Ollama launches experimental local image generation on macOS with Z-Image Turbo and FLUX.2

Ollama has introduced experimental image generation capabilities, expanding its open-source platform beyond text-based large language models (LLMs). At launch, this feature is available for macOS users, with support for Windows and Linux listed on the roadmap.

Building on the upgrade, generated images are saved to the current directory. Terminals that support inline image previews, such as Ghostty, iTerm2, and Kitty, can display these images directly within the terminal interface.

In terms of model support, Ollama enables image generation with Z-Image Turbo, a 6 billion parameter text-to-image system developed by Alibaba. This model can generate photorealistic content and accurately render bilingual English and Chinese text. Additionally, FLUX.2 Klein from Black Forest Labs, available in 4B and 9B parameter formats, is supported for projects requiring readable text in images, such as UI mockups and typography-centric designs.

To tailor image outputs, users can adjust width and height using the /set width and /set height commands, and specify the number of steps to control how many iterations the model runs, and set a seed for reproducible results. Looking ahead, the Ollama team plans to add image generation for Windows and Linux, support more model types, and introduce image editing features.