 percent 4. These steps support recurring revenue as companies run generative AI at scale on AWS.