๐ŸŽจ Add-it: Training-Free Object Insertion

Add objects to images using pretrained diffusion models

๐ŸŒ Project Website | ๐Ÿ“„ Paper | ๐Ÿ’ป Code

Status: Model ready!

Generate a base image and add objects to it

1 1.3
0 10
Localization Model
Example Prompts
Source Prompt Target Prompt Subject Token
  • Prompt Design: The Target Prompt should be similar to the Source Prompt, but include a description of the new object to insert
  • Seed Variation: Try different values for Object Seed - some prompts may require a few attempts to get satisfying results
  • Localization Models: The most effective options are attention_points_sam and attention. Use Show Attention to visualize localization performance
  • Object Placement Issues: If the object is not added to the image:
    • Try decreasing Structure Transfer Step
    • Try increasing Extended Scale
  • Flexibility: To allow more flexibility in modifying the source image, leave Blend Steps empty to send an empty list