Z-Image Turbo for Beginners on an 8 GB Mac (vs SDXL Turbo)

Short answer: Open the Z-Image Turbo CosyTemplate in Promptus and generate from its prompt box, from Playground, or on the ComfyUI canvas. On a Mac with 8 GB of unified memory, use a low-memory recipe — compressed models and a memory helper — and start small: 256 pixels first, then 512. To try another model like SDXL Turbo, open its own template; models don’t share recipes.

Three ways to start

Route	How	Best for
Template page	Open the Z-Image Turbo CosyTemplate, type in its prompt box, choose Local, Generate	The quickest picture
Playground	Template menu › Run offline in Playground; check the mode says Local	Describing an image without touching nodes
ComfyUI canvas	Template menu › Load in ComfyUI, then Fit view	Seeing and changing every step

Whichever route you choose, the required models and nodes must be installed, and the recipe has to fit in your computer’s memory.

A simple map of the canvas

The template opens as one compact block; click the open icon on its top-right corner to see the steps inside. Three loaders supply the ingredients:

Z-Image Turbo — the model that builds the image.
Qwen — the text encoder that reads your words.
VAE — turns the finished latent into pixels.

For the 8 GB Mac the video switched to a low-memory turbo recipe: compressed models plus an extra node that releases loaded weights before sampling.

The settings that matter

Setting	Value in the video	Why
Prompt	Subject, material, setting, light source	Gives you something to check the result against
Size	256×256 first, then 512×512	Doubling the side needs four times the pixels — test small first
Seed	42, fixed	Keep it fixed while changing one detail; randomise to start fresh
Steps / CFG	8 steps, guidance 1	More steps are not always an upgrade
Batch	1	One image per run on a small machine

The first prompt — a yellow ceramic robot on a blue desk, lit by a window — gave a 256-pixel image that matched every choice. The second — a glossy red robot beside a green succulent at 512 pixels, same seed — is a new example, not a pure resolution comparison, because the prompt changed too.

Switching to SDXL Turbo

Don’t swap the file name inside the Z-Image workflow — open the SDXL Turbo CosyTemplate and Load in ComfyUI. Its canvas looks different: one checkpoint loader supplies the model, text encoders and VAE together, and it runs in a single step at 512×512, guidance 1.

	Z-Image Turbo	SDXL Turbo
Loaders	Separate model, Qwen text encoder and VAE	One checkpoint with all three
Steps	8	1
Guidance	1	1
Same prompt and seed 42	Red robot, succulent	Same scene, different composition

Matching the seed across models doesn’t give matching pictures. Even with one step, loading the model still takes time.

Keep the picture and the recipe

Open Assets › Generated to inspect and download the image.
Export the workflow from the graph menu so you can rerun it later.
Both workflows stay open as tabs — click a tab to choose which recipe runs.

The same 8 GB limits are tested more widely in what runs on an 8 GB MacBook, and the ControlNet side of Z-Image is in Z-Image Turbo ControlNet: Canny settings. Download Promptus.

Frequently Asked Questions

Can Z-Image Turbo run on an 8 GB Mac?

In this video, yes, using a low-memory Z-Image Turbo recipe with compressed models and an extra memory helper node. It generated a 256-pixel test image and then a native 512-pixel image locally.

What settings does Z-Image Turbo use?

The recipe in the video used 8 steps and a guidance (CFG) of 1, with the sampler and scheduler kept together as set in the template.

Why can’t I use the same workflow for Z-Image and SDXL Turbo?

Different models need different recipes. Z-Image uses separate loaders for the model, the Qwen text encoder and the VAE; SDXL Turbo loads all three from one checkpoint and runs in a single step.

Does the same seed give the same picture in different models?

No. A fixed seed keeps results steady within one workflow, but matching the seed across different models does not give matching pictures.

What are the three ways to generate in Promptus?

From the CosyTemplate page’s own prompt box, from Playground with Run offline in Playground, or on the ComfyUI canvas with Load in ComfyUI.

Written by:

Claudia Perez

Claudia Perez is a tech creator and founder at Promptus. She loves to design and share expressive image workflows. Her work focuses on visual storytelling, refined aesthetics, and ComfyUI workflows.

Try Promptus Cosy UI today for free.