As a powerful node-based AI image generation interface, ComfyUI provides us with unlimited creative possibilities. Today, we will focus on a recently popular technology—In Context Edit (ICEdit), which allows you to intelligently edit your images through simple text instructions, just like communicating with a designer, without the need for cumbersome masking operations. This ComfyUI ICEdit tutorial will take you step by step to understand its workflow and core concepts.

What is In Context Edit (ICEdit)?
Simply put, In Context Edit (ICEdit) is an AI image editing method that can understand the original image you provide, the description of the original image (source context), and the editing instructions you wish to perform (target context). Based on this information, ICEdit can accurately modify specific parts of the image while maximizing the retention of details and overall style in the unedited areas. Its greatest charm lies in "context awareness" and "no masking required," making AI image editing unprecedentedly intuitive and efficient.

Technical Architecture
ICEdit adopts a context editing paradigm implemented based on DiT (FLUX) without training. The model generates edited outputs (right panel of the diptych) by processing "context prompts" and the source image (left panel). Although there are still some ongoing failure cases, the achieved advantages establish a strong baseline that helps improve accuracy through efficient fine-tuning. Within the DiT framework, a parameter-efficient LoRA adapter and mixed expert (MoE) routing are implemented, which dynamically activate task-specific experts during the editing process. The model is trained on minimal public data (50,000) and can improve editing success rates in various scenarios without architectural modifications or large-scale retraining. Additionally, ICEdit has designed an inference-time extension strategy to enhance editing quality. For more detailed information, click to visit the project page.

Examples:
Style conversion: convert the image into anime illustration
Change clothing color: change the white shirt to xxxx color
Change hairstyle: change the girl hair to xxxxxx hair
Slightly modify pose: make the girl eyes closed, make the girl smile
Add objects: the girl wears/ a cat
Remove objects/text/watermark: remove picture the xxx/text/watermark