OpenAI’s new ChatGPT image generator makes faking photos easy (arstechnica.com)

OpenAI has unveiled its new GPT Image 1.5, an AI image synthesis tool that simplifies photo manipulation to the extent of merely typing a sentence. This release follows Google's earlier success with its Nano Banana image model and responds to the growing demand for more accessible image-editing capabilities in the AI community. GPT Image 1.5 is designed to generate images up to four times faster than its predecessor and at a reduced cost through the API, marking a significant evolution in how users can create and alter images without needing advanced visual skills. What sets GPT Image 1.5 apart is its “native multimodal” architecture, which integrates language processing with image generation within the same neural network. By treating images and text as interchangeable data "tokens," the model can seamlessly alter photos, modifying elements like poses or backgrounds while maintaining realism. Users can interactively converse with the model to iteratively refine images, a feature reminiscent of drafting text in ChatGPT. This innovation not only accelerates the creative process but also raises important discussions about the implications of easily generating and altering photorealistic images.