ComfyUI.org

From Fabric to Leather: Mastering Material Style Transfer with AI

CN
ComfyUI.org
2025-05-23 08:55:02

1. Workflow Overview

mb0ke82g5x4bloz8f4jb1b216cbe15ca23ecc2ee6880f7fb8a7b8d0fd84dfb5ff5ddbb9da73747f4138.png

  • Purpose:
    This workflow replaces materials in specified image regions (e.g., fabric, leather) using manual masking, leveraging IPAdapter for style transfer and ControlNet for structural consistency.

  • Core Models:

    • juggernautXL_v9-Lightning_4S_V9 + RDPhoto 2: Base model for photorealistic generation.

    • IPAdapter: Transfers material style from reference images.

    • ControlNet (sai_xl_depth_256lora): Constrains structure via depth maps.

    • DepthAnythingPreprocessor: Generates depth maps for ControlNet.

2. Key Nodes

Node Name

Function

Installation

ImageCompositeMasked

Blends generated material with original image using masks.

Built-in

IPAdapterAdvanced

Applies material style (default weight: 0.85).

Install IPAdapter plugin

ControlNetLoaderAdvanced

Loads depth ControlNet for structure preservation.

Install ControlNet plugin

DepthAnythingPreprocessor

Requires model depth_anything_vitl14.pth.

Download from HuggingFace

Dependency

Place ip-adapter_sdxl_vit-h.safetensors in models/ipadapter.

CivitAI/official repository

3. Workflow Structure

  • Group 1: Image Input

    • Inputs:

      • Original image (e.g., furniture).

      • Material reference image.

      • Manually painted mask (right-click to edit).

    • Outputs: Cropped/aligned images and masks.

  • Group 2: Material Processing

    • Steps:

      1. Encodes material style via IPAdapter.

      2. Generates depth maps with DepthAnything for ControlNet.

      3. Smooths mask edges using GrowMask and FeatherMask.

  • Group 3: Generation & Composite

    • Generation:

      • KSampler merges IPAdapter and ControlNet outputs.

      • ImageCompositeMasked blends results with the original image.

    • Output: Final image (saved to ComfyUI/output).

4. Inputs & Outputs

  • Input Parameters:

    • Resolution: Default 1024x1024 (adjustable).

    • Seed: Fixed or randomized.

    • IPAdapter weight: Default 0.85 (controls style strength).

  • Output:

    • Image with replaced material (PNG).

5. Notes

  • Common Errors:

    • Missing IPAdapter/ControlNet models triggers file path errors.

    • Incorrect masking causes misalignment.

  • Optimization:

    • Tweak FeatherMask radius for natural edges.

    • Reduce IPAdapter weight (e.g., 0.6) to avoid overfitting.

  • Hardware:

    • Recommended 12GB+ VRAM (due to IPAdapter + ControlNet).

previous
Mastering Light Trails and Particle Dispersion: A Comprehensive Workflow

Recommend