According to Ars Technica, on Tuesday, OpenAI released a new AI image synthesis tool called GPT Image 1.5 to all ChatGPT users. This model reportedly generates images up to four times faster than its predecessor and costs about 20 percent less through the API. It’s a “native multimodal” model, meaning it processes images and text prompts within the same neural network, allowing for conversational photo editing. The release comes after Google beat OpenAI to market with a public prototype in March, later refining it into the popular Nano Banana image models. OpenAI’s new tool is designed to match those editing features, particularly in preserving facial likeness across edits and rendering text more clearly. The company’s CEO of applications, Fidji Simo, announced a new dedicated image creation space in ChatGPT’s sidebar to accommodate the visual task.
Strategy and timing
Here’s the thing: this isn’t just another incremental update. The timing is a direct, competitive response. Google’s Nano Banana models gained serious social media traction after their August release, especially for keeping faces consistent and rendering legible text. That got OpenAI‘s attention, and fast. So they’re pushing out GPT Image 1.5 to claw back mindshare and developer usage. Making it faster and 20% cheaper via the API isn’t an accident—it’s a classic land-grab move to make their model the default for anyone building apps that need this stuff. They’re basically commoditizing advanced image manipulation before our eyes.
The friction keeps dropping
And that’s the real story. For most of history, faking a photo well took real skill. Now? It’s a chat. You upload a pic of your dad and type “put him in a tuxedo at a wedding.” The model treats the pixels and your words as the same kind of data, predicting new pixels like it predicts the next word in a sentence. The barrier isn’t just lower; it’s practically gone. Sure, the model isn’t perfect—it still messes up prompts sometimes. But when it works, the results are scarily convincing. A consultant named Shaun Pedicini has even set up a useful “GenAI Image Editing Showdown” site to compare outputs, and the progress is stark.
Obvious utility, obvious danger
This capability is a double-edged sword. The power to tweak a photo while keeping someone’s face perfectly intact is a dream for legitimate editors. But it’s also a nightmare for creating deepfakes and non-consensual imagery. OpenAI has filters for overtly sexual or violent content, but what about just putting someone in an embarrassing, fake situation? That’s still possible. They’re adding C2PA metadata to label images as AI-generated, but let’s be real—that can be stripped with a quick re-save. We’re in a new era where every image is suspect. And the improved text generation, shown off in a demo of a fake newspaper, just pushes us further toward a world where even historical-looking documents can’t be trusted.
Where this is all heading
OpenAI admits the model has flaws with certain drawing styles and scientific accuracy. But they believe, as they wrote in their blog post, that “we’re still at the beginning.” And if the last few years are any indication, they’re probably right. The tech will only get better, faster, and more accessible. This forces a huge cultural question: what do visual images even mean anymore when anyone can fabricate reality over a coffee break? It’s a tool of incredible creative power and equally incredible destructive potential, and it’s now sitting in the sidebar of ChatGPT for millions. The friction is gone. Now we have to deal with the fallout.
