#Text-only request with image attachment incorrectly triggers image-generation UI state

8 messages · Page 1 of 1 (latest)

restive ospreyBOT
#

Reported by @mellow shore

Bug Report: Text-only request with image attachment incorrectly triggers image-generation UI state
`Steps to Reproduce`
  1. Open ChatGPT with GPT-5.5.
  2. Attach a reference image or screenshot.
  3. Ask for a text-only response based on the image.
  4. Explicitly state that no image should be generated, for example:
    "Create a prompt for GPT Image based on this reference image. Output: text prompt only. Do not create an image in this chat."
  5. Observe the assistant/UI behavior after sending the message.
`Expected Result`

ChatGPT should answer with a normal text response only.

If the user explicitly asks for a text prompt and explicitly says not to generate an image, the system should not enter an image-generation-like UI state. The response should be treated as a text-only vision/reasoning task.

`Actual Result`

ChatGPT sometimes shows an image-generation-style loading state even though the request is explicitly text-only.
The UI displays messages such as: "Thinking", "A more detailed image is being created — one moment."
A large image-generation-style placeholder/loading card is shown. In my observed cases, no final image was generated, but the UI strongly suggests that image generation has started. This is confusing because the user expects a text answer, not an image-generation workflow.

`Environment`

Web: - OS: Windows 10 Pro, Version 22H2, OS Build 19045.7291 - Browser: Google Chrome 148.0.7778.179, Official Build, 64-bit Android: - Device: Google Pixel 8a - OS: Android 16, Security update: May 5, 2026 - App: ChatGPT for Android 1.2026.125 (19), latest official version iOS: - Device: iPhone 16 Plus - OS: iOS 26.5 - App: ChatGPT for iOS 1.2026.132 (26051406691), latest official version Observed primarily on: - ChatGPT Web UI - Model: GPT-5.5

#
Additional Information

Please provide relevant details to help resolve the issue, such as:

  • ChatGPT Shared Link (if applicable).
  • Screenshots or videos demonstrating the problem.

-# ➜ Need to contact support? Visit the OpenAI Help Center.

mellow shore
mellow shore
mellow shore
mellow shore
#

PLEASE WORK ON THIS.

Additional context:

A major part of the issue is that simply attaching an image seems to heavily bias the system toward image generation.

In most of my workflows, I’m uploading an image because I want analysis, prompt writing, style breakdowns, scene descriptions, recreation prompts, or other text-based outputs. Despite explicitly asking for text, the system sometimes starts an image generation workflow instead.

The attached image should not override the user’s actual instruction. If the request is clearly asking for text output, the assistant should prioritize that and avoid triggering image generation unless it is explicitly requested.

#

Additional feedback:

Today I encountered a much more serious version of this issue. Normally the unwanted image generation is just annoying, but in this case it appears to have overwritten a valid text response.

Observed sequence:

  1. User attaches an image.
  2. Assistant incorrectly starts image generation.
  3. User cancels the generation.
  4. User explicitly clarifies that a text response is desired.
  5. Assistant generates the requested text response.
  6. User temporarily switches to another app.
  7. User returns to the chat and discovers that an image has appeared instead, while the previously generated text response is gone.
  8. User force-closes and restarts the app to rule out a temporary caching issue, but the problem persists.

[1.2026.139 (26320156439)]

The concerning part is not just the unwanted image generation itself, but the apparent replacement of an already generated text response. From a user perspective, it looks as if a canceled or background image-generation process later resumed and overwrote the final text output.

If this is intended behavior, it is highly confusing. If it is not intended, it may indicate a synchronization, state-management, or cancellation-handling bug.

hybrid shore
#

I have been experiencing the same problem! It's been going on for weeks now. However, in my experience, it doesn't just happen when my prompt includes an image, it happens for text-only prompts as well.

It seems to be completely random. I can ask ChatGPT, "What day is it today?" Then for some strange reason, it will start generating an image!

After a while, it seems to self-correct, cancel the image generation, and produce a text-based response instead. It takes a while though and it's very annoying!

Hope this gets fixed soon. It's one of those bugs that happens just often enough to be very annoying but infrequently enough that most people probably won't report it.