ComfyUI.org

Unlock the Power of E-commerce Model Pose Transfer with This Advanced Workflow

CN
ComfyUI.org
2025-05-08 09:48:46

1. Workflow Overview

maf6pv8js0jr4wtblae4026bd18117375b8c998383fcfe01bdd76b02af477d84d6f210a7b47fcbec5f.jpg

This workflow specializes in e-commerce model pose transfer, accurately migrating poses from reference images to target characters (e.g., "Ari") while preserving clothing textures and skin details. Key features:

  • Multi-ControlNet: OpenPose + Depth + Canny for pose/depth/contour control

  • LoRA Fusion: Blends "ZOEY Real Skin" and "Ari E-commerce" LoRAs

  • Flux Architecture: Uses FP8-precision FLUX.1-ControlNet for stability

2. Core Models

Model Name

Function

Source

基础算法_F.1

Base model (FP8 optimized)

Load via UNETLoader

FLUX.1-ControlNet-Union-Pro

Triple-ControlNet (OpenPose+Depth+Canny)

Download .safetensors file

ZOEY Skin LoRA

Enhances skin realism (weight=0.4)

Place in models/loras

Ari E-commerce LoRA

Injects fashion photography style (weight=0.4)

Stack with skin LoRA

3. Key Nodes

Node Name

Function

Installation

Dependencies

ControlNetApplySD3

Next-gen ControlNet (multi-modal)

Update ComfyUI-ControlNet

FLUX-specific models required

AIO_Preprocessor

Unified preprocessor (Canny/OpenPose)

Install Impact-Pack

None

CLIPTextEncodeFlux

Flux-specific text encoder

Flux plugin required

Dual-CLIP models

easy positive

Structured positive prompt generator

Built-in

None

4. Workflow Structure

  • Group 1: Base Models

    • UNETLoader + DualCLIPLoader: Load FP8 base model

    • VAELoader: Uses ae.sft decoder

  • Group 2: LoRA Stacking

    • LoraLoader×2: Blends skin & style LoRAs (0.4 weight each)

  • Group 3: Multi-ControlNet

    • OpenPose: Extracts skeleton from reference image

    • DepthAnything: Generates depth map

    • CannyEdge: Constrains outlines

  • Group 4: Generation

    • KSampler //Inspire: 30 steps, Euler-beta

    • SaveImage: Auto-saves with timestamp

5. Inputs & Outputs

  • Inputs:

    • Required: Pose reference image (768x1024), clothing prompts

    • Optional: Seed (random), ControlNet weights (0.6-0.7)

  • Output:

    • Images saved to %date%/image_*

    • Real-time ControlNet previews

6. Critical Notes

  1. Hardware:

    • ≥10GB VRAM (batch processing for 2048x1024)

    • RTX 30/40 series recommended

  2. Model Specs:

    • Must use fp8_e4m3fn precision

    • FLUX.1-specific ControlNet required

  3. Debug:

    • Pose inaccuracy? Check OpenPose preprocessing

    • Skin artifacts? Adjust ZOEY LoRA (0.3-0.5)

  4. Maintenance:

    • Regularly update Impact-Pack for latest processors


previous
Unlock High-Quality Game Character Art with This Cutting-Edge Workflow
next
Unlock Liquid Magic: Advanced I2V Workflow for Stunning Visual Effects

Recommend