OpenAI has introduced a new image generation model named ChatGPT Image 2.0. According to OpenAI, this model can create accurate visuals that are immediately usable. This development follows the recent discontinuation of the Sora video generation model by the company, as resources are being redirected towards coding capabilities due to the increasing popularity of Anthropic.
The company stated in a blog post that the new image generation model excels at following detailed instructions, accurately representing objects, and incorporating text into images. OpenAI highlighted that the model’s strength lies in producing images that appear less AI-generated and more intentionally designed, thanks to its composition skills and visual aesthetics.
Furthermore, Image 2.0 is capable of generating images across various aspect ratios. OpenAI has been testing this model extensively, and it is now accessible to all ChatGPT and Codex users. Notably, ChatGPT Image 2.0 is the first model from the company to incorporate thinking capabilities.
Users can leverage this feature by selecting thinking or pro mode in ChatGPT, allowing Image 2.0 to search the web for real-time information, generate multiple distinct images from a single prompt, and self-validate its outputs. Advanced thinking outputs are accessible to ChatGPT Plus, Pro, and Business users.
OpenAI emphasized the improved usability and attention to fine details offered by this model. While AI-based image generation is not novel, the challenge has often been producing outputs that align closely with users’ intentions. Image 2.0 aims to address this gap by delivering practical and usable results.
Additionally, the model excels at rendering intricate elements such as small text, iconography, UI components, complex compositions, and stylistic nuances. OpenAI also presented a demo of ChatGPT Image 2.0, showcasing its capabilities through a screenshot of a ChatGPT conversation, emphasizing the realism it can achieve.
Building on the success of GPT-4o in image generation, OpenAI’s new model enables users to create visuals in diverse styles, including Japanese seinen manga, movie posters, midcentury pastel comics, and more.
Moreover, Image 2.0 offers robust text and multilingual support. Users previously faced challenges with inaccurate text and spelling errors in images, which this model aims to rectify. OpenAI highlighted its multilingual capabilities, asserting that it can accurately render text in languages like Japanese, Korean, Chinese, Hindi, and Bengali.
The model proves beneficial for users seeking to create books or comics, as it streamlines the creative process by generating up to eight coherent outputs in one go. In thinking mode, Image 2.0 can produce multiple unique images simultaneously, accommodating various aspect ratios and languages.