GPT Image 2 is OpenAI's new flagship image model. It launched straight to the top of the public image leaderboards and it is live in AI Image Lab today, selectable alongside Nano Banana 2, Nano Banana Pro, Kling Image, Grok Imagine, and the rest of the model line-up.
Three strengths stand out in real work, not just in benchmarks. This is where the model earns its place in the line-up.
GPT Image 2 produces coherent text in Japanese, Korean, Chinese, Hindi, Bengali, and other non-Latin scripts. You can ask for a manga page, a Korean advert, or a Chinese poster and get characters that are correctly formed and laid out. Useful for localised marketing, global packaging mockups, and multilingual infographics without running a separate design pass per language.
GPT Image 2 supports square, landscape, and portrait orientations at 1024 resolution, plus HD (1920×1080), QHD (2560×1440), and 4K (3840×2160) for landscape assets. Pick the size that fits your final surface: social tile, slide deck, banner, print.
AI Image Lab already lets you write one creative direction, expand it into prompt variations, and run the result across several models in the same batch. GPT Image 2 slots into that flow. Tick it alongside Nano Banana 2 and a Kling model, for instance, and every prompt runs on all three in parallel. You get a side-by-side comparison without re-entering anything.
It also works with reference images. Upload a photograph, a product shot, or a previous generation, and GPT Image 2 will edit from that starting point. Combine it with image markers to point at specific regions of the reference image and describe what should change.
Start your free trial . No credit card required. Open AI Image Lab, write a creative direction, and tick GPT Image 2 in the model list. Run it on its own or alongside the other models to compare.