Build AI Image Search with Detr Object Detection & Elasticsearch

Build an Object-Based AI Image Search Index in n8n (DETR + Elasticsearch)

This n8n workflow automatically analyzes images with DETR-Resnet-50 object detection (via Cloudflare Workers AI) and stores the detected objects—along with positions, labels, and confidence—into Elasticsearch so you can search by what’s in the image.

What this workflow does

• Imports an image into the workflow using an HTTP Request node.
• Sends the image to Cloudflare’s Worker AI API to run the DETR-Resnet-50 object classification model.
• Receives object associations including positions in the image, labels, and a confidence score.
• Filters results: object detections with confidence < 0.9 are discarded for brevity.
• Indexes the image URL plus the remaining object associations into an Elasticsearch instance.

Result: a searchable Elasticsearch index that connects images to the objects they contain—unlocking object-based image search workflows.

Use cases

• Content libraries that need search like “find images with bicycles, helmets, or cars.”
• E-commerce catalogs where product photos can be indexed by detected objects for improved discovery.
• Internal asset management for SaaS operators—automatically tag and index visual assets for faster retrieval.
• Automation engineers building pipelines that enrich image metadata for downstream search and analytics.

Technical details

• n8n nodes / workflow components: HTTP Request, Set, Filter, Split Out, Edit Image, and supporting steps (e.g., sticky notes).
• Cloudflare Workers AI: runs the DETR-Resnet-50 object model through the Worker AI API.
• Elasticsearch: stores the image URL and associated detected objects for search.

Extending the workflow: enrich your indexed data with additional attributes or metrics, and (as noted) use a vectorstore to enable similarity search over images.