ChatGPT Images and GPT Image 1.5 Release

OpenAI has launched a new version of ChatGPT Images, powered by a new flagship image generation model. This update introduces significant improvements in generation speed, precise editing, and instruction following, allowing users to create and modify images while maintaining consistent details across iterations.

Precise Image Editing and Preservation

GPT Image 1.5 allows for more reliable edits to uploaded images, ensuring that changes only occur where requested while preserving essential elements such as lighting, composition, and the appearance of people. This capability enables more believable clothing and hairstyle try-ons, practical photo edits, and conceptual transformations that retain the original image's essence.

Editing Capabilities

The model excels at adding, subtracting, combining, blending, and transposing elements within an image. For example, users can combine multiple subjects from different images into a single scene or change the style of specific elements (e.g., changing a person to a retro anime style) while keeping other parts of the image intact.

Creative Transformations

The model can perform complex transformations, such as turning a photo into a movie poster with specific text and layout changes, while maintaining the likeness of the subjects. These transformations can be triggered via text prompts or through preset styles in the new Images feature.

Instruction Following and Text Rendering

Instruction following has been improved over the initial version, allowing for more intricate original compositions and better preservation of relationships between elements. Additionally, the model has made advancements in rendering denser and smaller text, enabling it to create realistic newspaper articles or infographics with precise formatting and numbers.

A New Dedicated Creation Space

OpenAI has introduced a dedicated Images experience within the ChatGPT sidebar (available on the mobile app and chatgpt.com). This space is designed to streamline creative exploration through:

  • Preset Filters and Prompts: Dozens of regularly updated filters and trending prompts to jump-start inspiration.
  • Likeness Upload: A one-time upload of a user's appearance to reuse across future creations without needing to upload a new photo each time.
  • Parallel Generation: Users can continue generating new images while others are still in progress, reducing wait times.

Performance and API Availability

Images now render up to four times faster than previous versions. While OpenAI notes that results remain imperfect and there is room for improvement in areas like multilingual support and multiple faces, the model shows clear progress in vivid graphics and avoiding premature cropping.

GPT Image 1.5 in the API

Developers can access the new model via the API as GPT Image 1.5. This version provides stronger image preservation and editing than GPT Image 1, making it particularly useful for brand work, logo creation, and e-commerce product catalogs.

Key API updates include:

  • Cost Reduction: Image inputs and outputs are now 20% cheaper compared to GPT Image 1.
  • Industry Adoption: Companies such as Wix, Canva, Figma, and Shutterstock are already utilizing the model.

Availability

The new ChatGPT Images model is rolling out globally to all ChatGPT users and the API. The previous version of ChatGPT Images remains available as a custom GPT.

Sources