OpenAI has introduced a new image generation model named ChatGPT Image 2.0, emphasizing its ability to create precise and readily usable visuals. This launch follows the closure of OpenAI’s Sora video generation model to shift focus towards coding capabilities due to the rising popularity of Anthropic.
According to OpenAI, the ChatGPT Image 2.0 model excels in following detailed instructions, accurately depicting objects, and embedding text in images. The company highlights that the model’s output feels more intentionally designed and less AI-generated, thanks to its strong sense of composition and visual aesthetics.
One notable feature of Image 2.0 is its capacity to generate images across various aspect ratios. OpenAI had been testing this model extensively before making it accessible to all ChatGPT and Codex users. Notably, this model marks the company’s first to incorporate thinking capabilities.
Users opting for thinking or pro mode within ChatGPT can leverage Image 2.0 to search for real-time information online, create multiple unique images from a single prompt, and self-verify its outputs. Advanced thinking features are accessible to ChatGPT Plus, Pro, and Business users.
OpenAI addresses the historical challenge of AI-generated images often falling short of user expectations by offering a tool that delivers practical and usable results. The model excels in rendering intricate elements like small text, iconography, UI components, dense compositions, and stylistic nuances.
In a demonstration of ChatGPT Image 2.0’s capabilities, OpenAI shared a mock ChatGPT conversation screenshot, emphasizing the model’s realism and attention to detail. Building on the success of GPT-4o in image generation, users can now create visuals in diverse styles like Japanese seinen manga, movie posters, and midcentury pastel comics.
Moreover, Image 2.0 enhances text rendering in images by supporting multiple languages beyond English. OpenAI states that the model now comprehends and accurately renders text in languages such as Japanese, Korean, Chinese, Hindi, and Bengali.
For users looking to create books or comics, Image 2.0 streamlines the process by autonomously handling repetitive tasks. Users can request the generation of up to eight cohesive outputs at once, with the tool ensuring continuity in characters and objects.
In thinking mode, the model can simultaneously produce multiple unique images in various aspect ratios and languages, offering enhanced flexibility and creative possibilities for users.
– Ends