I’d like to suggest adding a simple Image Generation: ON/OFF toggle to individual ChatGPT conversations.
The problem is that uploading an image does not necessarily mean the user wants ChatGPT to generate or edit an image. Often, we simply want ChatGPT to look at a screenshot, photograph, graph, UI, etc. and discuss or analyze it.
At present, conversational instructions such as “Do not generate images in this chat” do not appear to be completely reliable. This matters particularly because image generations can be subject to usage limits. An unsolicited generation can therefore consume a finite resource the user intended to save for later.
I recently had a particularly good example:
I explicitly told ChatGPT several times not to generate images unless I specifically requested one. ChatGPT understood and acknowledged the instruction.
Later, I uploaded a screenshot and asked for its opinion about using that screenshot as a YouTube thumbnail.
Instead of simply discussing the screenshot, ChatGPT generated a new image.
The amusing part is that immediately beforehand, ChatGPT had actually congratulated itself for successfully analyzing another visual without unnecessarily generating anything.
A hard per-chat permission would solve this completely.
Suggested behavior:
When Image Generation = OFF, ChatGPT can still view, analyze and discuss uploaded images normally, but it cannot invoke the image-generation tool.
If the user subsequently asks ChatGPT to create or edit an image, ChatGPT could simply say something like:
“Image generation is disabled for this conversation. Would you like to enable it?”
The user could then enable it with one click.
This would be preferable to relying exclusively on conversational instructions because it gives users deterministic control over a quota-limited tool. It could potentially be extended to other expensive or specialized tools in the future as well.
The ideal implementation could be as simple as a small toggle in the chat/tool bar:
Image Generation: ON / OFF
This would be especially useful for conversations involving lots of screenshots or photographs where the user wants visual analysis but has no intention of generating new images.
Thanks for considering it.