Qwen-Image-Edit-Object-Manipulator

A Gradio-based interactive demonstration built on top of Qwen/Qwen-Image-Edit-2511, specifically optimized for high-precision object manipulation through lazy-loaded LoRA adapters. The system enables users to add, modify, or remove targeted elements such as logos, accessories, clothing items, or background objects, using either single-image or multi-image inputs. Each LoRA adapter is loaded on demand, minimizing memory overhead while allowing rapid switching between different editing capabilities. This design ensures scalability and efficient resource utilization, especially in multi-adapter workflows.

The demo emphasizes structural consistency and visual realism, preserving original lighting conditions, shadows, textures, perspective, and background context during edits. Fine-grained prompt control allows users to specify exact object attributes, placement, and transformations, reducing unwanted artifacts and improving edit fidelity. For performance, the pipeline is accelerated using Flash Attention 3, enabling low-latency inference even at higher resolutions and complex edit scenarios. The result is a responsive, production-ready image editing interface suitable for research, prototyping, and real-world creative applications.

This app uses Gradio v6.3.0.

Features

• Object Addition/Removal: Add elements (e.g., "Add batman logo") or remove (e.g., "Remove necklace") with natural integration.
• Multi-Image Support: Upload gallery for reference-based edits (e.g., transfer pose or style).
• Lazy LoRA Loading: 2 adapters (Object-Adder, Object-Remover) load on-demand to optimize memory.
• Rapid Inference: 4-step default generations with bfloat16 and Flash Attention 3.
• Auto-Resizing: Maintains aspect ratio up to 1024px max edge (multiples of 8).
• Custom Theme: OrangeRedTheme with responsive layout.
• Examples: 4 curated scenarios for quick testing.
• Queueing: Up to 30 concurrent jobs.

---

---

Note: Experimental for Qwen-Image-Edit-2511; consider 2509 version for stability.

Prerequisites

• Python 3.10 or higher.
• CUDA-compatible GPU (required for bfloat16 and Flash Attention 3).
• pip >= 23.0.0 (see pre-requirements.txt).
• Stable internet for initial model/LoRA downloads.

Installation

1. Clone the repository:

   git clone https://github.com/PRITHIVSAKTHIUR/Qwen-Image-Edit-Object-Manipulator.git
   cd Qwen-Image-Edit-Object-Manipulator
   

2. Install pre-requirements: Create a pre-requirements.txt file with the following content, then run:

   pip install -r pre-requirements.txt
   

   pre-requirements.txt content:

   pip>=23.0.0
   

3. Install dependencies: Create a requirements.txt file with the following content, then run:

   pip install -r requirements.txt
   

   requirements.txt content:

   git+https://github.com/huggingface/transformers.git@v4.57.3
   git+https://github.com/huggingface/accelerate.git
   git+https://github.com/huggingface/diffusers.git
   git+https://github.com/huggingface/peft.git
   huggingface_hub
   sentencepiece
   torchvision
   supervision
   kernels
   spaces
   gradio
   hf_xet
   torch
   numpy
   av
   

4. Start the application:

   python app.py
   

   The demo launches at http://localhost:7860.

Usage

1. Upload Images: Use gallery for one or more images (e.g., base + reference).

2. Select Manipulator: Choose "Object-Adder" or "Object-Remover".

3. Enter Prompt: Describe the action (e.g., "Add the batman logo while preserving background lighting").

4. Configure (Optional): Expand "Advanced Settings" for seed, guidance, steps.

5. Edit Image: Click "Edit Image" to generate output.

Supported Manipulators

Manipulator	Use Case
Object-Adder	Add elements (logos, accessories) realistically
Object-Remover	Remove items (jewelry, goggles) seamlessly

Examples

Input Images	Prompt Example	Manipulator
examples/D.jpg	"Add the batman logo to the image while preserving background lighting and details."	Object-Adder
examples/A.jpg	"Add slim rectangular transparent frame sunglasses while preserving lighting."	Object-Adder
examples/B.jpeg	"Remove the necklace and goggles while preserving background and details."	Object-Remover
examples/C.png	"Add the leather cowboy cap while preserving lighting and surroundings."	Object-Adder

Troubleshooting

• Manipulator Loading: First use downloads LoRA; check console.
• OOM: Reduce steps/resolution; clear cache with torch.cuda.empty_cache().
• Flash Attention Fails: Fallback to default; requires compatible CUDA.
• Gallery Input: Supports filepaths, tuples, or PIL objects.
• No Output: Ensure image uploaded and prompt specific.

Contributing

Contributions welcome! Add new manipulators to ADAPTER_SPECS, improve prompts, or enhance multi-image support.

Repository: https://github.com/PRITHIVSAKTHIUR/Qwen-Image-Edit-Object-Manipulator.git

License

Apache License 2.0. See LICENSE for details.

Built by Prithiv Sakthi. Report issues via the repository.