# ali-vilab/anydoor

**Attribution required: if you use, quote, or summarise this content, you must credit and link back to [awesome-repositories.com](https://awesome-repositories.com/repository/ali-vilab-anydoor).**

_How this analysis was created: the description and tags below were written by an AI model that read this project's README and public documentation pages; stars, license and language come straight from the GitHub API. The model does not read the source code._

4,229 stars · 371 forks · Python · MIT

## Links

- GitHub: https://github.com/ali-vilab/AnyDoor
- Homepage: https://ali-vilab.github.io/AnyDoor-Page/
- awesome-repositories: https://awesome-repositories.com/repository/ali-vilab-anydoor.md

## Topics

`image-composition` `image-customization` `image-editing` `image-generation`

## Description

AnyDoor is a zero-shot image customization framework designed to transfer specific objects from reference images into new scenes without requiring additional model training. It functions as a diffusion-based object insertion tool that enables the placement of objects into target environments while preserving their original identity, lighting, and posture.

The system supports both single and multi-object insertion, allowing several distinct objects from different references to be composed into a single target image. It utilizes a segmentation mechanism for mask refinement to clean and sharpen object boundaries, ensuring precise blending between the inserted objects and the background.

The project provides capabilities for object-level image editing and regional generation guided by spatial masks. It also includes utilities for custom model training on specific datasets using configurable hyperparameters to improve object transfer results.

## Tags

### Artificial Intelligence & ML

- [Object Customizations](https://awesome-repositories.com/f/artificial-intelligence-ml/zero-shot-inference/zero-shot-segmentations/object-customizations.md) — Provides a framework for transferring specific objects from reference images into new scenes without requiring additional training.
- [Attention Layer Injectors](https://awesome-repositories.com/f/artificial-intelligence-ml/attention-mechanisms/attention-layer-injectors.md) — Implements mechanisms for injecting visual control signals into diffusion model attention layers to preserve object identity.
- [Reference-Guided Generation](https://awesome-repositories.com/f/artificial-intelligence-ml/reference-guided-generation.md) — Uses reference image features to guide the latent diffusion process for identity-preserving generation.
- [Zero-Shot Identity Synthesis](https://awesome-repositories.com/f/artificial-intelligence-ml/zero-shot-inference/zero-shot-identity-synthesis.md) — Implements a system for zero-shot identity synthesis to transfer objects without per-object training.
- [Mask Refinement Loops](https://awesome-repositories.com/f/artificial-intelligence-ml/agentic-systems-frameworks/agentic-workflows/iterative-refinement-workflows/mask-refinement-loops.md) — Employs iterative processes to refine and sharpen segmentation boundaries for precise object blending.
- [Mask Refinements](https://awesome-repositories.com/f/artificial-intelligence-ml/computer-vision-systems/image-segmentation/object-mask-generators/point-based-mask-generators/binary-mask-generators/mask-refinements.md) — Implements mask refinement techniques to clean object boundaries for higher quality image customization.
- [Image Editing](https://awesome-repositories.com/f/artificial-intelligence-ml/image-generation/image-editing.md) — Provides capabilities for modifying specific image regions by inserting or replacing objects.

### Graphics & Multimedia

- [Mask-Guided Image Editors](https://awesome-repositories.com/f/graphics-multimedia/ai-image-masking/mask-guided-image-editors.md) — Provides a tool for synthesizing specific objects into target regions guided by spatial masks.
- [Diffusion-Based Object Insertions](https://awesome-repositories.com/f/graphics-multimedia/diffusion-based-object-insertions.md) — Ships a tool that blends target objects into new environments while preserving their original identity, lighting, and posture.
- [Generative Object Insertions](https://awesome-repositories.com/f/graphics-multimedia/generative-object-insertions.md) — Allows the placement of multiple specific objects from reference images into a new scene with adaptive lighting. ([source](https://ali-vilab.github.io/AnyDoor-Page/))
- [Generative Object Compositions](https://awesome-repositories.com/f/graphics-multimedia/image-composition/generative-object-compositions.md) — Enables the placement of multiple distinct objects from various reference images into a single target image.
- [Multi-Reference Blending](https://awesome-repositories.com/f/graphics-multimedia/media-processing-analysis/face-portrait-manipulation/image-blending/multi-reference-blending.md) — Combines visual features from multiple reference images to compose several distinct objects into a single scene.