# IP-Composer: Semantic Composition of Visual Concepts

arXiv: https://arxiv.org/abs/2502.13951 · HTML: https://mlsanity.com/paper/2502.13951/

## Key facts

- Cited by 14 papers in the map (data through 14 Oct 2026).
- Builds on 7 of its 7 references that are in the map.
- First posted on arXiv 19 Feb 2025.

## Builds on (most cited first, 7 of 7)

- [T2I-Adapter: Learning Adapters to Dig out More Controllable Ability for Text-to-Image Diffusion Models](https://mlsanity.com/paper/2302.08453/) (2023)
- [Encoder-based Domain Tuning for Fast Personalization of Text-to-Image Models](https://mlsanity.com/paper/2302.12228/) (2023)
- [Sketch-Guided Text-to-Image Diffusion Models](https://mlsanity.com/paper/2211.13752/) (2022)
- [Domain-Agnostic Tuning-Encoder for Fast Personalization of Text-To-Image Models](https://mlsanity.com/paper/2307.06925/) (2023)
- [CLIP2StyleGAN: Unsupervised Extraction of StyleGAN Edit Directions](https://mlsanity.com/paper/2112.05219/) (2021)
- [Concept Decomposition for Visual Exploration and Inspiration](https://mlsanity.com/paper/2305.18203/) (2023)
- [TexSliders: Diffusion-Based Texture Editing in CLIP Space](https://mlsanity.com/paper/2405.00672/) (2024)

## Built on by (most cited first, 14 of 14)

- [Stand-In: A Lightweight and Plug-and-Play Identity Control for Video Generation](https://mlsanity.com/paper/2508.07901/) (2025)
- [SAEdit: Token-level control for continuous image editing via Sparse AutoEncoder](https://mlsanity.com/paper/2510.05081/) (2025)
- [Piece it Together: Part-Based Concepting with IP-Priors](https://mlsanity.com/paper/2503.10365/) (2025)
- [Distribution-Conditional Generation: From Class Distribution to Creative Generation](https://mlsanity.com/paper/2505.03667/) (2025)
- [VLM-Guided Adaptive Negative Prompting for Creative Generation](https://mlsanity.com/paper/2510.10715/) (2025)
- [Learning Joint ID-Textual Representation for ID-Preserving Image Synthesis](https://mlsanity.com/paper/2504.14202/) (2025)
- [Composing Concepts from Images and Videos via Concept-prompt Binding](https://mlsanity.com/paper/2512.09824/) (2025)
- [Inspiration Seeds: Learning Non-Literal Visual Combinations for Generative Exploration](https://mlsanity.com/paper/2602.08615/) (2026)
- [CLAY: Conditional Visual Similarity Modulation in Vision-Language Embedding Space](https://mlsanity.com/paper/2604.11539/) (2026)
- [Muses: Designing, Composing, Generating Nonexistent Fantasy 3D Creatures without Training](https://mlsanity.com/paper/2601.03256/) (2026)
- [Graph-PiT: Enhancing Structural Coherence in Part-Based Image Synthesis via Graph Priors](https://mlsanity.com/paper/2604.06074/) (2026)
- [CMAG: Concept-Scaffolded Retrieval for Marketplace Avatar Generation](https://mlsanity.com/paper/2605.18680/) (2026)
- [Latent-Identity Tuning in Text-to-Image Personalization Models](https://mlsanity.com/paper/2607.11885/) (2026)
- [Mise-en-Sc`ene: Implicit Layout Emergence in Diffusion Transformers for Human-AI Design Co-Creation](https://mlsanity.com/paper/2608.19000/) (2026)

## Abstract (arXiv)

Content creators often draw inspiration from multiple visual sources, combining distinct elements to craft new compositions. Modern computational approaches now aim to emulate this fundamental creative process. Although recent diffusion models excel at text-guided compositional synthesis, text as a medium often lacks precise control over visual details. Image-based composition approaches can capture more nuanced features, but existing methods are typically limited in the range of concepts they can capture, and require expensive training procedures or specialized data. We present IP-Composer, a novel training-free approach for compositional image generation that leverages multiple image references simultaneously, while using natural language to describe the concept to be extracted from each image. Our method builds on IP-Adapter, which synthesizes novel images conditioned on an input image's CLIP embedding. We extend this approach to multiple visual inputs by crafting composite embeddings, stitched from the projections of multiple input images onto concept-specific CLIP-subspaces identified through text. Through comprehensive evaluation, we show that our approach enables more precise control over a larger range of visual concept compositions.