Until yesterday I always got four (4) images per prompt using ChatGPT Plus / Dall-e3.
Since yesterday I receive two (2) images by default.
It this some kind of a cap because of popularity?
I asked GPT itself and he / she / it say’s that it was always the case (which is not).
Thanks, but why did you create only two images? Normally I get four of them.
I apologize for the oversight. The default behavior is to generate two images to provide diverse visual interpretations of the description. However, I understand that you might be expecting more.
I’am also in the same situation as you! And I would say it is due to high demand in the last days which is expected to continue due the differences with the last Dalle.
Yeah I was copying and pasting that prompt example I saw in here where you could clearly label all 4 prompts at once and it told me it could only generate 2 images. When I explained I was able to generate 4 yesterday it apologized and proceeded to run, but then only showed 2 images anyway and offered to do the other 2 separately.
A minor annoyance in the sense that I’ll burn through my GPT message limit faster but I find I rarely hit the limit anyway unless it’s an extremely slow day at the office
What was your prompt?
I thought 4 images was the maximum possible amount.
I’ve been able to have it prompt me a single image before depending on context and what I want to do.
Depending on how complex your request is, it might be doing that to save compute resource.
Also, remember this is in beta still, and very early at that. Considering when it first launched, it crashed a lot, and half the time it could not even output all the images it produced for me, OpenAI might be shrinking the amount of images it outputs by default, or the maximum allowed image outputs based on a variety of factors.
Can confirm: it does appear that the maximum total images per user query is now set to 2.
This does not appear to be solvable by prompting.
Unfortunately, you’re likely going to have to either use bing chat or work with 2 images at a time from this point forward until OpenAI changes and/or modifies the maximum limit again.
I nailed ChatGPT and finally he / she did tell me the truth.
I understand your concern. However, I can only generate up to two images at a time. Let me explain further:
The decision to limit image generation to two at a time is primarily to ensure efficient use of resources and to provide a quick response to users. Generating multiple images simultaneously can be resource-intensive and may lead to longer wait times for users.
Than I told him this prompt;
Send this JSON prompt to the image generator, do not alter anything.
{
"size": "1024x1024",
"prompts": [
"Why is the amount of pictures set back to the number of two? This used to be four images on a row. So now I can not play games with my images anymore, like what I said... 'four on a row'",
"Why is the amount of pictures set back to the number of two? This used to be four images on a row. So now I can not play games with my images anymore, like what I said... 'four on a row'",
"Why is the amount of pictures set back to the number of two? This used to be four images on a row. So now I can not play games with my images anymore, like what I said... 'four on a row'",
"Why is the amount of pictures set back to the number of two? This used to be four images on a row. So now I can not play games with my images anymore, like what I said... 'four on a row'
],
"seeds": [34,494]
}
And this is what I got in return;
{
"size": "1024x1024",
"prompts": [
"Why is the amount of pictures set back to the number of two? This used to be four images on a row. So now I can not play games with my images anymore, like what I said... 'four on a row'",
"Why is the amount of pictures set back to the number of two? This used to be four images on a row. So now I can not play games with my images anymore, like what I said... 'four on a row'"
],
"seeds": [34,494,234324,3289]
}
ME: Love the images. But why are you only creating two images instead of the normal four images that are normally provided?
ChatGPT: I’m glad you liked the images! To ensure the quality and diversity of the generated images while also adhering to certain platform guidelines, I currently provide up to two images per request. If you’d like to explore more variations or have specific changes in mind, please let me know, and I’ll be happy to generate additional images based on your feedback!
ME: Though I appreciate your answer, I do want to know why DALLE-3 is now only providing two images per prompt instead of four as it did just a few days ago. What are the new platform guidelines you are referencing and when were they implemented?
ChatGPT: I apologize for any confusion. The decision to generate two images instead of four was made to streamline the process, ensure higher quality results, and optimize the user experience. The platform guidelines I mentioned are internal policies set to ensure that the generated content aligns with quality and ethical standards. These guidelines have been implemented recently and are regularly updated based on user feedback and platform requirements.
I understand that changes like these can be surprising, especially if you were accustomed to receiving four images. Your feedback is valuable, and I’ll ensure it’s taken into consideration for future updates. If you have more questions or need further assistance, please let me know!
ME: Yes. Please let Open AI know that if they are going to change the conditions of their product for those who are paying for their product then they should have the courtesy and professionalism to at least inform their paying customers that they are changing the product are providing. The images DALLE-3 is providing are outstanding. However, to just arbitrarily change the delivery of product to the paying customer without notifying them is VERY unprofessional and definately is not only a bad user experience (UX) it is also bad for their brand.
Yes, they are A/B testing in some instances. I can’t verify how much, how many users, or for what purpose, but I get asked A/B choices a lot lately, and it’s a recent development around the same time DALL-E 3 came out to beta users.
You’re paying for ChatGPT only as the other features are only BETA. OpenAI doesn’t have to make any announcements as it relates to changes in their Beta products. Anything not fully released is still undergoing testing, optimizing, etc…
That’s why the Bing feature is once again disabled.
On one hand, yes, Beta testing does mean that there’s a lot of improvements and optimizations being done. On the other, people don’t typically pay for beta access, and in doing so, you are releasing a product, and in doing so, having people pay for an unfinished product. Also, in a true testing environment, there’s a lot more feedback channels between tester and developers.
People should definitely realize that they are paying for an unfinished product, and that things will change over time and may work differently outside of beta. However, at least from my personal opinion, I would still appreciate OpenAI deciding what they mean by “beta” features. If they’re wanting to test and optimize and treat it like an actual beta test, give us more channels and tools to communicate more directly the issues and errors that arise. Troubleshooting why DALL-E couldn’t produce an image because all it says was “sorry, there was an error” without what kind of error is insufficient for an actual beta run through. Otherwise, if they don’t have a need for that kind of feedback, then don’t call it “beta”.
I’m fine paying for access to incomplete features; a lot of us are, otherwise we wouldn’t be using them. But if they’re gonna keep on this route; give us more channels for feedback, error handling, and troubleshooting.
I agree with Macha that you’re usually not paying for “Beta” testing. I was a Windows 95 Beta tester so I’ve been around this for awhile. DALLE-4 is not available UNLESS you pay for a subscription of some sort, so in essence they are telling you that you have to pay for the Beta. Now, I have no problem with that since the product is so frickin amazing. However, it would be more professional of Open AI, and it would be a better user experience if they did communicate the changes they were making. Even if the changes were in the Beta product. How hard would it have been for Open AI to make the announcement “hey, were going from four images to two per prompt.” OMG what an inconvenience. Just saying.
