Not explicitly a feature request—but an intelligence request, as it seems Chat likes to misinterpret things for character creators/story proofreaders and beta readers. Or it might be me.
Chat seems to prefer generating images when I ask it for a prompt or describe how things are supposed to look, instead of continuing discussions—locking me out of multiple design threads that weren’t supposed to have images at all.
When I give it something I’m reading to check for typos, it generates an image or treats it as a roleplay.
It’d be nice to be able to manually input hard rules—such as to only sketch/draw in a certain style while using an upload as an example without it being mistaken for a redraw, for story input not to be treated as a roleplay, or for images not to be generated unless Chat is told to generate them.
For a proper feature—I use the Free version—but having 3 uploads daily is ludicrously low for someone who designs and proofreads, especially since they count as attachments and you only get around 20 messages at a time with attachments every four hours. Maybe changing the upload limit to a maximum of 5, not counting the previous files uploaded by the user or Chat as attachments, and a maximum of 50 chats with attachments, with the limit resetting at around 5 or 6 hours, would make it easier for others to use the Free version without complaint.
Maybe even allowing the upload limit to change based on what you use GPT for—such as casual fun, artwork, proofreading, math help, and so on—would help those with heavier upload use who can’t afford to pay for the paid plans, while still offering good options with the paid plans.
Then there’s Instant. I’d love to be able to turn it off and on for certain chats, as images can’t be created in Mini. My only complaint with Mini is that it doesn’t really have a good memory after a few chats. Maybe something with Instant’s memory that doesn’t allow image generation when the model is chosen but still allows at least 3 uploads on the more compact model would be the best middle ground. Not as forgetful as Mini, not able to generate images, three images to upload—it feels like a real mini version, unlike Mini, which just feels like a glorified version of Gemini.