Image Metadata

Metadata lets you attach custom key-value pairs to images in your Roboflow workspace. Use metadata to store structured information alongside your images - such as capture conditions, device identifiers, quality scores, or any domain-specific attributes - and then search, filter, and organize your data around those attributes.

Overview

Each image can hold any number of metadata entries. An entry is a key (a name like camera_id) paired with a value (a string, number, or boolean).

Use cases

• Capture context - record camera ID, GPS coordinates, weather, lighting conditions
• Quality tracking - attach confidence scores, review status, annotator IDs
• Data slicing - filter your dataset by any attribute to build targeted training sets
• External system linking - store identifiers that connect images back to your internal tools

Adding Metadata

You can add metadata to images through the web UI, the Python SDK, the REST API, or automatically via S3 Bucket Mirror.

Web Application

Values are automatically parsed by type:

Python SDK

Pass a metadata dictionary when uploading an image:

import roboflow

rf = roboflow.Roboflow(api_key="YOUR_API_KEY")
project = rf.workspace("your-workspace").project("your-project")

project.upload(
    image_path="image.jpg",
    metadata={
        "camera_id": "cam001",
        "location": "warehouse-3",
        "temperature": 72.5,
        "is_night": False
    }
)

CLI

Use the roboflow image metadata command to update metadata and tags on existing images:

# Set metadata on a single image
roboflow image metadata <image_id> -m '{"camera_id": "cam001", "location": "warehouse-3"}'

# Add tags to an image
roboflow image metadata <image_id> --tags "reviewed,v2"

# Remove metadata keys
roboflow image metadata <image_id> --remove-metadata "old_key,deprecated_field"

# Remove tags
roboflow image metadata <image_id> --remove-tags "draft"

# Combine: set metadata, add tags, and remove tags in one call
roboflow image metadata <image_id> -m '{"quality_score": 95}' --tags "reviewed" --remove-tags "pending"

# Batch update multiple images (async)
roboflow image metadata img1,img2,img3 -m '{"batch": "june-2026"}' --tags "processed" --poll

A single image ID updates synchronously. Multiple comma-separated IDs (up to 1,000) use the batch async endpoint. Add --poll to wait for the batch to finish; without it the command returns a taskId you can check later with roboflow asynctasks get <task-id>.

REST API

Add metadata during upload

Include a metadata field (JSON-stringified) in the multipart form data when uploading an image:

curl -X POST "https://api.roboflow.com/dataset/your-dataset/upload?api_key=$ROBOFLOW_API_KEY" \
  -F "name=image.jpg" \
  -F "split=train" \
  -F "file=@image.jpg" \
  -F 'metadata={"camera_id":"cam001","temperature":72.5}'

S3 Bucket Mirror

When using Datasources to sync images from an S3 bucket, metadata is imported via JSON sidecar files placed alongside each image. See Datasources for sidecar file format, constraints, and update strategies.

MCP Server

Connect your AI agent to the MCP Server and it can tag images and set metadata on them with these tools:

Searching by Metadata

Metadata is indexed and searchable in the Asset Library. Use the search bar to filter images by metadata values:

metadata:camera_id="cam001"
metadata:quality_score>80
metadata:reviewed=true

You can combine metadata filters with other search filters:

metadata:location="warehouse-3" AND class:forklift

The Asset Library also provides autocomplete for metadata keys and values based on what exists in your workspace.

Key Naming Rules

Metadata keys must follow these rules:

Valid keys: camera_id, capture.temperature, _internal_ref, v2_score

Invalid keys: camera/id (contains /), .starts_with_dot (starts with .), has spaces (contains spaces)

Metadata vs. Tags

Both metadata and tags help you organize images, but they serve different purposes:

You can use both on the same image. For example, tag an image as reviewed and also store reviewer: "alice" and confidence: 0.95 as metadata.