README.md

---
license: mit
language:
  - en
base_model:
  - black-forest-labs/FLUX.1-dev
  - Qwen/Qwen2-VL-7B-Instruct
library_name: diffusers
tags:
  - flux
  - qwen2vl
  - stable-diffusion
  - text-to-image
  - image-to-image
  - controlnet
pipeline_tag: text-to-image
inference: false
---

# Qwen2vl-Flux

<div align="center">
  <img src="landing-1.png" alt="Qwen2vl-Flux Banner" width="100%">
</div>

Qwen2vl-Flux is a state-of-the-art multimodal image generation model that enhances FLUX with Qwen2VL's vision-language understanding capabilities. This model excels at generating high-quality images based on both text prompts and visual references, offering superior multimodal understanding and control.

## Model Architecture

<div align="center">
  <img src="flux-architecture.svg" alt="Flux Architecture" width="800px">
</div>

The model integrates Qwen2VL's vision-language capabilities into the FLUX framework, enabling more precise and context-aware image generation. Key components include:
- Vision-Language Understanding Module (Qwen2VL)
- Enhanced FLUX backbone
- Multi-mode Generation Pipeline
- Structural Control Integration

## Features

- **Enhanced Vision-Language Understanding**: Leverages Qwen2VL for superior multimodal comprehension
- **Multiple Generation Modes**: Supports variation, img2img, inpainting, and controlnet-guided generation
- **Structural Control**: Integrates depth estimation and line detection for precise structural guidance
- **Flexible Attention Mechanism**: Supports focused generation with spatial attention control
- **High-Resolution Output**: Supports various aspect ratios up to 1536x1024

## Generation Examples

### Image Variation
Create diverse variations while maintaining the essence of the original image:

<div align="center">
  <table>
    <tr>
      <td><img src="variation_1.png" alt="Variation Example 1" width="256px"></td>
      <td><img src="variation_2.png" alt="Variation Example 2" width="256px"></td>
      <td><img src="variation_3.png" alt="Variation Example 3" width="256px"></td>
    </tr>
    <tr>
      <td><img src="variation_4.png" alt="Variation Example 4" width="256px"></td>
      <td><img src="variation_5.png" alt="Variation Example 5" width="256px"></td>
    </tr>
  </table>
</div>

### Image Blending
Seamlessly blend multiple images with intelligent style transfer:

<div align="center">
  <table>
    <tr>
      <td><img src="blend_1.png" alt="Blend Example 1" width="256px"></td>
      <td><img src="blend_2.png" alt="Blend Example 2" width="256px"></td>
      <td><img src="blend_3.png" alt="Blend Example 3" width="256px"></td>
    </tr>
    <tr>
      <td><img src="blend_4.png" alt="Blend Example 4" width="256px"></td>
      <td><img src="blend_5.png" alt="Blend Example 5" width="256px"></td>
      <td><img src="blend_6.png" alt="Blend Example 6" width="256px"></td>
    </tr>
    <tr>
      <td><img src="blend_7.png" alt="Blend Example 7" width="256px"></td>
    </tr>
  </table>
</div>

### Text-Guided Image Blending
Control image generation with textual prompts:

<div align="center">
  <table>
    <tr>
      <td><img src="textblend_1.png" alt="Text Blend Example 1" width="256px"></td>
      <td><img src="textblend_2.png" alt="Text Blend Example 2" width="256px"></td>
      <td><img src="textblend_3.png" alt="Text Blend Example 3" width="256px"></td>
    </tr>
    <tr>
      <td><img src="textblend_4.png" alt="Text Blend Example 4" width="256px"></td>
      <td><img src="textblend_5.png" alt="Text Blend Example 5" width="256px"></td>
      <td><img src="textblend_6.png" alt="Text Blend Example 6" width="256px"></td>
    </tr>
    <tr>
      <td><img src="textblend_7.png" alt="Text Blend Example 7" width="256px"></td>
      <td><img src="textblend_8.png" alt="Text Blend Example 8" width="256px"></td>
      <td><img src="textblend_9.png" alt="Text Blend Example 9" width="256px"></td>
    </tr>
  </table>
</div>

### Grid-Based Style Transfer
Apply fine-grained style control with grid attention:

<div align="center">
  <table>
    <tr>
      <td><img src="griddot_1.png" alt="Grid Example 1" width="256px"></td>
      <td><img src="griddot_2.png" alt="Grid Example 2" width="256px"></td>
      <td><img src="griddot_3.png" alt="Grid Example 3" width="256px"></td>
    </tr>
    <tr>
      <td><img src="griddot_4.png" alt="Grid Example 4" width="256px"></td>
      <td><img src="griddot_5.png" alt="Grid Example 5" width="256px"></td>
      <td><img src="griddot_6.png" alt="Grid Example 6" width="256px"></td>
    </tr>
    <tr>
      <td><img src="griddot_7.png" alt="Grid Example 7" width="256px"></td>
      <td><img src="griddot_8.png" alt="Grid Example 8" width="256px"></td>
      <td><img src="griddot_9.png" alt="Grid Example 9" width="256px"></td>
    </tr>
  </table>
</div>

## Usage

The inference code is available via our [GitHub repository](https://github.com/erwold/qwen2vl-flux) which provides comprehensive Python interfaces and examples.

### Installation

1. Clone the repository and install dependencies:
```bash
git clone https://github.com/erwold/qwen2vl-flux
cd qwen2vl-flux
pip install -r requirements.txt
```

2. Download model checkpoints from Hugging Face:
```python
from huggingface_hub import snapshot_download

snapshot_download("Djrango/Qwen2vl-Flux")
```

### Basic Examples

```python
from model import FluxModel

# Initialize model
model = FluxModel(device="cuda")

# Image Variation
outputs = model.generate(
    input_image_a=input_image,
    prompt="Your text prompt",
    mode="variation"
)

# Image Blending
outputs = model.generate(
    input_image_a=source_image,
    input_image_b=reference_image,
    mode="img2img",
    denoise_strength=0.8
)

# Text-Guided Blending
outputs = model.generate(
    input_image_a=input_image,
    prompt="Transform into an oil painting style",
    mode="variation",
    guidance_scale=7.5
)

# Grid-Based Style Transfer
outputs = model.generate(
    input_image_a=content_image,
    input_image_b=style_image,
    mode="controlnet",
    line_mode=True,
    depth_mode=True
)
```

## Technical Specifications

- **Framework**: PyTorch 2.4.1+
- **Base Models**: 
  - FLUX.1-dev
  - Qwen2-VL-7B-Instruct
- **Memory Requirements**: 48GB+ VRAM
- **Supported Image Sizes**:
  - 1024x1024 (1:1)
  - 1344x768 (16:9)
  - 768x1344 (9:16)
  - 1536x640 (2.4:1)
  - 896x1152 (3:4)
  - 1152x896 (4:3)


## Citation

```bibtex
@misc{erwold-2024-qwen2vl-flux,
      title={Qwen2VL-Flux: Unifying Image and Text Guidance for Controllable Image Generation}, 
      author={Pengqi Lu},
      year={2024},
      url={https://github.com/erwold/qwen2vl-flux}
}
```

## License

This project is licensed under the MIT License. See [LICENSE](LICENSE) for details.

## Acknowledgments

- Based on the FLUX architecture
- Integrates Qwen2VL for vision-language understanding
- Thanks to the open-source communities of FLUX and Qwen