Vertex AI - 429 Quota Exceeded on Imagen Models Despite Approved Quota in Console

Google Cloud Build with AI Generative AI & Foundational Models
,
1

Hello everyone,

I’m hoping to get some help with a very strange issue on a new project. My application is completely blocked by a Vertex AI quota error, even though the console shows I have an approved, non-zero quota. I did contact google and requested an increase and it was approved

image
image1924×188 36.1 KB

Application Architecture:

  • Backend: A Node.js Express server running on Cloud Run in europe-west2.

  • AI SDKs: Using the official (PII Removed by Staff) and (PII Removed by Staff) libraries.

  • Functionality:

    1. The service successfully generates text and JSON using the gemini-1.5-flash model. This works perfectly.

    2. The service then attempts to generate an image using an Imagen model. This step always fails.

The Problem:

Every API call to any available Imagen model (imagen-3.0-generate-002, imagen-3.0-fast-generate-001, etc.) fails with a 429 RESOURCE_EXHAUSTED (Quota Exceeded) error on the very first request.

However, the Quotas page in the Google Cloud Console clearly shows an approved, non-zero quota for these models. For example, base_model:imagen-3.0-generate has an approved limit of 5 requests/minute, and base_model:imagen-3.0-fast-generate has a limit of 20 requests/minute.

Summary of Our Troubleshooting (What We’ve Proven):

We have spent a long time debugging this and have successfully eliminated all common causes:

  1. It is not a code issue: The error happens even with a minimal, standalone test script that makes a single API call.

  2. It is not a model availability issue: We have confirmed via the interactive Vertex AI Studio UI that models like imagen-3.0-generate-002, imagen-3.0-fast-generate-001 are available and work for my account in the europe-west2 region. The API error is 429, not 404.

  3. It is not a permissions (IAM) issue: The service account running the code (firebase-deploy-service-accoun **at** mats4u-genai-app.iam.gserviceaccount.com) has the Editor role on the project, which includes all necessary permissions like aiplatform.user.

  4. It is not a billing issue: Text generation with Gemini models is working and being billed correctly.

  5. It is not an API enablement issue: The aiplatform.googleapis.com API is enabled.

  6. It is not a concurrency issue: The error occurs even when the application is architected to make only a single image request.

Has anyone else experienced a situation where the Quotas page in the console shows an approved limit, but the Vertex AI API behaves as if the quota is zero? This seems to be a project-specific quota enforcement or synchronization bug.

1 Like
4

Yes, same here. I increased the quota limit to 10, but i still get error 429 on first request. It’s Imagen issue and not client side error. I am using Imagen 002 for my query. It hasn’t worked at all.

5

Hi there,

I realize it’s been some time since you posted this, but I wanted to check if you managed to find a solution?

I am currently facing the exact same issue with the imagen-3.0-capability model. My Cloud Console clearly shows an approved, non-zero quota (limit: 60) for my region, yet the API fails with a 429 RESOURCE_EXHAUSTED error on the very first request.

I have already contacted Google Support, but I haven’t received a working solution from them yet (standard troubleshooting steps like backoff strategies didn’t help).

If you discovered what the root cause was or how to fix it, I would really appreciate a hint!