1. Workflow Overview

This workflow generates high-quality cherry blossom scenes using SDXL model, emphasizing vibrant colors and natural details.

Key Features:

• SDXL-based high-res (752x1192) image generation

• Precise scene control via text prompts

• Batch generation (3 images per run)

2. Core Model

3. Key Nodes

• CheckpointLoaderSimple (Node 4)
  ▶ Loads SDXL model (LEOSAM HelloWorld 新世界 | SDXL大模型_v7.0.safetensors)

• CLIPTextEncode (Node 6/7)
  ▶ Positive Prompt:

  A cherry blossom tree... [detailed scene description]  

  ▶ Negative Prompt: text, watermark

• KSampler (Node 3)
  ▶ Critical Settings:

  • Sampler: Euler

  • Steps: 70 (high precision)

  • CFG Scale: 8 (strong prompt adherence)

4. Workflow Structure

1. Model Load:

   • Load SDXL + VAE (Node 4)

2. Latent Space:

   • Create blank latent image (752x1192 x3) (Node 5)

3. Text Encoding:

   • Encode scene prompts (Node 6)

   • Filter artifacts (Node 7)

4. Generation:

   • High-step sampling (Node 3)

5. Output:

   • Images saved as ComfyUI_XXXX.png (Node 9)

5. Input/Output

• Inputs:

  • Built-in prompts (modify Node 6 for customization)

• Outputs:

  • Resolution: 752x1192

  • Batch: 3 images

  • Path: Default ComfyUI output folder

6. Notes

⚠️ VRAM: 12GB+ required for SDXL
⚠️ Generation Time: ~2-3 mins/image (70 steps)
💡 Tips:

• Reduce steps to 30-50 for faster generation

• Adjust dimensions in Node 5 for different resolutions