IP Composer ๐ŸŒ…โœš๐Ÿ–Œ๏ธ

compose new images with visual concepts extracted from refrence images using CLIP & IP Adapter

๐Ÿ› ๏ธ How to Use:

  1. Upload a base image
  2. Upload 1โ€“3 concept images
  3. Select a concept type to extract from each concept image:
    • Choose a predefined concept type from the dropdown (e.g. pattern, emotion, pose), or
    • Upload a file with text variations of your concept (e.g. prompts from an LLM).
      • ๐Ÿ‘‰ If you're uploading a new concept, don't forget to adjust the "rank" value under Advanced Options for better results.

Following the algorithm proposed in IP-Composer: Semantic Composition of Visual Concepts by Dorfman et al. [Project page] [arxiv]

Concept 1

Pick concept type

  1. Upload a file with text variations of your concept (e.g. ask an LLM)
  1. Prefereably with > 100 variations.

Emotion Description

a photo of a person feeling joyful

a photo of a person feeling sorrowful

a photo of a person feeling enraged

a photo of a person feeling astonished

a photo of a person feeling disgusted

a photo of a person feeling terrified

...

1 50
0.1 2

If a concept is not showing enough, try to increase the rank

1 150
1 150
1 150
Examples
Base Image (Required) Concept Image 1 Concept 1 Concept Image 2 Concept 2 Rank concept 1 Rank concept 2 Rank concept 3 Scale Seed Num steps