n8n Workflow: Object-Based Image Search with Cloudflare AI

Turn one image into searchable object crops using Cloudflare AI + Elasticsearch

This n8n workflow downloads an image, uses Cloudflare Workers AI (DETR ResNet-50) to detect objects, crops only high-confidence detections, uploads those crops to Cloudinary, and indexes the results in Elasticsearch to enable object-based image search.

What this workflow does

• Runs when you manually start the workflow in n8n.
• Downloads the source image from the URL you configure.
• Calls Cloudflare Workers AI (@cf/facebook/detr-resnet-50) to extract object labels, bounding boxes, and confidence scores.
• Splits detection output into one item per detected object and filters to keep only detections with a score ≥ 0.9.
• Re-downloads the original image for each retained detection and crops the object region using the bounding box.
• Uploads each cropped object image to Cloudinary.
• Inserts a document into Elasticsearch containing the Cloudinary URL, the original source URL, the object label, and detection metadata.

Use cases

• Object-based search for product catalogs (e.g., find images by detected items).
• Building visual indexes for SaaS content libraries where users search by what’s inside an image.
• Curating labeled image datasets with high-confidence object crops for downstream review or indexing.

Technical details

• Cloudflare Workers AI: DETR ResNet-50 endpoint for object detection (labels, bounding boxes, scores).
• Cloudinary: uploads cropped object images using configured cloud name, an unsigned upload preset, and query-auth credentials.
• Elasticsearch: indexes image documents into your configured index name.
• n8n nodes include: set, filter, split out, edit image, sticky note, and http request.

Before running, configure Cloudflare Workers AI credentials, Cloudflare Account ID, Cloudinary settings, Elasticsearch credentials/index, and the source image URL (and adjust the object score threshold if needed).