Ollama adds experimental image generation with Z-Image Turbo and FLUX.2 Klein

TL;DR

Ollama released experimental image‑generation support on macOS, exposing two models—Z‑Image Turbo (6 B parameters, photorealistic and bilingual) and FLUX.2 Klein (4 B/9 B parameters, strong text rendering)—so users can generate and preview images directly from the terminal.

Immediate Availability on macOS

Ollama’s new run command creates images that are saved to the current working directory. Terminals with inline image support (e.g., Ghostty, iTerm2) display the results instantly. Windows and Linux support is announced for a future release.

ollama run x/z-image-turbo "your prompt"

Z‑Image Turbo Model

Source: Alibaba’s Tongyi Lab Size: 6 B parameters License: Apache 2.0 (open weights, commercial use allowed)

• Photorealistic output: excels at realistic photographs, portraits, and scenes.
• Bilingual text rendering: accurately draws English and Chinese characters within images.

Example Prompts and Results

• Photorealistic portrait: "Young woman in a cozy coffee shop, natural window lighting, wearing a cream knit sweater, holding a ceramic mug, soft bokeh background…"
• Chinese calligraphy: "Traditional Chinese calligraphy brush painting style, the characters "山高水长" written in elegant black ink on rice paper…"
• Creative composition: "Surreal double exposure portrait, woman's silhouette filled with blooming cherry blossom trees…"

All examples are rendered inline in the blog post and can be reproduced with the same prompt using the ollama run x/z-image-turbo command.

FLUX.2 Klein Model

Source: Black Forest Labs Sizes: 4 B (Apache 2.0) and 9 B (FLUX Non‑Commercial License v2.1) Key strength: reliable readable text in generated images, useful for UI mockups and typographic designs.

Example Prompts and Results

• Text rendering: "A neon sign reading "OPEN 24 HOURS" in a rainy city alley at night, reflections on wet pavement."
• Product photography: "Matte black coffee tumbler on wooden desk, morning sunlight casting long shadows, steam rising, commercial product shot."

Both prompts demonstrate the model’s ability to place crisp, legible text within complex scenes.

Configuration Options

Ollama provides terminal commands to fine‑tune generation:

• Image location: Images are written to the current directory; change directories beforehand to store elsewhere.
• Image size: Use /set width and /set height to adjust dimensions; smaller sizes generate faster and consume less memory.
• Number of steps: Controls iteration count; fewer steps speed up generation at the cost of detail, while excessive steps may introduce artifacts. Defaults are model‑specific.
• Random seed: Setting a seed yields reproducible outputs, facilitating iterative design work.
• Negative prompts: Specify undesired elements to steer the model away from them.

Roadmap

• Extend native image generation to Windows and Linux.
• Add more text‑to‑image models and introduce image‑editing capabilities.

---

All images and prompts shown above are directly taken from Ollama’s official blog post dated 2026‑01‑20.