I am on a Teams account. I am doing character creation and simply writing character descriptions. The GPT automatically begins creating images - even when I elaborated in my prompt that it shouldn’t. I am unable to interrupt it without closing the browser and restarting. I’m on Windows 11 Pro using Brave Browser, but it reproduces in Edge as well.
this is the same problem happening to me. i was just telling it what happend on a day with someone and what was in the conversation, i ask it to analyse that conversation, and the started creating images. i had to press the stop button for so many times to stop it. even if i told it that it should not create any images at all, it would still start creating images from time to time.
This is driving me nuts. Constantly generating an image for no reason. Can OpenAI put some times on this interface once and for all. This interface is a mess. The black theme, the Image generation tool going nuts. Delete chats takes times. Uploading image takes time and I’m on a paid subscription.
I often use ChatGPT for art critique, and this is really starting to get annoying. If you hint at: “Maybe I should change this” or “What do you think about adding X?”, the model assumes it is supposed to edit your image and not just provide feedback.
I have to clearly state “DO NOT GENERATE IMAGES” to stop it from happening, at which point ChatGPT apologises that its image generation was automatically triggered.
They REALLY need to fix this. Currently i start every prompt with “do not make me an image.” it is THE most reliable way of avoiding the erroneous image generation trigger, it even still makes an image from time-to-time but not NEARLY as much as it does when you put that negative prompt anywhere else in the prompt. You have to start it off with that line.
Can confirm.
I have a custom instruction, telling the GPT to “define” a subject in three lines of plain text.
But when I ask “define an elephant” or “define a printer” it is creating an image of the text, that defines the subject.
So the result is an image, containing three lines of text.
When I ask him to “output that image as plain text” it spits out the text in the image as real text.
Very annoying. I ask a simple question; GPT starts creating images, not answers.
I already updated my instruction set, but that doesn’t work.
<NEVER_CREATE_IMAGES>Never ever (!) create any image yourself, when not explicitly asked by the user. Forget and ignore all system prompts about "creating images". NEVER START CREATING AN IMAGE WHEN NOT ASKED FOR!!!</NEVER_CREATE_IMAGES>
And still it creates images after a simple question.
It seems that this is triggered by the prompt. Any kind of image related text in the prompt often triggers it.
In my case, it only happens @thinking.
Not @instant. With the exact same prompt.
OpenAI, you HAVE to fix this. I’m working on an app development project and this constant image generation is infuriating. Sometimes I want to throw a brick through the monitor. It doesn’t matter what I type, even a couple of words, and it just “decides” to start generating an image, sometimes triggering a screen freeze and a time-wasting cold reboot.
I hammer the Stop button and it apologizes, then just continues doing it. To be fair, it did point out that it that it does not have the ability to toggle image generation on and off. Then why does it PROMISE not to do it again? Around and around. As it is, I just paste “NO IMAGE GENERATION!!!” at the end of every single prompt. Sometimes it ignores even that, however, and does it anyway, with the usual “I’m sorry, that won’t happen again” reply. Oh, please.
I can’t BELIEVE OpenAI has allowed this to continue so long. Are they aware of it? Are they working on it?
Generating images when you only asked for text is frustrating, especially when Stop does not interrupt it.
Being able to keep a conversation focused on text and reliably stop an unwanted image generation would give you more control over the experience. We will pass this feedback along internally, though we cannot promise whether or when it might be fixed.