Summary

This ComfyUI workflow, Amazing Z-Image Workflow v4.0, enhances Z-Image-Turbo for high-quality, stylized image generation. It targets users seeking a user-friendly experience with pre-configured GGUF and SAFETENSORS setups, integrating refinement and upscaling.

How It Works

The workflow expands ComfyUI with an 18-style selector, a refiner for quality, and a 50% resolution upscaler, featuring photo/illustration optimization modes. It offers distinct configurations for GGUF checkpoints (recommended for <=12GB VRAM) and SAFETENSORS (potentially better ComfyUI optimization, may need more VRAM). It includes a "Power Lora Loader" and automatic CivitAI prompt detection.

Quick Start & Requirements

• Primary install / run command: Installation is via ComfyUI-Manager.
• Non-default prerequisites and dependencies: Required custom nodes include rgthree-comfy and ComfyUI-GGUF (for GGUF). Specific GGUF (e.g., z_image_turbo-Q5_K_S.gguf) and SAFETENSORS (e.g., z_image_turbo_bf16.safetensors) checkpoints are detailed with sizes/directories. GGUF is suggested for <=12GB VRAM; SAFETENSORS may require 12GB+ VRAM. Links to low-VRAM alternatives are provided.

Highlighted Details

• 18 customizable image styles.
• Integrated Refiner and 50% resolution Upscaler.
• Photo/illustration optimization modes.
• Speed (7 Steps) and Smaller Image (1216x832) switches.
• Alternative Sampler, Landscape Orientation, and experimental "Spicy Impact Booster."
• Preconfigured GGUF (low VRAM) and SAFETENSORS workflows.
• "Power Lora Loader" for multiple LoRAs.
• Automatic CivitAI prompt detection.

Maintenance & Community

Acknowledges contributions from Tongyi-MAI Team and rgthree. No specific community channels or roadmap details are provided in the README.

Licensing & Compatibility

Licensed under the Unlicense, a permissive public domain equivalent license, generally allowing commercial use and integration into closed-source projects.

Limitations & Caveats

Checkpoint choice (GGUF vs. SAFETENSORS) depends heavily on system configuration (ComfyUI, PyTorch, CUDA, GPU, VRAM/RAM). Some quantizations may degrade output quality. The "Spicy Impact Booster" is experimental.