To run Z-Image in ComfyUI, import the official template and install its diffusion, text-encoder, and VAE files. Choose the matching Turbo or Base workflow. Turbo guide Base guide
What are the key facts about Z-Image in ComfyUI?
| Field | Verified information |
|---|---|
| Developer | Alibaba's Tongyi Lab develops Z-Image; ComfyUI documents the workflows. Turbo guide |
| Released | Turbo on November 26, 2025; Z-Image foundation weights on January 27, 2026. Project repository |
| Type | Text-to-image model family with official ComfyUI workflows. Turbo guide Base guide |
| Size or parameters | 6 billion parameters for the image model. Turbo model card |
| License and access | Apache 2.0 weights on Hugging Face. Turbo card Base card |
| Where it runs | Local ComfyUI or the documented Comfy Cloud workflow. Turbo guide |
What can the official Z-Image workflows generate?
ComfyUI's official templates make the model components and generation settings inspectable. The supplied Turbo workflow names the files it loads and includes its sampling configuration. This gives you a concrete baseline to save before making changes. [Official Turbo template][template]
The model developer highlights photographic generation, English and Chinese text rendering, and instruction following for Turbo. These are useful categories for evaluating your first outputs; they do not establish that every requested word or relationship will render correctly. Turbo model card
Save the unchanged official workflow before editing it. For a first prompt, choose one object on a plain background; add complexity after the graph completes successfully.
The Z-Image Base guide explains the undistilled model's guidance and negative-prompt controls. That is the next useful step when your goal moves from establishing a working graph to exploring more detailed generation settings.
Why do Z-Image settings and ComfyUI versions matter?
Workflow availability depends on your installed ComfyUI version. The official documentation identifies outdated installations and failed node imports as possible reasons for missing nodes, and notes that stable releases can lag the newest documented support. Check the startup log and release version before adding unrelated extensions. Turbo workflow guide
Sampling settings belong to the selected model. The fetched Turbo template uses eight steps, CFG 1, res_multistep, and the simple scheduler. Preserve the downloaded template's settings for the first run, since template maintainers may update them. [Official Turbo template][template]
Do not transfer a ComfyUI CFG number mechanically into a Python example. The Turbo model card's Diffusers recipe sets guidance_scale=0.0; the graph and pipeline expose their own controls. Use the documented recipe for each implementation. Turbo model card
Hardware fit is another separate question. The developer describes Turbo running within 16 GB of consumer GPU memory under its supported setup, while its subsecond latency claim refers to an H800. That claim does not predict the latency of a particular home computer. Project repository
How do you install Z-Image model files in ComfyUI?
Update ComfyUI using the instructions for your installation, then open the official Z-Image-Turbo text-to-image template. You can obtain its JSON from the workflow guide or download the template file directly with this command. Turbo guide [Official template][template]
curl -L https://raw.githubusercontent.com/Comfy-Org/workflow_templates/main/templates/image_z_image_turbo.json \
-o image_z_image_turbo.json
Load that JSON into ComfyUI. Follow the model download links in the official guide, then place the files in these locations: Turbo workflow guide
| Component | File location inside ComfyUI |
|---|---|
| Diffusion model | models/diffusion_models/z_image_turbo_bf16.safetensors |
| Text encoder | models/text_encoders/qwen_3_4b.safetensors |
| VAE | models/vae/ae.safetensors |
In the fetched graph, the model loader selects the Turbo weights, the CLIP loader selects the Qwen file with type lumina2, and the VAE loader selects ae.safetensors. Check those selections before running the graph. [Official Turbo template][template]
Write a modest first prompt: “A yellow ceramic bowl on a wooden kitchen table, side window light, close view.” Run the unchanged template and inspect the saved output. At this stage, success means that the intended files load and the graph completes; defer creative judgments until that baseline is established.
Duplicate the working graph before changing the prompt. Name each saved version after its change so that you can compare it with the original.
If a loader reports a missing file, compare its selected filename with the actual directory entry. Check whether the download completed and whether you placed it in the component-specific folder. Avoid changing the sampler while investigating a file-loading error; write down the error and resolve that stage first.
For missing nodes, check the installed ComfyUI version and startup import errors. Follow the update guide linked from the official workflow documentation. Turbo workflow guide
For the undistilled model, load the separate Base template and use z_image_bf16.safetensors in the diffusion-model folder. Its guide specifies the same Qwen text-encoder and VAE filenames. Use the Base workflow as a complete starting point instead of replacing a single loader in your Turbo graph. Base workflow guide
How do Z-Image Turbo and Base workflows differ?
The immediate alternative to Z-Image-Turbo is Z-Image Base. The Base card documents classifier-free guidance and negative prompting, with a recommended schedule of 28–50 steps; Turbo targets a short distilled schedule. Choose based on the controls your task needs, then assess output quality using the same brief. Base model card
The ComfyUI pillar explains the broader workflow concepts. Keep a clean copy of the official graph alongside your customized version so that you always have a known starting configuration.
What should you check before running Z-Image in ComfyUI?
How does Z-Image work with ComfyUI?
Z-Image is Alibaba Tongyi's image-model family. ComfyUI runs its weights through official templates with separate diffusion-model, text-encoder, and VAE loaders. Turbo guide Base guide
Which files does the Z-Image-Turbo workflow require?
Z-Image-Turbo uses z_image_turbo_bf16.safetensors, qwen_3_4b.safetensors, and ae.safetensors. Place them in models/diffusion_models, models/text_encoders, and models/vae, respectively. Guide
Can I reuse Z-Image-Turbo settings with Base?
Start Z-Image Base with its own template. Its model card recommends 28–50 steps and guidance of 3.0–5.0; the Turbo template uses eight steps and CFG 1. Base [Template][template]
Why is my Z-Image workflow missing nodes?
Missing Z-Image nodes can result from an outdated ComfyUI version or failed imports at startup. Check the startup log and update instructions; stable releases can lag new node support. Guide
Sources
- Project repository
- Turbo model card
- Base card
- Turbo workflow guide
- Base workflow guide
- [Official Turbo template][template]
[template]: https://raw.githubusercontent.com/Comfy-Org/workflow_templates/main/templates/image_z_image_turbo.json<!-- pz-related-guides -->
Top comments (0)