# Imagic: Text-Based Real Image Editing with Diffusion Models

**Authors:** Bahjat Kawar, Shiran Zada, Oran Lang, Omer Tov, Huiwen Chang, Tali Dekel, Inbar Mosseri, Michal Irani  
**arXiv:** [2210.09276](https://arxiv.org/abs/2210.09276)  
**Date:** 2022-10-17  
**Summary page:** https://www.summarizepaper.com/en/paper/2210.09276v1/

## Key points

- The field of text-conditioned image editing has gained significant attention recently
- Existing methods are limited to specific types of edits or only work with synthetically generated images
- Most methods require multiple input images of the same object
- "Imagic" is a groundbreaking approach that allows for complex semantic edits on a single real image using text guidance
- Imagic only requires a single input image and a target text describing the desired edit
- It operates on high-resolution natural images provided by the user without additional inputs such as masks or multiple views of the object
- The key component of Imagic is a pre-trained text-to-image diffusion model that aligns with both the input image and target text
- Imagic enables various sophisticated edits on real images while preserving their original characteristics, such as changing posture and composition, making objects sit down or jump, spreading bird wings, etc.
- This method demonstrates exceptional quality and versatility across numerous inputs from different domains
- It showcases an extensive range of high-quality complex semantic image edits within a single unified framework
- This breakthrough opens up new possibilities for real-world applications in photo editing, digital art creation, and visual storytelling.

## Summary

The field of text-conditioned image editing has gained significant attention recently. However, most existing methods are limited to specific types of edits or only work with synthetically generated images. They also often require multiple input images of the same object. In this paper, a groundbreaking approach called "Imagic" is introduced, which allows for complex semantic edits on a single real image using text guidance. Unlike previous methods, Imagic only requires a single input image and a target text describing the desired edit. It operates on high-resolution natural images provided by the user and does not rely on additional inputs such as image masks or multiple views of the object. The key component of Imagic is a pre-trained text-to-image diffusion model that produces a text embedding aligning with both the input image and the target text. The diffusion model is fine-tuned to capture the specific appearance of the image. With Imagic, it becomes possible to perform various sophisticated edits on real images while preserving their original characteristics. For example, users can change the posture and composition of objects within an image or make a standing dog sit down or jump. They can even cause a bird to spread its wings, among other possibilities. This method demonstrates exceptional quality and versatility across numerous inputs from different domains. It showcases an extensive range of high-quality complex semantic image edits within a single unified framework. This breakthrough opens up new possibilities for real-world applications in areas such as photo editing, digital art creation, and visual storytelling.