Facing the second half of the yearGoogle Nano BananaAfter a strong comeback in the field of visual generation, OpenAI finally couldn't sit still and announced the release of its latest image vision model - GPT-Image-1.5.
This update is not simply a matter of stacking parameters, but rather an improvement targeting the most troublesome pain points for creators: speed, controllability, and cost. OpenAI CEO Sam Altman even took to social media to share a calendar photo of a sexy hunk generated by the new model, demonstrating the new model's powerful ability to achieve consistency in portraits.
For example: pic.twitter.com/qcEEjfG8g0
- Sam Altman (@sama) December 16, 2025
Say goodbye to the mystical aspects of gacha pulls: Precise editing with the ability to change exactly what you point to.
In the past, the biggest fear when using AI to edit images was that "one small change would affect the whole picture," meaning that if you wanted to change your hairstyle, your face would also be changed. The biggest upgrade of GPT-Image-1.5 is its deep understanding of the structure of the image.
In the official case study, starting with a 2000s-style party photo, the process involved adding background figures, changing the style of specific objects (e.g., turning the left side into a hand-drawn illustration and the dog into a doll), and finally dressing everyone in OpenAI sweaters. Throughout the entire process, the unedited areas maintained an astonishing consistency. This means that AI can finally understand the instruction "change only here, leave everything else alone," making photo editing no longer a matter of luck.
Speed increased by 4 times, API price cuts to seize market share
In addition to accuracy, speed is also crucial. GPT-Image-1.5's generation speed is four times faster than its predecessor, almost reaching the level of "instant feedback," significantly reducing the time cost of trial and error.
In terms of business strategy, OpenAI has also launched a price war. The cost of image input and output through API calls has been reduced by 20% (priced at $8 per million input tokens), which is obviously aimed at attracting more enterprise users (such as Wix and Canva) to adopt the technology and build a more solid competitive advantage.
Independent creation portal: ChatGPT transformed into Photoshop?
To complement the new model, ChatGPT has also added a dedicated image creation portal on its website and app.
This interface is more like a simplified photo editing software, with a built-in filter library, popular prompt templates, and even the ability to upload a personal portrait with fixed facial features. This is indeed quite convenient for general users who don't want to repeatedly type long prompts.
Actual testing: Attention to detail is good, but Chinese remains a weak point.
According to actual tests, the new model performs remarkably well when handling instructions requiring precise proportions, such as "Mark Zuckerberg sitting on the beach watching the sunset while wearing Ray-Ban smart glasses." Its ability to convert real photos into line art has also been significantly improved.
However, despite its strong English proficiency, its Chinese comprehension remains a disaster. Even when asked to draw a Chinese comic like "Harry Potter vs. Transformers," the resulting text is still illegible, demonstrating that there's still room for improvement in the localization of culture and language.
Analysis Perspective: From "Arms Race" to "Pragmatism"
In my opinion, setting the version number to 1.5 instead of 2.0 suggests OpenAI's pragmatism and caution.
Faced with Google Nano Banana Pro's lead in logical reasoning and physics simulation, OpenAI has chosen a path closer to commercial application: extreme efficiency and a controllable workflow. For most users, the ability to solve complex mathematical puzzles may not be important, but the key factor in making a purchase is the ability to quickly and cheaply generate a "just-usable" poster template.
