Moments ago, OpenAI unveiled a fresh iteration of ChatGPT Images, which is powered by a cutting-edge, flagship-level image generation model.

This latest image generation model is capable of precisely editing images while keeping all details intact. Moreover, the speed of image generation has seen a boost, increasing up to fourfold.

The research team has also rolled out a brand-new image feature in ChatGPT. This feature is crafted to simplify and enhance the image generation process, sparking creativity and facilitating effortless creative exploration.

ChatGPT's new image model is now accessible to all users, with the API version being identified as GPT Image 1.5.

GPT Image can more accurately adhere to user instructions. When tasked with editing an uploaded image, the model only modifies the elements specified by the user, while ensuring consistency in aspects such as lighting, composition, and character appearance throughout the input, output, and any subsequent edits.

The model demonstrates exceptional proficiency in various editing tasks, including adding, removing, combining, blending, and rearranging elements.

The model's creativity shines through in its transformations of elements like text and layout, which can vividly convey ideas while preserving crucial details.

These transformations are applicable to both simple and complex concepts. They can be readily experimented with using preset styles and creative tools in the new ChatGPT image feature, without the need for any textual prompts.

The model can more reliably follow instructions, craft more intricate original works, and maintain the expected relationships between elements.

Simultaneously, the model has made strides in text rendering, now capable of handling denser and smaller text with ease.

The model has also seen enhancements in other areas, resulting in outputs that are more directly usable. For instance, it can render numerous small faces and produce outputs that appear more natural.

The results indicate that the model shows marked improvements across various scenarios, though it is not without flaws. While this release marks significant progress, there is still ample room for refinement in future iterations.

When combined with the aforementioned image, despite some scientific inaccuracies, approximately 70% of the content is correct. The image is more vivid and avoids premature cropping.

GPT Image 1.5 in the API offers all the enhancements found in ChatGPT Images: it surpasses GPT Image 1 in terms of image preservation and editing capabilities.

Compared to GPT Image 1, GPT Image 1.5 reduces the cost of image input and output by 20%, enabling you to generate and refine more images within the same budget.

Hila Gat, the Head of AI Research and Data Science at Wix, commented, "GPT Image 1.5 can generate high-fidelity images with remarkably fast response times, preserving composition, lighting, and fine details. The generated images are crisp, realistic, and dependable, supporting accelerated workflows from concept to production on platforms like Wix."

References:

https://openai.com/index/new-chatgpt-images-is-here/