Intermittent Errors with gpt-image-1 in LibreChat vs Azure Playground

[gitanna]

Are there additional content checks or flags applied by LibreChat when using gpt-image-1, compared to running the same prompt directly in the Azure playground?

I’m seeing intermittent failures in LibreChat for image prompts that consistently succeed in Azure. Are there known differences in how prompts are handled, or are there other triggers (e.g. session state, rate of requests) that could affect stability?

Also, are there any configurable parameters or recommended strategies to improve consistency with gpt-image-1 image generation in LibreChat?

[danny-avila]

The only intermittent errors I have with Azure are related to safety/moderation/copyright, perhaps there is something on Azure Portal side to help with this, something the playground doesn't include?

This is especially evident when using gpt-image-1 when served by OpenAI as opposed to Azure.

[gitanna]

Thanks, Danny. Just to clarify—there’s a prompt that consistently fails in LibreChat using Azure gpt-image-1 as the backend, but works reliably in the Azure Playground.

Here’s the original prompt that fails in LibreChat, but works in the Azure Playground:
“Create a photorealistic image of a nuclear submarine docked at a pier. In the foreground, four female officers of different ranks stand in casual formation, smiling for the camera.”

A simplified version succeeds in both environments:
“Create a photorealistic image of a nuclear submarine docked at a pier.”