Crop photos embedded with Voyage and saved to Qdrant

Crop photos embedded with Voyage and saved to Qdrant

An n8n workflow you start by hand. It lists the crop photos in a Google Cloud Storage bucket, turns each one into an embedding with Voyage AI and saves the vectors in a Qdrant collection, four photos at a time.

What is inside

• workflow.json: the workflow as n8n exports it, twenty-five nodes. Thirteen do the work and twelve are sticky notes that explain the steps. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• The step that splits the photos into batches and gives each one a UUID was a Python Code node. In n8n 2, native Python offers only _items and _item, has no _input, and blocks standard library imports such as uuid unless the instance allows them (n8n's docs, read on 12 November 2026). Mindket's copy does the same job in JavaScript.
• The Qdrant cluster address and the bucket name belonged to the original author. They are now the placeholders <YOUR_QDRANT_CLUSTER_URL> and <YOUR_BUCKET_NAME>.
• Several nodes were linked to credentials in the author's n8n. The links are gone, so select your own.
• Nothing else. The tomato filter, the batch size and the sticky notes are the file's own.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• A Qdrant cluster, on Qdrant Cloud or your own server, with its address and an API key.
• A Voyage AI API key.
• A Google Cloud Storage bucket whose pictures anyone can read without signing in, sorted in folders by class, such as agricultural-crops/cucumber/photo-1.jpg. Voyage downloads every picture from its address.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add a Qdrant API credential with your cluster address and key. Select it in Check Qdrant Collection Existence, Create Qdrant Collection, Payload index on crop_name and Batch Upload to Qdrant.
3. Add a Header Auth credential for Voyage. Name it Authorization and give it the value Bearer <YOUR_VOYAGE_KEY> with your own key. Select it in Embed crop image.
4. Add a Google Cloud Storage credential and select it in Google Cloud Storage.
5. In that node, replace <YOUR_BUCKET_NAME> with your bucket. Set prefix to the folder that holds the class folders. The file uses agricultural-crops.
6. Open Qdrant cluster variables and replace <YOUR_QDRANT_CLUSTER_URL> with your cluster address, with https:// in front and no slash at the end. Check the collection name, the batch size of 4 and the vector size of 1024.
7. Look at Filtering out tomato to test anomalies. It skips the tomato folder on purpose. Delete the node or edit it if you want every class.
8. Press Execute workflow.

How a run goes

1. The manual trigger starts the run when you press Execute workflow, and Qdrant cluster variables sets the cluster address, the collection name, the vector size and the batch size.
2. Check Qdrant Collection Existence (HTTP Request 4.2) asks Qdrant whether the collection exists, and If collection exists reads the answer.
3. If it does not exist, Create Qdrant Collection makes it with one named vector, voyage, compared by Cosine distance. Payload index on crop_name then adds a keyword index on the crop_name field.
4. Google Cloud Storage lists every object under the prefix. Get fields for Qdrant builds a publicLink for each photo and takes the cropName from the folder that holds it.
5. Filtering out tomato to test anomalies drops the tomato photos.
6. A Code node groups the photos into batches of batchSize and gives each photo a UUID, because Qdrant does not make point ids. Batches in the API's format reshapes each batch for Voyage and for Qdrant.
7. Embed crop image sends a batch to Voyage's multimodal embeddings API with the model voyage-multimodal-3 and input_type set to document. Batch Upload to Qdrant saves the batch with its ids, vectors and payloads.

Change the result

• Bucket and folder: bucketName and prefix in Google Cloud Storage.
• Collection, batch size and vector size: the values in Qdrant cluster variables.
• Skipped folders: the condition in Filtering out tomato to test anomalies.
• Fields saved with each photo: batchPayloadQdrant in Batches in the API's format.
• Embedding model: model in Embed crop image. Change the vector size to match and use a new collection.

Good to know

• Run it once for each collection. The UUIDs are random on every run, so a second run saves every photo again as new points.
• Voyage fetches each picture from its address, so the files must be public. Voyage's docs, read on 12 November 2026, say a picture must be a PNG, JPEG, WEBP or GIF, at most 20 MB and 16 million pixels. Its address must also answer with a content length and respect robots.txt.
• The file uses voyage-multimodal-3, which those docs still list as a recommended model.
• With the batch size of 4, one hundred photos make 25 calls to Voyage and 25 to Qdrant.
• Voyage bills the embeddings to your key, and Qdrant keeps the vectors in your cluster.
• The public address of each photo goes to Voyage, and the address and crop name go to Qdrant.
• The file holds no credential and no key.