OpenAI has officially launched ChatGPT Images 2.0, its latest image generation model, and the big upgrade here is not just prettier AI art. The company says this new version is much better at following detailed instructions, handling lots of text inside an image, and keeping objects in the right place relative to each other.
That already sounds useful, but the part that will probably catch attention across Malaysia and the wider SEA creative scene is this: OpenAI says the model has made major gains with non-Latin text, specifically Japanese, Korean, Chinese, Hindi and Bengali. If that claim holds up in real-world use, this could be a pretty meaningful upgrade for regional creators, indie game teams, fan artists and marketers who need visuals that do not instantly fall apart the moment you ask for Asian-language text.
According to OpenAI, Images 2.0 is also its first image model with reasoning capabilities. The company says that lets the system do things like search the web and verify outputs, with the goal of making results more reliable when accuracy, consistency and visual cohesion matter.
For anime, manga and game-adjacent work, that matters a lot. One of the biggest frustrations with AI image tools has been that they can look nice at first glance, then completely fumble the details when you need readable text, consistent characters, or a layout that actually makes sense. OpenAI says the new model is also better at recreating different visual languages faithfully, which it pitches as useful for game prototyping and storyboarding.
There are some practical upgrades too. Images 2.0 now supports aspect ratios as wide as 3:1 and as tall as 1:3, can generate at up to 2K resolution, and can produce up to eight outputs at once. That gives users more flexibility for banners, vertical social posts, manga-style panels and other formats that do not fit the usual square-image workflow.
In Engadget's early preview, the model was tested with three prompts: a tortoiseshell cat rendered in the pixel art style of Pokemon's third-generation games, a request to turn that image into a transparent PNG, and a four-page manga about the cat enjoying a sunny day by a city stream. The preview said the Pokemon-style result was surprisingly solid, especially since pixel art is usually a weak spot for AI models.
The transparent PNG test was a bit shakier. It took longer, and the output was slightly different from the original image, which meant it drifted from the prompt somewhat. Still, it reportedly managed to produce a proper transparent image, which is something other image models do not always handle well.
For readers in Malaysia and SEA, the non-Latin angle is probably the real headline. A lot of regional creative work sits across multiple language and visual cultures at once, especially when you're dealing with anime fandoms, Chinese-language promo art, Korean-inspired design trends or game assets aimed at mixed audiences. If Images 2.0 can genuinely keep text cleaner and style references tighter, that could save a lot of cleanup time.
The new model is available starting today for all ChatGPT users, including the Free and Go tiers. Plus and Pro subscribers get access to more advanced outputs. OpenAI is also rolling it out through its API and Codex coding app, which only recently got built-in image generation. The launch also comes just days after Anthropic pushed into visual design with its own design assistant, so the AI image race is definitely heating up.
Source: Engadget