1. Workflow Overview

This workflow, named "Star Migration", intelligently maps products (e.g., clothing, accessories) onto models while preserving poses and backgrounds. It combines:

• Stable Diffusion: Realistic generation (majicMIX realistic_v7).

• Florence-2: Visual understanding of product images.

• SAM2: Precise segmentation of products/models.

• ControlNet & FluxGuidance: Controls structural details.

Key Features:

• Auto-align products to models (wearing/holding).

• Background removal via BiRefNetUltra/SAM2.

• Optional upscaling (UltimateSDUpscale).

2. Key Components

Nodes:

1. Input & Preprocessing:

   • LoadImage: Load product/model images.

   • ImageResize+: Standardize resolution (e.g., 1600x1600).

2. Visual Understanding:

   • Florence2Run: Analyzes product images.

   • Sam2Segmentation: Segments product regions.

3. Generation & Inpainting:

   • KSampler: Uses Euler (Steps=20, CFG=8).

   • InpaintStitch: Blends products onto models.

4. Output:

   • UltimateSDUpscale: 2x upscaling (optional).

   • SaveImage: Saves final results.

Dependencies:

• Download Florence-2-large and SAM2 models manually.

• Install plugins via ComfyUI Manager:

  • ComfyUI-Florence2, ComfyUI-Inpaint-CropAndStitch.

3. Workflow Structure

Groups:

1. Product-to-Model Mapping:

   • Input: Product image, model photo, prompts (e.g., "best,4K").

2. Segmentation:

   • Uses SAM2/BiRefNetUltra for masking.

3. Upscaling (Optional):

   • Enabled via UltimateSDUpscale.

4. Inputs & Outputs

Inputs:

• Required:

  • Product image (transparent background or masked).

  • Model photo (with/without background).

• Optional: Seed, resolution, sampler settings.

Output:

• Final image (PNG) with product naturally fitted.

5. Notes

1. Issues:

   • GPU OOM: Reduce resolution or disable upscaling.

   • Model errors: Verify Florence-2/SAM2 paths.

2. Optimization:

   • Use FP16 for Florence-2/SAM2 if GPU supports.

   • Increase KSampler steps (30+) for complex products.