Unlock the Secrets of Traditional Chinese Art: A One-Click Landscape Painting Workflow

CN
ComfyUI.org
2025-06-10 13:33:00

1. Workflow Overview

mbqk9834o8dm1e358p图片压缩d2cd9d35b44ddc6521a95b6592e71885b7cca762caad489ed836b1ac50498249.png

This workflow, named "One-Click Chinese Style Landscape Painting", transforms input images into traditional Chinese blue-green landscape art with gold-accented effects.

  • Core Models:

    • Stable Diffusion: Custom checkpoint 小资禅意山水-青绿山水人物-鎏金人物山水-国风插画系列_1.0.

    • ControlNet: control_v11p_sd15_canny (edge-based structure control).

  • Function: Image style transfer + detail enhancement.

2. Key Nodes

Node Name

Description

Installation

CheckpointLoaderSimple

Loads the custom Chinese-style SD model.

Manually place .safetensors in models/checkpoints.

ControlNetLoader

Loads Canny edge ControlNet model.

Download .pth to models/controlnet.

CannyEdgePreprocessor

Extracts edge maps from input images for ControlNet.

Install ControlNet Preprocessors via ComfyUI Manager.

CLIPTextEncode

Encodes text prompts into embeddings.

Built-in.

KSampler

Controls generation (steps: 25, CFG: 7, sampler: dpmpp_2m).

Built-in.

VAEDecode

Decodes latent images to pixels.

Built-in.

ClipInterrogator

Auto-generates prompts from input images (optional).

Install via ComfyUI Manager.

ImageConcanate

Concatenates input/output images for comparison.

Install custom node (e.g., ComfyUI-Image-Concatenate).

3. Workflow Structure

  • Group 1: Model & Preprocessing

    • Nodes: CheckpointLoaderSimple, ControlNetLoader, CannyEdgePreprocessor.

    • Input: Source image (via LoadImage).

    • Output: Edge map + loaded models.

  • Group 2: Generation

    • Nodes: CLIPTextEncode, KSampler, VAEDecode.

    • Input: Prompts, edge map, model params.

    • Output: Generated landscape image.

  • Group 3: Post-Processing

    • Nodes: ImageConcanate, SaveImage, PreviewImage.

    • Input: Source + generated images.

    • Output: Side-by-side comparison + saved result.

4. Inputs & Outputs

  • Inputs:

    • Resolution: 1104x1432 (set in EmptyLatentImage).

    • Prompts: Auto-generated by ClipInterrogator or manual input.

    • Sampling: Steps 25, CFG 7, Sampler dpmpp_2m.

  • Output:

    • Final image: Blue-green landscape with gold accents (PNG).

5. Notes

  • Model Dependencies:

    • Download 小资禅意山水 model and control_v11p_sd15_canny.

    • Paths: ComfyUI/models/checkpoints and models/controlnet.

  • Hardware:

    • ≥8GB VRAM recommended (e.g., RTX 3060+).

  • Troubleshooting:

    • Edge detection issues: Adjust thresholds in CannyEdgePreprocessor (default: 100/200).

    • Style mismatch: Tweak prompts or reduce ControlNet weight (in CR Apply ControlNet).