EliGen: Entity-Level Controlled Image Generation

Introduction

We propose EliGen, a novel approach that leverages fine-grained entity-level information to enable precise and controllable text-to-image generation. EliGen excels in tasks such as entity-level controlled image generation and image inpainting, while its applicability is not limited to these areas. Additionally, it can be seamlessly integrated with existing community models, such as the IP-Adpater and In-Cotext LoRA.

• Paper: EliGen: Entity-Level Controlled Image Generation with Regional Attention
• Github: DiffSynth-Studio
• Model:
  • ModelScope
  • HuggingFace
• Online Demo: ModelScope EliGen Studio
• Training dataset: ModelScope Dataset

Methodology

We introduce a regional attention mechanism within the DiT framework to effectively process the conditions of each entity. This mechanism enables the local prompt associated with each entity to semantically influence specific regions through regional attention. To further enhance the layout control capabilities of EliGen, we meticulously contribute an entity-annotated dataset and fine-tune the model using the LoRA framework.

1. Regional Attention: Regional attention is shown in above figure, which can be easily applied to other text-to-image models. Its core principle involves transforming the positional information of each entity into an attention mask, ensuring that the mechanism only affects the designated regions.

2. Dataset with Entity Annotation: To construct a dedicated entity control dataset, we start by randomly selecting captions from DiffusionDB and generating the corresponding source image using Flux. Next, we employ Qwen2-VL 72B, recognized for its advanced grounding capabilities among MLLMs, to randomly identify entities within the image. These entities are annotated with local prompts and bounding boxes for precise localization, forming the foundation of our dataset for further training.

3. Training: We utilize LoRA (Low-Rank Adaptation) and DeepSpeed to fine-tune regional attention mechanisms using a curated dataset, enabling our EliGen model to achieve effective entity-level control.

Usage

This model was trained using DiffSynth-Studio. We recommend using DiffSynth-Studio for generation.

git clone https://github.com/modelscope/DiffSynth-Studio.git
cd DiffSynth-Studio
pip install -e .

1. Entity-Level Controlled Image Generation EliGen achieves effective entity-level control results. See entity_control.py for usage.
2. Image Inpainting To apply EliGen to image inpainting task, we propose a inpainting fusion pipeline to preserve the non-painting areas while enabling precise, entity-level modifications over inpaining regions. See entity_inpaint.py for usage.
3. Styled Entity Control EliGen can be seamlessly integrated with existing community models. We have provided an example of how to integrate it with the IP-Adpater. See entity_control_ipadapter.py for usage.
4. Entity Transfer We have provided an example of how to integrate EliGen with In-Cotext LoRA, which achieves interesting entity transfer results. See entity_transfer.py for usage.
5. Play with EliGen using UI Download the checkpoint of EliGen from ModelScope to models/lora/entity_control and run the following command to try interactive UI:

   python apps/gradio/entity_level_control.py
   

Examples

Entity-Level Controlled Image Generation

1. The effect of generating images with continuously changing entity positions.

1. The image generation effect of complex Entity combinations, demonstrating the strong generalization of EliGen. See entity_control.py example_1-6 for generation prompts.

1. Demonstration of the robustness of EliGen. The following examples are generated using the same prompt but different seeds. Refer to entity_control.py example_7 for the prompts.

Image Inpainting

Demonstration of the inpainting mode of EliGen, see entity_inpaint.py for generation prompts.

Styled Entity Control

Demonstration of the styled entity control results with EliGen and IP-Adapter, see entity_control_ipadapter.py for generation prompts.

We also provide a demo of the styled entity control results with EliGen and specific styled lora, see ./styled_entity_control.py for details. Here is the visualization of EliGen with Lego dreambooth lora.

Entity Transfer

Demonstration of the entity transfer results with EliGen and In-Context LoRA, see entity_transfer.py for generation prompts.