Image creation has bonked. Is this intentional?
The story began on March 22, when an OpenAI community post said image creation had “bonked” and asked whether this was intentional. The question cast a routine failure as something more suspicious: had the tool broken, or had someone decided it should behave this way?
On June 17 and June 19, ChatGPT posts asked whether image generation in custom GPTs was broken. A July 4 complaint focused more broadly on ChatGPT art generation. On July 10 came the less technical but familiar diagnosis: “Did something happen?”
By August, the reports had become more specific. An August 10 post said Chat image generations had degraded. On August 14, another asked how to create multiple images with ChatGPT, suggesting that basic workflow expectations were causing friction.
The sharpest claim arrived through OpenAI on August 25: gpt-image-1 was being shut down, and one user said they could not reproduce their indie game’s visual identity with gpt-image-2. On August 26, a ChatGPT user said image preservation during edits had become so poor that they received a refund. A final post on August 27 asked whether anyone else was having image-generation problems that day.
The pattern has widened from one “bonked” tool to complaints about custom GPTs, visual consistency, editing and model replacement. The original question remains unanswered. Thirteen signals describe recurring frustration without establishing intentional degradation.
Taking this as factual, impact is 10%. The immediate effects fall on ChatGPT image users, custom GPT builders and creators who depend on repeatable visual identities. They include the indie game developer unable to reproduce work made with gpt-image-1 in gpt-image-2. Broken generation, weaker image preservation and inconsistent outputs create extra retries and rework. A refund shows that at least one complaint crossed from irritation into a failed paid experience.
Watch for an OpenAI explanation of the gpt-image-1 shutdown, documented changes to image generation or editing, and reproducible comparisons between gpt-image-1 and gpt-image-2. Those would show whether this is a product decision, a recurring fault, or several complaints sharing one convenient theory.
<!-- SC_OFF --><div class="md"><p>Just spent an hour prompting the new Image 2.0 and the quality jump is ridiculous. Complex scenes, accurate lighting, and consistent details on
<!-- SC_OFF --><div class="md"><p>i fed it some social media data to see what interesting insights it would come up with, instead, it started yoinking my hog about how much of a
<table> <tr><td> <a href="https://www.reddit.com/r/ChatGPT/comments/1t6pfv6/gptimage2_is_not_a_better_midjourney_its_the/"> <img alt="GPT-image-2 is not a “better Midjourney.”
<!-- SC_OFF --><div class="md"><p>I just got this pop-off saying that thinking mode now actually improves the accuracy of the images, and it does... like quite a lot. Suddenly, it
<!-- SC_OFF --><div class="md"><p>So I just noticed that the web interface got a silent update, now when you paste very long text, instead of pasting it inline it becomes a
<!-- SC_OFF --><div class="md"><p>Is anyone else running into weird tiling / mosaic artifacts with ChatGPT Image Gen 2?</p> <p>I’m seeing it across different kinds of images. The
<!-- SC_OFF --><div class="md"><p>I use ChatGPT only to create images of my DnD characters. The past day it keeps having problems. I will give it some images to referance, ask it
<table> <tr><td> <a href="https://www.reddit.com/r/ChatGPT/comments/1tf9l3h/watermarks_have_now_been_added_to_images/"> <img alt="Watermarks have now been added to images."
NewsHive monitors these sources continuously. All signal titles above link to the original reporting.
Intelligence by NewsHive. Need help navigating what this means for your business? Contact GeekyBee →