Very slow gemini node

Questions
1

I have been using gemini api for the past month, recently I realsied that the gemini API is extremely not responsive and very slow compared to the past. One node could take up to 12minutes to complete and sometimes even not giving output with errors 503.

Information on your n8n setup

  • n8n version: not sure
  • Database (default: SQLite): not sure
  • n8n EXECUTIONS_PROCESS setting (default: own, main):
  • Running n8n via (Docker, npm, n8n cloud, desktop app): Docker
  • Operating system: Mac
2

Hi @YEE_HAN_AW

If you are on the free tier of AI Studio, this is expected.
You may want to switch to other models

3

Free tier in n8n? Or free tier in Gemini?

4

I meant the free tier in Gemini (Google AIStudio)
The model you are using will also play a part, especially if it is very popular

5

I am on Tier 1 though, does that make a big difference compared to free tier?

6

A 503 error means the model is overloaded at a global level. It is not a reflection of your account status or your quota.

  • Global Demand: During peak hours, the shared infrastructure simply cannot handle the volume of requests.
  • Preview Models: If you are using a “Preview” model (like Gemini 3 Pro or certain 2.5 versions), these are often less stable and more prone to 503s than “GA” (General Availability) models.

You may:

  1. Switch to Gemini Flash: If you are using Gemini Pro, switch to the Flash version. It is significantly faster and much less likely to throw 503 errors.
  2. Check n8n Retry Settings: Look at the node’s “Settings” tab. If “Retry on Fail” is on, reduce the number of retries or increase the delay between them to avoid hammering an already overloaded server.
  3. Reduce Context: Ensure you aren’t sending unnecessary data in the prompt, as larger contexts increase the likelihood of timeouts and 503s.
7

Welcome @YEE_HAN_AW!

One option worth trying alongside kjooleng’s suggestions: switch to the Vertex AI node instead of the Gemini node. It hits Google Cloud’s infrastructure rather than AI Studio, so it tends to be more stable for sustained production workloads. You’ll need a Google Cloud project with Vertex AI API enabled and a service account credential in n8n, but the Gemini models are identical underneath. If you want to stay with AI Studio, gemini-2.0-flash (GA, not a preview version) is currently the most reliable model for consistent throughput.

8

I see, do I need to pay anything for that? I already have credits in Google AI Studio

9

Hi @YEE_HAN_AW

Credits in Google AI Studio cannot be transferred to or used with Vertex AI.
So yes, you will likely need to pay

10

I am currently using the gemini nano banana for editing photos, does vertex ai can so that?

Screenshot 2026-06-24 at 2.48.16 PM
Screenshot 2026-06-24 at 2.48.16 PM4140×536 219 KB

11

Yes. The Gemini Nano Banana models—Nano Banana 2 (Gemini 3.1 Flash Image) and Nano Banana Pro (Gemini 3 Pro Image)—are available on Vertex AI.