Running the Z-Image Workflow Locally

This guide explains how to run Alibaba’s Z-Image workflow inside Promptus, providing a streamlined, no-code experience for ComfyUI users. Whether you are using the Promptus Web interface or the Desktop App, follow these steps to master high-performance image generation.

📋 1. Prerequisites

Before you start, ensure your environment meets these requirements:

• App: Latest version of Promptus.
• Hardware: 12–16GB VRAM GPU recommended (Quantized GGUF versions can run on 4GB VRAM).
• Automation: Promptus automatically handles workflow loading, model downloading, and file management so you can focus on creating.

📂 2. Loading the Z-Image Workflow

Promptus ships one‑click CosyFlows templates for Z-Image — every one currently in the catalog, so you can pick the exact variant for your hardware and task instead of hand-building a workflow:

• Z-Image-Turbo: Text to Image — the efficient single-stream diffusion transformer variant, built for speed (6–9 steps). Supports both English and Chinese prompts; the fast, general-purpose starting point.
• Z-Image: Text to Image — the non-Turbo foundation model. Slower than Turbo, but offers more diverse aesthetics, higher photorealistic quality, stronger response to negative prompts, and is the better base if you plan to fine-tune.
• Z-Image-Turbo: Fun Union ControlNet — adds multi-control guidance on top of Turbo: Canny edges, HED, Depth, Pose, or MLSD, so you can constrain composition instead of relying on prompting alone.
• Promptus: Z-Image Text to Image [bf16] — the full-precision preset. Highest quality output, but needs real GPU headroom; on 8GB Macs, use one of the GGUF presets below instead.
• Promptus: Z-Image Text to Image [GGUF-Q2] — the smallest quantized preset (z-image-Q2_K.gguf), tuned for 512×512 output in 4 steps. Fastest and lightest on VRAM, at the cost of some detail versus Q4/Q8.
• Promptus: Z-Image Text to Image [GGUF-Q4] — the mid-size quantized preset (z-image-Q4_K_M.gguf). A balance point: noticeably better detail retention than Q2, still far lighter than the full bf16 model.
• Promptus: Z-Image Text to Image [GGUF-Q8] — the largest quantized preset (z-image-Q8_0.gguf). Closest to bf16 quality among the GGUF options, at the cost of more VRAM and slower generation than Q2/Q4.
• Text to Image — the Z-Image-Turbo BF16 preset pulled from Promptus's Getting Started library. Same model as the bf16 preset above, packaged as a first-run template if you're new to Z-Image.

Steps:

1. Open Promptus.
2. Navigate to the CosyTemplates tab and search “z-image.”
3. Pick the variant above matching your hardware, then click Load in ComfyUI or Run Online in Playground.
4. The visual pipeline (UNet Loader, CLIP Text Encode, KSampler, etc.) will appear on your canvas, fully configured.

📥 3. Model Configuration

Promptus detects missing models and prompts you to download them automatically.

💡 Tip: Simply click "Download All Models" and Promptus will place them in the correct directories for you.

✍️ 4. Prompting & Sampling

To get the best results from the Z-Image architecture, use these optimized settings:

• Positive Prompt: "A cinematic portrait of a young woman standing in warm sunset light, shot on a 50mm lens, ultra-realistic, detailed skin texture."
• Steps: 6–9 (Z-Image Turbo is highly efficient).
• CFG: 1.0–3.0.
• Sampler/Scheduler: Euler or Euler A / Simple.

🖼️ 5. Image-to-Image (img2img) Capabilities

Transform existing visuals using the dedicated img2img workflow:

1. Add your image via the Load Image node.
2. Adjust Denoise Strength:
   • 0.2–0.4: Subtle retouching.
   • 0.5–0.7: Medium style changes.
   • 0.8–1.0: Heavy transformation (e.g., Anime to Realism).

🎨 6. Using LoRAs

Enhance your style or characters by adding LoRAs:

1. Add a LoRA Loader node after the UNet.
2. Set strength between 0.5–1.0.
3. Include specific trigger words in your positive prompt.
4. Note: You can chain multiple LoRAs for combined effects.

✨ Best Practices for Success

• Keep Prompts Concise: Z-Image understands complex concepts without "prompt salad."
• Camera Styles: Use keywords like 50mm, IMAX, or Fujifilm to influence the render's "feel."
• GPU Optimization: If you experience slow renders, switch to the Promptus GPU Marketplace for high-end cloud power.

🏁 Conclusion

You are now ready to run Z-Image without a complex manual ComfyUI setup. Promptus provides the infrastructure; you provide the vision.