I use the same exact prompt all the time but ever since everything went down the other day it seems images that are generated are ignoring uploaded reference photos and generating degraded lower quality images. The same exact prompt with the same exact uploaded referance image is now generating entirely different end results. I’m also frequently getting an “error” on my original prompt but chat will still generate an image. Its very weird and frustrating
Yes, I’m in the exact same situation - and so far, my bug reports, support requests have been ignored. There are other users reporting this exact issue - with the same timeframe - in other threads and in other areas like X and reddit, so at least know you’re not alone, which might mean something is done about it eventually. For me, it has completely ruined my productivity and projects, as essentially image generation is broken.
There is a great difference between ChatGPT Classic and ChatGPT Work, even more so with Astra. I still have to make some more digging to see more about it. One thing I did noticed is that with Astra images have great adherence and stay in scope of the text. GPT 5.6 tends to enrich the prompts more.
Hi @showaemergency and @AthenaWV ,
I would also like to see the prompts that were used, if anyone would like to share it here? That way others could test it.
Also if you use ChatGPT, are you using Chat or Work when generating?
Easier to help, when we have little more to go from.
Personally, I haven’t experienced downgrading quality.
Chat GPT I just use “instant” I’m on the highest paid pro plan and usually get gorgeous images. Sorry not a pro at navigating in here I have always just used this model
Same! I am so annoyed and stressed. I have reported a bug about 100 times over the last couple of days. I’m on the highest paid pro plan but about to cancel as I cannot use this any longer
I have always worked in regular chat GPT. Have not touched chat gpt work but am going to go play around and see if that makes any difference
How many images per day are you producing on average?
I think / wonder if ChatGPT goes to lower quality if you’re mass producing?
Have you looked into any API tools where you can control more things like quality?
I produce alot but honestly that has never been an issue up until now. I have worked the same prompt and same amount of generations everyday for months always with great results =/ I have not looked into the API tools but I am now to see if I can find some sort of alternative that will gives similar results as my old work
Exactly how I’d describe my workflow. I do produce a lot of images, but it has never impeded my results in the many months that I’ve been using a Pro account. I create all of my images through the standard GPT images area of the web browser.
Here’s an example:
The Prompt:
Style: A shot from a fictional live action 1974 Japanese Tokusatsu film with Japanese special effects.
Scene: A scene from a live-action 1970s Japanese tokusatsu special-effects movie, shot on a Tokusatsu scale model miniature set of Tokyo harbour. The harbour is a large practical water tank set. A kaiju is walking through the water.
1970s Chiaroscuro Tokusatsu studio lighting.
Shot on a vintage telephoto long lens with telephoto compression and shallow depth of field.
Aspect ratio: 4:3
*Do not reference previous image generations.
*
And here are the results of multiple separate generations:
As you can see, the image is nearly identical every time I generate it, even in new chats. This has never been an issue before. Not only that, but the quality is lacking in many of the photo-realistic aesthetics I require for my work, that always previously generated just fine with this prompt. Changing the prompt doesn’t matter, it always has this same basic AI-looking aesthetic, and similar composition.
Here’s an example from a week ago, with the same prompt, showing the ideal quality that I was getting consistently:
Notice that the new generations have the nearly identical layouts, compositions, lighting, backgrounds, regardless of iteration, prompt changes or done in separate new chats. It constantly just produces the exact same low quality image, despite the fact that prompt instructs not to reference previous generations. It’s as if it’s producing the exact same image every time, with lower quality results.
I am currently playing around in Astra and so far while not the same as previous work the quality is MUCH better than anything I am getting in regular chat. I can absolutely seethe difference in the quality of your uploads. Am experiencing the exact same. Are you in regular chat or in work?
I always use the “images” section of the main web browser. I’ve experimented with using Astra in work mode, here are the latest results:
As you can see, it’s the identical composition of the images created with various models and effort levels in the standard images mode. I have never gotten identical compositions - not to mention such low quality with just repeated imagery, props (look at the telephone wires just sprayed and duplicated all over the image) and poor photo-realism since using ChatGPT. All of this aligns with the major crash the other day, as well as the rollout of Astra.
I’ve submitted this to the support team, and all I’ve received is that they acknowledge the issue being submitted, yet the issue is not officially acknowledge by Open AI anywhere, nor have I been given a timeline to when it might be fixed.
There are users all over reddit experiencing the same issue, so it’s not just me. It’s clearly a big enough issue that it should be acknowledged and addressed. Images is essentially useless at a Pro level account for me, which is what I use the service for to begin with.
No I completely agree what ever the issue is needs resolved esp with us being pro paid users. I am just trying to find ANY sort of work around right now to be able to meet my clients needs. This service isn’t cheap and my business relies on it. I hope they hear us and get this fixed soon 
Are we talking dozens of images per day? Hundreds? Thousands?
Yeah, OpenAI is kinda giving away AI for very low cost and losing money with ChatGPT, even the $200/month accounts.
If it’s business critical, I would definitely look into the API.
There’s a thread here with Python code you can run on your own. Anyone have the link handy?
Not every day - but by the end of a busy week, probably several hundred, but certainly not 1,000.
This is a helpful post to understand the costs behind the various sizes and qualities…
Another good post…
Are all images coming out “bad” / “low quality” or just after you’ve been running a lot?
For the past 4-5 days, every image generation has been creating identical low quality images when I attempt to generate, rendering chatGPT useless for me.
I tested your exat prompt with gpt-image-2 using ai API image editor, you can find it in this topic : Gpt-image-2 image creator and editor app - Python, desktop UI (download in latest _J reply!) - #20 by _j
This is quality:medium
This is quality: high:
