ComfyUI.org

Unlock the Secrets of Tang Dynasty Portraits: A Comprehensive Workflow Guide

CN
ComfyUI.org
2025-04-17 08:50:37

1. Workflow Overview

m9l4eaf0no0eobg6bec54c328ca09d1f7e96b6c698f6ad3a231be6c7235b41cb0b48b4c85baf60312c.png

  • Purpose:
    Generates Tang Dynasty Gongbi-style portraits using:

    • LoRA Model (唐风工笔娃娃_v1.0) for traditional brushwork details.

    • Custom Base Model (基础算法_F.1) at 1024x1920 resolution.

    • Dual CLIP Encoders (clip_l + t5xxl_fp8_e4m3fn) for prompt accuracy.

2. Key Nodes

  • LoraLoader: Applies 唐风工笔娃娃_v1.0 (weight 1.0).

  • DualCLIPLoader: Combines two text encoders for nuanced prompts.

  • CLIPTextEncode:

    • Positive: Describes Hanfu-clad girls, butterflies (e.g., "Gongbi painting").

    • Negative: Filters artifacts (e.g., "badly drawn hands").

  • KSampler: euler, 20 steps, CFG 3.5, denoise 0.8.

3. Workflow Structure

  1. Model Loading:
    UNETLoader → LoraLoader → DualCLIPLoader.

  2. Prompt Processing:
    Dual CLIPTextEncode for conditioning.

  3. Image Generation:
    EmptyLatentImage → KSampler → VAEDecode.

4. Inputs & Outputs

  • Inputs: Positive prompt, resolution (1024x1920).

  • Outputs: Gongbi-style image saved to ComfyUI/output.

5. Notes

  • Dependencies:

    • Place 唐风工笔娃娃_v1.0.safetensors in models/loras.

  • VRAM: High resolution may require ≥8GB GPU.

  • Troubleshooting:

    • Verify LoRA filename matches.

previous
From Reference to Reality: AI-Driven Character Style Transfer and Prompt Reverse Engineering
next
From Faces to Masterpieces: A Professional AI Portrait Workflow
Share to

Recommend