This is directly from ChatGPT after we discussed what had just happened during our session:
Subject: Please don’t count failed image generations against users’ limits
I’d like to offer feedback about the image-generation experience, particularly how failed attempts interact with usage limits.
In a recent conversation, a user and I tried to create a wholesome, non-explicit illustration: a young, handsome Fred Rogers depicted as a patriotic superhero, kneeling to talk to a smiling child. The first attempts were rejected by the image-generation safeguards. When we tried again after adjusting the prompt, the image-generation tool reported that the user’s limit had been reached.
The result was that the user received no images at all, despite multiple attempts.
I recognize that image-generation safeguards serve important purposes, and I’m not asking for those safeguards to be removed. I’m asking for the failure-handling design to be reconsidered.
When an image is not successfully generated, users should not lose the opportunity to generate an image they can actually use. Ideally, unsuccessful attempts would not count against successful-generation quotas. If that’s technically impractical, the system should offer a retry allowance, restore the consumed usage, or clearly distinguish between a content rejection and a usage-limit restriction.
It is particularly frustrating when a user responds constructively to a rejection by revising their prompt, only to discover that they can no longer try at all. That experience makes the system feel arbitrary and punishes users for cooperating with its safeguards.
A good user experience should encourage people to refine requests when necessary, while giving them a fair opportunity to receive the result they’re asking for.
Please consider separating failed generation attempts from successful image-generation usage, and providing clearer explanations and recovery options when generation fails.
Thank you for considering this feedback.