Workflow using Google Vertex API (Veo 3.0 Fast Generate) stops without error when triggered from app

đź’¬ Workflows
, ,
1

  1. My goal: We’ve built a Retool workflow that uses the Google Vertex AI API (Veo 3.0 Fast Generate model) to generate videos from multiple images. The goal is to trigger this workflow from a Retool app so that users can generate videos directly through the app interface.

  2. Issue: When the workflow is triggered from the app, it appears to hang and eventually stops without returning any error or result to the app.

    • The workflow stops executing partway through — it doesn’t complete all nodes.

    • There’s no visible error message or timeout message in the app or workflow logs.

    • When we run the same workflow manually (executing nodes one by one in the workflow editor), everything completes successfully.

    We’re aware that Retool apps have a 15-minute timeout, but this failure occurs well before reaching that limit (typically within a few minutes).

    The issue happens consistently when generating videos from more than 3 images, but smaller workloads complete without any issues.

  3. Steps I've taken to troubleshoot:

    1. Verified the Google Vertex AI API is working correctly — all API calls succeed when tested outside Retool.
    2. Tested the same workflow manually in Retool (executing each node in sequence) — all nodes complete successfully.
    3. Checked Retool logs and workflow run history — no explicit errors or timeouts appear.
    4. Tried reducing workload size — works fine with fewer images, but fails silently when the workload increases.
    5. Confirmed the workflow is within the 15-minute app timeout limit.

  4. Additional info: (Cloud or Self-hosted, Screenshots)

    1. Environment: [Cloud-hosted]
    2. Workflow type: Workflow triggered via “Run workflow” block in Retool app
1 Like
3

Hi @Jenna_Hilden,

Thank you for the detailed reporting of this issue and the steps you have tested out!

How are you getting the image data into the workflow when you run it manually?

The team and I were looking at your org's logs in Datadog and it seems the payload is too big for workflows on the cloud. It looks like one manual run recently errored out with 143MB of data as the payload.

It seems you are hitting max body size limits. Which is about 100MB. I am surprised that it is failing silently and not throwing an error in the app, or in the workflow run logs :thinking:

4

Hi! @Jenna_Hilden
This issue is likely happening because the generated video is being returned directly to Retool as a binary payload. Large binary responses can exceed memory limits in Retool workflows and cause them to terminate silently—especially when the output size grows with more than 3 images.

For your workflow first upload pictures to cloudnery than use retrive in string format after that give those string to ai model and than genrated video should be uploaded to clodnery .reterive the string from cloudnery use retool email block and send it to you self,now You would be able to see the ai genrated video from your mobile over internet

1 Like
5

Hi @Jenna_Hilden,

Let me know if the above suggestion was helpful! You may be running into some limitations on the file size for downloading the videos back to Retool.

In which case you may need to test out using strings or other data formats to capture the video data to transfer and then transform back, or using a third party service to host the video data and sending something such as a link to download or access the video from that intermediate video holder.