ChatGPT Images 2 shows realism, styles

OpenAI on April 21 introduced ChatGPT Images 2.0, a new image-generation model the company said improves text rendering, multilingual support and visual reasoning. The release also expanded image creation and editing inside ChatGPT, according to OpenAI’s product post and help documentation. Users this week circulated examples on X and elsewhere showing more photorealistic outputs, style changes and broader editing workflows, but OpenAI’s published materials focus on the model and product features rather than those individual posts. QwenLM, meanwhile, has been marketing Qwen-Image-2.0 since February with specifications that include 1,000-token instructions and native 2K output, giving developers another benchmark in the current image-model cycle. ### What did OpenAI actually announce on April 21? OpenAI said ChatGPT Images 2.0 is a “state-of-the-art image generation model” with improved text rendering, multilingual support and advanced visual reasoning. The product page says the model is available in ChatGPT. The OpenAI API documentation lists a GPT Image 2 model, while an earlier OpenAI post from late 2025 described a prior image release as “GPT-Image-1.5.” That sequence indicates OpenAI has moved its public image stack forward since the earlier rollout. (openai.com) ### Where do the editing and “in-chat” claims come from? OpenAI’s help documentation says ChatGPT Images lets users create new images and edit existing ones directly in ChatGPT on web, iOS and Android. (openai.com) The same page says users can upload an image, open the editor, select part of an image with a selection tool and describe changes in chat, or describe edits directly in the conversation panel. (developers.openai.com) OpenAI’s academy materials also describe using multiple uploaded images to guide generation or editing, including style references. That is the clearest official basis for claims about style transforms and iterative prompt-edit-create workflows inside ChatGPT, though OpenAI does not use the phrase “infinite canvas” on the pages reviewed here. ### Did OpenAI document an “infinite canvas” feature? OpenAI has a separate product called canvas for writing and coding projects, and its help center says that feature is available on web, Windows and macOS, with mobile support coming later. (help.openai.com) The canvas documentation reviewed here does not describe a dedicated image-editing canvas tied to ChatGPT Images 2.0. That means the “infinite canvas” description in social posts is not confirmed by the official OpenAI materials surfaced in this reporting. (openai.com) What OpenAI has documented is image editing inside ChatGPT, including region selection and conversational edits. ### How does Qwen-Image-2.0 compare on published specs? QwenLM’s GitHub repository says Qwen-Image-2.0 launched on February 10, 2026 as a next-generation image model. (openai.com) The repository says the model supports 1,000-token instructions, native 2K resolution, stronger semantic adherence and image editing, with a particular emphasis on typography rendering. Alibaba Cloud’s Model Studio documentation lists qwen-image-2.0 and qwen-image-2.0-pro as generation-and-editing models. (help.openai.com) The documentation says the Pro series offers stronger text rendering, realistic textures and detailed scenes, while the standard 2.0 model is positioned as a balance of quality and speed. ### Why are users focusing on realism and styles? OpenAI’s own release describes improvements in text rendering and visual reasoning, while Alibaba Cloud describes Qwen’s newer image models in terms of realistic textures and naturalness. (github.com) Those are the product-level claims companies are making as users compare outputs on social platforms. The OpenAI help and academy pages also show why style has become a common test. (alibabacloud.com) Users can create images from prompts, upload reference images and ask for edits in chat, which makes side-by-side comparisons of realism, typography and style consistency easy to produce and share. ### What should readers watch next? OpenAI’s documentation says ChatGPT Images is already available across ChatGPT surfaces, and the API documentation lists GPT Image 2 as a current model. (openai.com) That makes developer documentation and product updates the next places to watch for any formal description of broader canvas-style editing features. QwenLM’s GitHub repository and Alibaba Cloud’s Model Studio pages remain the primary public references for Qwen-Image-2.0 specifications. (help.openai.com) Future changes to prompt limits, output sizes or editing features are likely to appear there first. (github.com)