KO | EN

GitLite

Minimalist Open Source Repository Search Engine

gitlite — search

#typescript #ai-agents #cli #open-source #ai #claude-code #codex #developer-tools #react #windows #deepseek-harness #ai-agent

QWEN-AI-Compositing

★ 90 Open GitHub ↗

AI compositing workflow built on Qwen Image Edit 2509 with Fusion LoRA. Import BG and subject, remove background, composite and mask areas (including shadows), then prompt the model to auto-match lighting, color, and shadows for a seamless, realistic result. Credits to the Qwen model and LoRA creators.

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# QWEN-AI-Compositing
**An enhanced AI compositing workflow** built on Qwen Image Edit 2509 with Fusion LoRA.
Unlike the default setup on the QWEN Fusion LoRA page — which often alters color balance, positioning, and fine details — **this workflow preserves the original plate perfectly**, modifying only the masked regions you define.
Simply import your background and subject, remove the background, composite and mask areas (including shadows), then provide a natural-language prompt describing your desired final look.
Credits to the Qwen model and LoRA creators.

# Video Demo - https://youtu.be/np99SCBUNo4

# 🧠 AI Compositing Workflow – Subject Integration in ComfyUI

This repository contains a **ComfyUI workflow** for realistic subject integration using **Qwen Image Edit 2509** with **Fusion LoRA**.  
It’s designed to automate the process of compositing a subject over a background — intelligently matching lighting, color grading, and shadows for a seamless final result.

---

## 🎬 Overview

Traditional compositing requires manual relighting, color correction, and shadow work to unify the subject and background.  
This workflow leverages **diffusion-based AI harmonization** to handle those tasks automatically while keeping full artistic control over layout and intent.

The process involves:
1. Importing and resizing the **background** and **subject**.
2. **Removing** the subject’s background using AI segmentation.
3. **Positioning and scaling** the subject over the background.
4. **Masking** areas for compositing and shadow interaction.
5. Adding a **natural language prompt** describing the final look.
6. Letting **Qwen Image Edit + Fusion LoRA** infer lighting, grading, and shadow adjustments.
7. Comparing and saving the **final photorealistic composite**.

---

## ⚙️ Requirements

- [ComfyUI](https://github.com/comfyanonymous/ComfyUI)
- Models:
  - `qwen_image_edit_2509_fp8_efdm3m.safetensors`
  - `Qwen-Fusion-LoRA`
  - `Qwen-Image-Lightning LoRAs (V1.0 / V1.1)`
  - `qwen_2_5_v1_7b_fp8_scaled.safetensors` (CLIP)
  - `qwen_image_vae.safetensors` (VAE)

Place models in:
    ComfyUI/models/checkpoints/
    ComfyUI/models/loras/
    ComfyUI/models/vae/
    ComfyUI/models/clip/

---

## 🧩 Workflow Breakdown

| Step | Description |
|------|--------------|
| **1.** | Import and resize background image |
| **2.** | Import subject and remove its background |
| **3.** | Composite subject over background |
| **4.** | Mask subject and shadow regions |
| **5.** | Apply prompt and run diffusion harmonization |
| **6.** | Compare AI composite with base comp |
| **7.** | Save final image to disk |

---

## 💡 Example Prompt

A cinematic photo of a spacecraft on a Martian desert beside an astronaut, realistic lighting, soft shadows, 50mm lens.

No need to specify color correction or light direction — the model automatically adapts these elements.

---

## 🧠 How It Works

The workflow combines:
- **Qwen Image Edit 2509** for diffusion-based inpainting and relighting.  
- **Fusion LoRA** for blending subject and background colors naturally.  
- **Lightning LoRAs** for fast, detailed refinement.  
- **CLIP + VAE** for high-fidelity text/image understanding and decoding.

Together, they turn ComfyUI into a **context-aware compositing environment** — fusing manual control with AI intuition.

---

## 📸 Output Example

| Rough Composite | AI Composite |
|------------------|--------------|
| ![Before](https://raw.githubusercontent.com/rik-python/QWEN-AI-Compositing/main/Output_Images/Park4runner/QwenCompositeRaw_00002_.png) | ![After](https://raw.githubusercontent.com/rik-python/QWEN-AI-Compositing/main/Output_Images/Park4runner/Qwen-IE-2509_Inpaint_00009_.png) |
| ![Before](https://raw.githubusercontent.com/rik-python/QWEN-AI-Compositing/main/Output_Images/DummasBeach/QwenCompositeRaw_00001_.png) | ![After](https://raw.githubusercontent.com/rik-python/QWEN-AI-Compositing/main/Output_Images/DummasBeach/QwenComposite_00001_.png) |
| ![Before](https://raw.githubusercontent.com/rik-python/QWEN-AI-Compositing/main/Output_Images/Forest/QwenCompositeRaw_00001_.png) | ![After](https://raw.githubusercontent.com/rik-python/QWEN-AI-Compositing/main/Output_Images/Forest/Qwen-IE-2509_Inpaint_00021_.png) |
| ![Before](https://raw.githubusercontent.com/rik-python/QWEN-AI-Compositing/main/Output_Images/Mars2/QwenCompositeRaw_00001_.png) | ![After](https://raw.githubusercontent.com/rik-python/QWEN-AI-Compositing/main/Output_Images/Mars2/Qwen-IE-2509_Inpaint_00009_.png) |
| ![Before](https://raw.githubusercontent.com/rik-python/QWEN-AI-Compositing/main/Output_Images/Moon/QwenCompositeRaw_00002_.png) | ![After](https://raw.githubusercontent.com/rik-python/QWEN-AI-Compositing/main/Output_Images/Moon/Qwen-IE-2509_Inpaint_00009_.png) |

---

## 🙏 Credits

- **Qwen Image Edit 2509** by the original authors  
- **Fusion LoRA** and **Lightning LoRAs** by their respective creators  
- Workflow design and optimization by [Your Name]

All rights and model ownership belong to their respective authors.

---

## 🧾 License

This project is released under the **MIT License**.  
You are free to modify and adapt it with proper credit to the model and LoRA creators.

---

## 🌐 Connect

If you use or modify this workflow, feel free to share results or improvements — pull requests and suggestions are welcome!

---

> _"Human intent meets AI precision — compositing made effortless."_

QWEN-AI-Compositing

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README