Ollama Image Generation: Can It Make Images? (Nov 2026)

No — not on a current build. As of 2 November 2026 (Ollama v0.40.13), ollama run x/z-image-turbo answers image generation models are not currently supported. Ollama shipped text-to-image as a macOS-only experiment on 20 January 2026 with two models, x/z-image-turbo and x/flux2-klein, then "temporarily removed" it in v0.32.6 on 4 August 2026. No release since — v0.34 through v0.40.13 — has brought it back. The only way to generate an image with Ollama today is to run Ollama 0.32.5 on a Mac; it has never worked on Windows or Linux. For actual local image generation on any OS, use ComfyUI or Forge with the same FLUX.2 Klein and Z-Image weights — our FLUX.2 local setup guide is the walkthrough.

This page keeps the full model specs, commands and memory tables from when the feature worked, because they still apply on 0.32.5 and because the same two models run in ComfyUI. It also answers the question most people arrive with, which is simply whether "Ollama does images" — and today that means image understanding (vision models), not image generation.

Did Ollama always generate images? (Clearing up the misconception)

No. This trips a lot of people up, so let's be precise. Before January 2026, Ollama ran language models and vision models — meaning it could read an image you handed it (with multimodal models like Gemma 4, Qwen3-VL or Llama 3.2 Vision) and describe it. It could not create images. There was no text-to-image in Ollama at all.

The 20 January 2026 release added experimental image generation as a distinct feature, for macOS only. Six and a half months later, v0.32.6 took it out again. So the honest timeline is:

If you remember someone saying "Ollama does images," before 2026 they meant image understanding, during February–July 2026 they may have meant generation on a Mac, and today they mean understanding again. Vision models are covered in the best Ollama models guide; this page is about generation.

What happened in v0.32.6, and is it coming back?

The v0.32.6 release notes say only that the feature was "temporarily removed" and that 0.32.5 keeps it. The maintainer who wrote the Windows support PR explained the reason when closing it: image generation was "temporarily dropped" so the team could focus on maturing the MLX engine — the runtime behind Ollama's Apple Silicon builds, which the v0.40.13 release is now also packaging for Linux and Windows (ollama-linux-amd64-mlx, ollama-windows-amd64-mlx.zip). Nobody at Ollama has published a return date, and the model pages still carry the January note that "Image generation models currently only work on macOS."

What that means in practice:

• If you are on a Mac and want it today: install Ollama 0.32.5 and do not accept the "Restart to update" prompt — Ollama's FAQ says the macOS app downloads updates automatically, so the moment you restart into a newer build the runner is gone again. You also lose every model and fix shipped after 27 July 2026 (Qwen3.8, Gemma 4 updates, LFM3, the MLX speed work).
• If you are on Windows or Linux: there is nothing to install. Image generation never ran there. Skip to Ollama image generation on Windows for what to use instead.
• If you can wait: watch the releases page; the models stay in the library, so when the runner returns the commands below should work unchanged.

What models can Ollama generate images with in 2026?

There are two, both in Ollama's x/ (experimental) namespace, and both still listed with download sizes read from the library on 2 November 2026 (every tag shows "9 months ago" — nothing has been touched since launch).

A few details that matter:

• Z-Image-Turbo is a few-step "turbo" model. Ollama's model page describes it as "a 6 billion parameter text-to-image model" and says "1024x1024 resolution is recommended." Its standout feature is text rendering in English and Chinese — text inside the image actually reads correctly, which most open models botch.
• FLUX.2 Klein comes in two sizes. The 4B is the default (x/flux2-klein and x/flux2-klein:4b are the same 5.7 GB build) and the one to start with: small, fast, and Apache 2.0 so you can use outputs commercially. The 9B is sharper but carries Black Forest Labs' FLUX Non-Commercial License v2.1, so do not ship its outputs in a paid product without a commercial agreement.
• Every size has fp4, fp8 and bf16 tags (x/flux2-klein:4b-fp8, x/flux2-klein:9b-bf16, x/z-image-turbo:bf16) if you want to trade download size for precision.

The same two model families run outside Ollama with full control: our FLUX.2 local setup guide covers Klein in ComfyUI on any OS, and the Z-Image-Turbo in ComfyUI walkthrough does the same for Z-Image.

How do I generate images with Ollama? (Exact commands)

These commands work on macOS running Ollama 0.32.5 or earlier (back to the 20 January 2026 release that introduced the feature). On any newer build they return the "not currently supported" error, and on Windows and Linux they have never run. The model downloads on first run.

# Z-Image-Turbo (Alibaba Tongyi, 6B) — quality + bilingual text
ollama run x/z-image-turbo "a cat holding a sign that says hello world"

# FLUX.2 Klein 4B (Black Forest Labs) — fast, commercial-friendly
ollama run x/flux2-klein "a neon-lit Tokyo street at night, photorealistic"

# FLUX.2 Klein 9B — higher fidelity (non-commercial license)
ollama run x/flux2-klein:9b "a watercolor fox in a misty forest"

By default the generated image is saved to your current directory. If your terminal supports inline image rendering — Ghostty, iTerm2 and similar — the picture also previews right in the terminal window, no external viewer needed.

Inside an interactive session you can tune generation with slash commands:

# Inside an interactive run:
/set width 1024      # output width in pixels
/set height 1024     # output height in pixels

Ollama's launch post adds that "smaller images generate faster and use less memory," that fewer steps are faster but less detailed while "too many steps can cause artifacts" (Ollama defaults to each model's recommended count), that a fixed seed gives reproducible results, and that a negative prompt excludes unwanted elements. The official details live on the Ollama image generation blog post and the x/z-image-turbo model page.

How much VRAM / unified memory does Ollama image generation need?

Because the feature only ever ran on macOS, the practical resource is Apple Silicon unified memory (the same pool the GPU and CPU share). Use the download size as your floor and add headroom — the model has to live in memory while it runs. These are practical targets, framed approximately:

The takeaway: a 16 GB Apple Silicon Mac comfortably runs FLUX.2 Klein 4B and can handle Z-Image-Turbo fp8 with little else open; for the bf16 full-precision Z-Image you really want 36 GB+ of unified memory. The same files in ComfyUI on a Windows PC want the same amount of VRAM — the best local image models compared page has the per-card fit for FLUX.2 Klein, Z-Image and the SD family. To sanity-check any model against your machine before downloading 13-33 GB, our VRAM calculator is the quickest gut-check.

How fast is Ollama image generation on a Mac?

There is no controlled cross-machine benchmark published for these two models inside Ollama. Rather than quote a second-count that would not transfer to your Mac anyway, here is the arithmetic that bounds it — you can run it for your own machine in about ten seconds.

A diffusion model is streamed through memory once per sampling step, so the floor on generation time is set by memory bandwidth:

seconds per image (floor) = steps × (model size in GB ÷ memory bandwidth in GB/s)

Two inputs. Steps: Ollama uses each model's recommended step count by default; both are distilled few-step models, so single digits is the right order of magnitude — eight is used below as a worked figure, not a measurement. Bandwidth: your chip's unified-memory bandwidth is a fixed, published number; look it up on the tech-spec page for your exact Mac.

Worked example at 120 GB/s (substitute your own chip's published figure and redo the division):

Read that as a floor, not a prediction. Real generation lands well above it: text encoding, the VAE decode pass, attention compute and macOS memory pressure all stack on top, and the very first run of a model includes the download and load. What the arithmetic does tell you reliably is the ratio — on the same Mac, the 4B is roughly twice as quick per step as Z-Image fp8 and around six times quicker than bf16. That ordering is what should drive your model choice. It also explains why the same model can be several times faster on a higher-bandwidth Mac with identical software: nothing changed except the divisor.

How is Ollama image generation different from ComfyUI?

This is the honest heart of the article, and the removal made the gap wider. Even when it worked, Ollama's image generation was fast to start and good for one-off prompts from the terminal, but intentionally minimal. ComfyUI (and Forge/A1111) are full image pipelines, they run everywhere, and they run today.

So when does each win?

• Use Ollama image generation only if you are on a Mac, are willing to stay on 0.32.5, want a quick image from the command line, and do not need fine control. It was the lowest-friction way to do ollama text to image; it is currently a pinned-version curiosity.
• Use ComfyUI or Forge for everything else — which today means everyone on Windows or Linux and anyone on a Mac who wants current Ollama for their language models. Our complete ComfyUI guide covers that workflow end to end, the FLUX.2 local setup guide gets Klein running in it, and the broader local FLUX image generation guide goes deeper on the FLUX family.

Ollama image generation on Windows

It has never worked on Windows. The launch post listed Windows and Linux as "coming soon"; a maintainer opened "win: Image Generation support" the day after launch, marked it ready for review on 25 January 2026, and closed it unmerged on 24 August 2026 after the feature was removed from macOS too. Pulling x/z-image-turbo on Windows downloads 13 GB and then fails at run time.

What a Windows user should do instead, in order of effort:

1. ComfyUI with FLUX.2 Klein 4B — the same Apache 2.0 weights Ollama used, with LoRA, ControlNet and img2img on top. The FLUX.2 local setup guide is the step-by-step; the fp8 build is a 9.5 GB file, so a 12 GB card has room for it.
2. ComfyUI with Z-Image-Turbo — the Z-Image-Turbo in ComfyUI walkthrough; the fp8 build is 13 GB, so budget 16 GB of VRAM.
3. Keep Ollama for the text side. Run your language and vision models in Ollama (the Windows install guide gets it set up in five minutes) and let ComfyUI own the images. The two coexist happily; they just use the GPU at different times.

Which is the best Ollama model for image generation?

For the narrow case where it still runs (macOS, Ollama 0.32.5): Z-Image-Turbo for quality, FLUX.2 Klein 4B for speed.

• Pick x/z-image-turbo if you want the most photorealistic output and especially if your images contain text (signs, logos, UI mockups, English or Chinese). Its bilingual text rendering is the single most differentiated thing in this lineup.
• Pick x/flux2-klein:4b if you want fast iteration, a small 5.7 GB download, and commercial usage rights (Apache 2.0). It is the friendliest starting point on a 16 GB Mac.
• Pick x/flux2-klein:9b only if you need extra fidelity and your use is non-commercial — its FLUX Non-Commercial License v2.1 rules out shipping outputs in a paid product without a separate agreement.

For everyone else the "best Ollama model for image generation" is the same weights loaded in ComfyUI, where the ranking is the same: Z-Image-Turbo for photorealism and text, FLUX.2 Klein 4B for speed and licence.

Key Takeaways

1. Ollama cannot generate images on a current build. The feature shipped 20 January 2026 (macOS only, experimental) and was "temporarily removed" in v0.32.6 on 4 August 2026; v0.40.13 (28 October 2026) still does not have it.
2. Ollama 0.32.5 on a Mac is the only version that works — and the app auto-updates, so you have to decline "Restart to update" every time.
3. It never worked on Windows or Linux. The Windows pull request was closed unmerged on 24 August 2026.
4. The two models are still in the library: x/z-image-turbo (Alibaba Tongyi, 6B, fp8 13 GB, Apache 2.0, bilingual text) and x/flux2-klein (Black Forest Labs, 4B 5.7 GB Apache 2.0 / 9B 12 GB non-commercial). Same weights run in ComfyUI on any OS.
5. It was never a ComfyUI replacement. No LoRA, no ControlNet, no inpainting/img2img. For serious or repeatable work, ComfyUI/Forge were always the tool — and today they are the only tool.

Next Steps

• Want local image generation that works today, on any OS? Start with the FLUX.2 local setup guide and the best local image models compared.
• Need real control — LoRAs, ControlNet, inpainting? Read the ComfyUI complete guide and our Z-Image-Turbo in ComfyUI walkthrough.
• Keeping Ollama for text and vision models? See the best Ollama models roundup (the vision section covers image understanding), the complete Ollama guide, and the Windows install guide.
• Going deeper on FLUX outside Ollama? The local FLUX image generation guide covers the full family.