1. Workflow Overview

This workflow, named "One-Click Realistic to Fantasy Ancient Male Portrait", is designed to automatically transform real-life photos into ancient fantasy-style male character illustrations. It combines Stable Diffusion models with LoRA fine-tuning, leveraging image recognition, style transfer, and detail enhancement to achieve high-quality anime-style results.

2. Core Models

• Base Model: 基础算法_F.1 (Custom Stable Diffusion variant)

• LoRA Model: F.1-【超人】最强古风玄幻推文_V1 (Optimized for ancient male characters)

• CLIP Model: t5xxl_fp8_e4m3fn (Enhances text comprehension)

• VAE Model: ae.sft (Handles image encoding/decoding)

• Image Tagger: WD14Tagger (Auto-generates tags from input images)

3. Key Nodes

4. Workflow Structure

1. Input & Preprocessing

   • Input: Real-life photo (loaded via LoadImage)

   • Processing:

     • WD14Tagger extracts tags (e.g., clothing, pose)

     • Auto-generates positive/negative prompts (e.g., Simpsons style, ng_deepnegative_v1_75t)

2. Model Loading

   • Loads base model, LoRA, and CLIP models.

   • Configures VAE (ae.sft).

3. Image Generation

   • Generates images via KSampler (30 steps, Euler sampler, CFG=3.5).

4. Output

   • Final image saved as PNG (ComfyUI/output).

5. Inputs & Outputs

• Input Parameters:

  • Required: Real-life photo (PNG/JPG)

  • Optional: Seed value (default: random), sampling steps (default: 30)

• Output: Ancient fantasy-style male character illustration.

6. Notes

1. Dependencies:

   • LoRA models must be manually downloaded to models/loras.

   • WD14Tagger requires installation via ComfyUI Manager.

2. Hardware:

   • Recommended: ≥8GB VRAM (NVIDIA GPU).

3. Troubleshooting:

   • Adjust KSampler CFG or switch LoRA models if results are unsatisfactory.