OpenAI today introduces a new version of ChatGPT Images, powered by its main image-generation model. What does that mean for you? More faithful edits to your intent, faster results, and an experience inside ChatGPT designed to inspire and explore without friction.
What's new in this version
-
More precise edits: the model makes only the changes you ask for and preserves lighting, composition, and people’s features across multiple iterations. Want to change the color of a jacket without touching the photo's lighting? Now it’s more likely to come out right.
-
Speed: it generates images up to 4x faster than the previous version. Less waiting, more creative iterations.
-
Better preservation and realism: it’s better at keeping logos, brand elements, and small details, which makes it useful for marketing and product catalogs.
-
Text and small faces: noticeable improvement in rendering dense text and small faces inside scenes.
-
Integration in ChatGPT: there’s a new Images section in the sidebar (mobile app and chatgpt.com) with dozens of filters and presets to get started without writing a complex prompt.
-
In the API it appears as GPT Image 1.5: it includes the same improvements and also reduces image input and output costs by 20% compared to GPT Image 1, letting you iterate more within the same budget.
Practical use cases
-
E-commerce: generate product variants, scenes, and angles from a single photo to build catalogs faster.
-
Marketing and branding: keeping logos and visuals consistent across edits makes it easier to create cohesive campaign assets.
-
Photography and quick retouching: spot corrections, background changes, or stylistic tweaks that respect the essence of the original image.
-
Creativity and prototypes: moodboards, mockups, and conceptual transformations with presets to explore ideas without starting from scratch.
Not a designer? No problem. It works like a creative studio in your pocket: practical and accessible.
How to try it now
-
Go to chatgpt.com or open the mobile app and look for the Images section in the sidebar.
-
Upload a photo and ask for a specific edit: you’ll see it keeps details like lighting while applying the change.
-
If you use the API, look for the GPT Image 1.5 model and try it in the OpenAI Playground or in your current flows; image inputs and outputs are 20% cheaper.
Limitations and precautions
-
Not perfect: OpenAI regrouped examples and confirms clear improvements, but there are still cases where results fail or don’t match the user’s intent.
-
Mind copyright and responsible use: edits and recreations may touch protected material or pose risks of improper representation.
-
Biases and safety: the safety and policy teams are still working on mitigations; it’s important to verify sensitive outputs before using them commercially.
Deployment and team
The new version launches today for all ChatGPT users and for API users globally. You don’t need to select anything to use it: it works inside ChatGPT and as GPT Image 1.5 in the API. The previous version is available as a custom GPT for anyone who prefers it.
Behind the launch are research, product, inference, and safety teams, led by people like Gabriel Goh and Adele Li, with contributions from dozens of researchers and product folks.
Today this is a significant step, not the final destination. You’ll see useful improvements now and more iterations to come: finer edits, more detailed outputs, and broader language support.