Image generation (experimental)

Ollama now supports experimental image generation on macOS, and states that Windows and Linux support is coming soon. Images are generated locally by running a model through the ollama run command and are saved to the current directory. Terminals that support image rendering, such as Ghostty and iTerm2, can preview images inline.

Two models are listed for the feature. Z-Image Turbo is a 6 billion parameter text-to-image model from Alibaba's Tongyi Lab, described as generating photorealistic images and handling bilingual text rendering in both English and Chinese, with open weights under Apache 2.0 for commercial use. FLUX.2 Klein is described as Black Forest Labs' fastest image-generation model to date and is available in 4B and 9B parameter sizes; the 4B model is Apache 2.0, while the 9B model uses the FLUX Non-Commercial License v2.1. Both are run with the ollama run command using the x/z-image-turbo and x/flux2-klein model names respectively.

Configuration options for image generation include changing the save location by changing directories in the terminal, setting width and height with the /set width and /set height commands, controlling the number of steps, setting a random seed for reproducible results, and using negative prompts to guide the model on what not to include. Ollama notes that smaller images generate faster and use less memory, that fewer steps are faster but less detailed, and that too many steps can cause artifacts, with defaults set to the recommended step count for each model.

Ollama lists Windows and Linux support, additional image generation models, and image editing as what's next.