🎨 R-Image-Base

Where Z-Image Base Meets the Signature Rebel Aesthetic.

---

🚨 UPDATE: GGUF Model Now Available!

The GGUF version is officially live. To get it running, follow these steps:

• Custom Node Required: Download and extract the Rebels_RIB_Nodes folder, Place it inside your custom_nodes directory.
• Custom Node Required: city96s gguf custom node pack 'ComfyUI-GGUF'. (download in comfyui manager or github)
• Required Files: You must download the provided text encoder (qwen_3_4b_fp8_mixed) and VAE (ae) if you do not already have them in your text encoder and vae folders.
• Plug & Play: The node is configured to call all necessary files automatically.
• Important: Please USE THE PROVIDED WORKFLOW to ensure everything scales and loads correctly.

---

📌 About the Model

R-Image-Base is a merged checkpoint consisting of Z-image Base and a custom selection of LoRAs. It displays a heavy focus on portraits with extreme photorealistic detail and aesthetic enhancement, taking the foundation of Z-Image and pushing it into a high-fidelity space.

⚙️ Recommended Settings

For the best results, stick to these parameters:

Setting	Recommended Value
Resolution	2048 x 2048
Steps	30 - 50
CFG Scale	4.0 - 5.0

---

🛠️ Installation

Option 1: Full Weight Installation (Standard)

Use this if you want the uncompressed, full-fidelity experience.

• Download: The full weight Checkpoint file (place in checkpoints folder).
• Workflow: Use the provided JSON bf16 workflow for the best results.

Option 2: GGUF Installation (Optimized)

Use this for faster performance or lower VRAM usage.

• Download: the zip file 'Rebels_RIB_Nodes'. Place the Rebels_RIB_Nodes folder inside your custom_nodes directory.
• Requirements: You must also have the text encoder (qwen_3_4b_fp8_mixed) and the VAE (ae) downloaded.
• Workflow: You must use the provided JSON gguf workflow for the GGUF node to function correctly.

---

Developed by Rebel AI ```

EXAMPLES:

Model Description

This is a merged checkpoint of Z-image Base and 2 of my own custom trained LoRAs. values are what i would consider perfect for any image generation and the loras were trained to be detail enhancing.