ComfyUI.org

Mastering Multi-Image Composition and Face Swapping for Creative Portraits

CN
ComfyUI.org
2025-06-16 09:05:58

1. Workflow Overview

mbyvcmc9kl7yq2w8asd写真_00002_压缩.png

This workflow is designed for consistent character photography across environments, featuring:

  • Multi-image composition: Layer merging via Compositor3

  • Prompt inversion: Auto-generates prompts from reference images using CLIPtion

  • FLUX PULID face swap: Maintains facial identity in new scenes

  • 4x upscaling: Outputs HD images via UltimateSDUpscale

Ideal for creative portraits and cross-context character placement.

2. Core Models

Model Name

Function

Source/Installation

flux1-dev-fp8-e4m3fn

FLUX main model (FP8 optimized)

Manual download to checkpoints

FLUX PULID v0.9.1

Face swapping & identity preservation

Download .safetensors file

4xLexicaDAT2_otf

Super-resolution upscaler

Install via ComfyUI-Manager

RMBG-1.4

Automatic background removal

Requires ComfyUI-RemBG

3. Key Nodes

Node Name

Function

Installation

Compositor3

Multi-image layering & warping

Install ComfyUI-CompositorPack

CLIPtion Inversion

Generates prompts from images

Manual custom node install

PerturbedAttentionGuidance

Enhances details via noise perturbation

Built-in

easy imageRemBg

One-click background removal

Install ComfyUI-EasyImage

4. Workflow Groups

  • Group 1: Image Composition

    • Input: Multiple source images (e.g., image (38).png)

    • Process: Remove BG → Composite layout → Output merged image

  • Group 2: FLUX PULID Face Swap

    • Input: Character reference + Target scene

    • Process: Face feature extraction → Model fine-tuning → Fusion

  • Group 3: Control & Generation

    • Depth ControlNet for composition

    • KSampler generation → UltimateSD upscale

5. Inputs & Outputs

  • Inputs:

    • Minimum: 1 character image + 1 scene image (PNG/JPG)

    • Optional prompts (auto-inverted if empty)

    • Lora triggers: e.g., bailing_fire (fire spirit style)

  • Output:

    • Resolution: 1024x768 → 2048x1536 (4x)

    • Format: PNG with metadata

6. Notes

⚠️ VRAM: Recommended 16GB+ (PULID & UltimateSD are resource-intensive)
⚠️ Dependencies: Must install:

  • ComfyUI-Impact-Pack (ControlNet preprocessors)

  • ComfyUI-RemBG (background removal)
    ⚠️ Model Paths:

  • PULID models in models/pulid

  • Download clip-vit-large-patch14 for inversion

previous
Precision Image Creation: Harnessing the Power of Dual ControlNets
next
Discover the Beauty of Eastern Romanticism with this AI Art Workflow

Recommend