I am reporting a persistent, recurring failure in ChatGPT’s image-generation workflow that affects literally every image-generation task I attempt.
This is not an one off incident, a particularly difficult prompt, or a single failed image gen It is an ongoing pattern across completely unrelated image gen requests.
Despite providing explicit instructions, reference images, annotated corrections, and repeated clarification, the assistant consistently substitutes unrequested elements, ignores clearly defined constraints, changes details that were never requested, and repeatedly generates incorrect results.
More concerning, the assistant frequently demonstrates that it understands the instructions perfectly when discussing them, yet submits image gen requests that directly contradict those same instructions.
The assistant then acknowledges its mistakes, promises corrections, and repeats the exact same failures.
This raises serious questions about instruction adherence, prompt construction, and quality control between the conversational model and image gen system.
I am a paying customer, and this persistent behavior has resulted in substantial wasted time, repeated failed generations, and an inability to complete otherwise straightforward tasks.
I am requesting an actual technical investigation into this recurring behavior, not another generic explanation of image gen limitations.
The issue is not that the image gen occasionally produces imperfect images. The issue is that the GPT repeatedly fails to execute explicit instructions correctly, even when it can accurately restate those instructions moments beforehand.