[Feature Request/Question] Support for higher-dimension Gemini Embedding node (e.g., gemini-embedding-exp-03-07 @ 3072d)

Questions
,
1

Describe the problem/error/question

I’m building a RAG workflow using Supabase (pgvector) and the Embeddings Google Gemini node. My goal is to use Google’s latest embedding model, gemini-embedding-exp-03-07, for its high-quality 3072-dimension output.

The issue is that the Embeddings Google Gemini node seems to default to, or is limited to, the text-embedding-004 model which only outputs 768-dimension vectors. This causes a dimension mismatch when my Supabase table is correctly configured for vector(3072).

The workflow only succeeds when I downgrade both the Supabase table and the n8n node to use the 768-dimension default. Manually typing models/gemini-embedding-exp-03-07 into the Model field does not resolve the issue, suggesting the node is not handling the 3072d output correctly.

Has anyone found a way to use the 3072-dimension models, or is this a current limitation?

What is the error message (if any)?

When the Supabase table is set to vector(3072), the Supabase Vector Store node fails with the following error, which originates from the dimension mismatch:

Error inserting: vector must have at least 1 dimension 400 Bad Request

Share the output returned by the last node

The Embeddings Google Gemini node returns [empty array] when the Supabase table is expecting 3072 dimensions, even though the upstream nodes (Default Data Loader, Text Splitter) are confirmed to be passing non-empty data.

Information on your n8n setup

  • n8n version:latest
  • Database (default: SQLite): supabase
  • n8n EXECUTIONS_PROCESS setting (default: own, main): own
  • Running n8n via (Docker, npm, n8n cloud, desktop app): docker_google cloud
  • Operating system: linux debian
2

If you need 3072 embeddings right now:

  • Install the n8n-nodes-google-gemini-embeddings-extended fork, or
  • Use an HTTP Request node pointing directly to the Google Generative AI endpoint.

If you’d rather wait for integrated support:

  • Stay tuned for updates to the official node with the “dimensions” option.
  • In the meantime, you can continue using the default 768-dimensional model (text-embedding-004) aligned with your vector store.
1 Like
3

Thanks for the reply! I’ve switched back to using 004 with 768 dimensions, which is working normally for me. I plan to try n8n-nodes-google-gemini-embeddings-extended.

Just wanted to add that the Gemini Embedding 0307 model works fine when querying my 3072-dimension vector store in Supabase using the n8n AI Chat node.

Looking forward to an official update for the Embeddings Google Gemini node!

1 Like