Suggestions for Improving ChatGPT’s Image Generation and Free-Plan Experience
I would like to offer some constructive feedback about ChatGPT, particularly regarding image generation and how limited image generations are handled on the Free plan.
I understand that AI is not perfect, and I don’t expect it to be. In fact, I think expecting AI to be completely error-free is unrealistic. AI is created by people, trained using human-created information, language, documentation, images, and other material, and built through human-designed systems. Mistakes are therefore inevitable—and I think that is completely okay.
What matters is how the system handles those mistakes.
1. Make Image Allowances Completely Transparent
It would be extremely helpful if ChatGPT clearly displayed:
-
How many image generations a user has remaining.
-
What actually counts as a generation.
-
What the Free plan’s image-generation limits are.
-
When the allowance resets.
-
Whether a failed generation counts against the allowance.
Users shouldn’t have to guess how many attempts they have left or wonder whether an error consumed one.
I appreciate that ChatGPT has started providing warnings about image limits. I think expanding that transparency would make the experience even better.
2. Don’t Penalize Users for Failed Generations
If an image generation produces an error, fails to display, produces no usable image, or otherwise clearly fails because of a system or AI problem, that attempt ideally should not consume the user’s limited generation.
This is especially important on the Free plan.
There is a major difference between:
“The user generated an image and decided they didn’t like it.”
and:
“The system failed to generate or deliver the image correctly.”
Those situations should not necessarily be treated as the same thing.
If a user never received a usable image because of an AI or system error, it feels unfair for them to lose one of their limited opportunities.
3. Allow Confirmation Before Complicated Image Generations
For complicated requests, I think ChatGPT should sometimes briefly confirm its understanding before using a limited image generation.
For example:
“Let me make sure I understand: You want a portrait-oriented image with X, Y, and Z, while keeping A and B unchanged. The people should remain in their original positions, and the background should be changed to ____. Is that correct?”
This would be especially helpful when a request contains:
-
Multiple people
-
Specific dimensions or orientations
-
Lots of text
-
Specific colors
-
Numerous objects
-
Detailed positioning
-
Photo-editing requirements
-
Elements that must remain unchanged
Giving the user an opportunity to correct the AI’s interpretation before consuming a limited generation could prevent a lot of frustration.
4. Explain the Visual Interpretation, Not Just Repeat the Request
AI should do more than simply repeat the user’s words.
It should briefly translate the request into what it believes it is actually going to create.
For example, if someone says they want a portrait image, the AI could confirm:
“I’ll create a portrait-oriented image, with the subject centered vertically, rather than a landscape composition.”
That gives the user an opportunity to say:
“Wait—that’s not what I meant.”
before the image is generated.
This could be particularly valuable for detailed image requests where one misunderstood specification can make the entire generation unusable.
5. Distinguish AI Mistakes From User Changes
I think this is one of the most important improvements.
There needs to be a clear distinction between:
A user changed their mind:
-
“I don’t like that color.”
-
“Can you make the dinosaur larger?”
-
“Actually, I want landscape instead.”
-
“Can you try a completely different style?”
versus:
The AI or system made a mistake:
-
The requested image orientation was ignored.
-
Important requested details were omitted.
-
Text was incorrectly spelled.
-
The image failed to generate.
-
The image did not display.
-
The AI misunderstood a clearly stated requirement.
-
A person in a photo was substantially changed when the user specifically asked for them to remain unchanged.
Those situations feel fundamentally different when image generations are limited.
6. Image Quality and Accuracy Matter When Users Are Deciding Whether to Pay
I also think this affects whether people feel that a paid subscription is worthwhile.
If someone repeatedly experiences:
-
An image-generation error,
-
A generation that doesn’t follow the instructions,
-
A generation with incorrect spelling,
-
A generation that changes important details,
-
Or a generation that becomes progressively worse after revisions,
then it can become difficult to justify paying for additional usage.
I understand that mistakes happen. I don’t expect perfection.
But if a user is paying for a service, they reasonably want to feel that the system is reliable enough to justify the cost.
The more transparent and fair the system is when mistakes happen, the more confidence users may have in continuing to use—and eventually pay for—the service.
7. I Don’t Expect AI to Be Perfect—and I Don’t Want It to Be
This is important to me.
I don’t think AI needs to be perfect.
I actually think expecting AI to never make mistakes is unrealistic. AI is a human-created technology, built from human-designed systems and trained using enormous amounts of human-created information.
Of course there will be mistakes.
That’s okay.
The goal shouldn’t necessarily be:
“Make AI incapable of making mistakes.”
Instead, I think the goal should be:
“Make AI better at understanding users, make its limitations transparent, and make the consequences of its mistakes fair.”
If AI makes a mistake, users should be able to understand what happened, correct it, and continue without feeling that they were punished for something they didn’t cause.
Final Thoughts
I appreciate being able to use the Free plan and having the opportunity to try these technologies without immediately paying for them. I think that is valuable.
My feedback is ultimately about making that experience better.
If ChatGPT clearly tells users what their limits are, confirms complicated requests when appropriate, accurately communicates what it is about to create, and doesn’t charge users’ limited generations for obvious system failures, I think users would have much more confidence in the product.
AI will make mistakes.
That’s not the problem.
The important thing is making the system transparent enough that users understand its limitations—and fair enough that users aren’t penalized when the system itself is responsible for the failure.
Thank you for taking the time to consider this feedback.