gpt-image-1.5-rev (0.1 per time)
gpt-image-1.5
OpenAI has officially launched a new version of ChatGPT Images, powered by its latest flagship image generation model, GPT Image 1.5. This new model is designed to be significantly more efficient, generating images up to four times faster than its predecessor. It is currently rolling out to all ChatGPT users, with a dedicated “Images” space accessible via the sidebar on both the web and mobile app. For developers, the model is available in the API as GPT Image 1.5, offering a 20% reduction in cost for both image inputs and outputs compared to the previous version.
A primary focus of this update is the ability to perform precise edits while maintaining the integrity of the original image. When users request changes to an uploaded photo, the model more reliably adheres to their intent, modifying only the specified areas while keeping elements like lighting, composition, and a person’s appearance consistent. This capability allows for practical applications such as virtual clothing or hairstyle try-ons and conceptual transformations that retain the essence of the source material. The model excels at various editing tasks, including adding, subtracting, blending, and transposing elements, effectively acting as a portable creative studio.
Technical improvements in instruction following and text rendering further distinguish GPT Image 1.5. The model can now handle intricate compositions, such as generating specific 6×6 grids of diverse objects, with much higher accuracy. Text rendering has also seen a significant leap, enabling the model to produce dense and small text found in infographics, diagrams, or code snippets with greater clarity. Additionally, the update enhances overall image quality, particularly in rendering natural-looking scenes and many small faces within a crowd, which were previously challenging for generative models.
To make the creative process more accessible, OpenAI has introduced a new Images feature within ChatGPT that includes preset filters and prompt suggestions. While the model shows marked improvements in scientific accuracy and visual vividness, OpenAI notes that it is not yet perfect; limitations remain in areas such as multilingual support and specific complex styling. Despite these hurdles, the model is already being utilized by enterprises for brand-consistent marketing graphics and e-commerce catalogs, where maintaining visual identity across multiple iterations is essential.
