Label satellite images fast using existing examples. The flow takes an image link and returns one land type, like forest or beach. Teams that manage maps, drones, or field data can speed up sorting and reporting without training a custom model.

Here is how it works. A calling workflow sends an image URL into an Execute Workflow Trigger. The image is turned into a vector using the Voyage AI multimodal embeddings API. That vector is sent to Qdrant to find the closest images that are already labeled. A code step counts the most common label among the neighbors. If there is a tie, the loop asks Qdrant for more neighbors by increasing the limit by five until the tie breaks or a cap is reached. The final class is returned to the caller. In tests noted in the build, it reached about 93 percent accuracy with no fine tuning.

To use it, you need a Qdrant collection with your labeled images and a Voyage AI API key. Set your Qdrant URL, collection name, and the vector space name to match your setup. The result is faster review, lower manual work, and consistent labels across large image sets. Good for mapping teams, agritech, and inspection workflows that need quick land type tags.

• Accepts an image URL through an Execute Workflow Trigger and passes it through the flow
• Creates an image embedding with the Voyage AI multimodal API using a simple HTTP request
• Stores Qdrant URL, collection name, and neighbor limit in Set nodes for easy tuning
• Queries Qdrant points with the voyage vector space and returns labeled neighbors with payloads
• Counts the most common class using a Python code block for a clear majority vote
• Resolves ties by increasing the neighbor limit by five and looping until a clear winner or a cap
• Outputs the final class to the calling workflow for downstream use

• Reduce manual image review from hours to minutes by returning a label automatically
• Automate up to 90 percent of sorting work using majority vote and a tie breaking loop
• Reach about 93 percent test accuracy reported in notes without custom training
• Handle large batches by increasing the neighbor limit as needed
• Connect Voyage AI embeddings with Qdrant search for consistent results

1. Import the template into n8n: Create a new workflow in n8n > Click the three dots menu > Select 'Import from File' > Choose the downloaded JSON file.
2. You'll need accounts with Voyage AI and Qdrant. See the Tools Required section above for links to create accounts with these services.
3. Create a Voyage AI API key on the official site. In the n8n credentials manager, create a new HTTP Header Auth credential named Voyage API. Add the header Authorization with the value Bearer YOUR_API_KEY and save it. Assign this credential to the Embed image HTTP Request node.
4. In Qdrant Cloud, create a cluster and a collection for your labeled images. Use a vector space named voyage and store a payload field like landscape_name for the class label. Generate your Qdrant API key and note the Cluster URL.
5. In n8n credentials, create a new QdrantApi credential. Enter the Cluster URL and API key. Save it and select it in the Query Qdrant node.
6. Open the Set node named Qdrant variables + embedding + KNN neigbours. Fill in qdrantCloudURL, collectionName, and set an initial limitKNN such as 20. Ensure the using value in the Qdrant request is voyage to match your vector space.
7. Prepare your Qdrant collection by uploading embeddings for your labeled images using the same Voyage model so the vector dimensions match. Include payloads with landscape_name so the vote step can read the label.
8. To test, manually run the workflow and provide input JSON that includes a field query.imageURL with a valid image link. The Image Test URL step maps this into the embedding request.
9. Check the execution. Confirm Qdrant returns points with payloads. Watch the Majority Vote and Check tie steps. If a tie occurs, you will see the limitKNN increase by five until a clear class is found or the cap is hit.
10. Troubleshoot common issues: if you get empty results, verify the collection name and vector space match. If you see 401 errors, recheck Qdrant or Voyage credentials. If labels come back wrong, confirm your payload field name matches the code node or adjust the code to your schema.