mlsanityData to 7 Oct 2026

JEDI: The Force of Jensen-Shannon Divergence in Disentangling Diffusion Models

arXiv 2505.19166 ↗First posted 25 May 2025cs.CV cs.LGcited by 1

On the map: Diffusion models → Diffusion foundations and sampling

Abstract (arXiv)

We introduce JEDI, a test-time adaptation method that enhances subject separation and compositional alignment in diffusion models without requiring retraining or external supervision. JEDI operates by minimizing semantic entanglement in attention maps using a novel Jensen-Shannon divergence based objective. To improve efficiency, we leverage adversarial optimization, reducing the number of updating steps required. JEDI is model-agnostic and applicable to architectures such as Stable Diffusion 1.5 and 3.5, consistently improving prompt alignment and disentanglement in complex scenes. Additionally, JEDI provides a lightweight, CLIP-free disentanglement score derived from internal attention distributions, offering a principled benchmark for compositional alignment under test-time conditions. Code and results are available at https://ericbill21.github.io/JEDI/.

Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Placed in Diffusion foundations and sampling, one of 14 directions in Diffusion models. Builds on 18 of its 21 references that are in the map. First posted on arXiv 25 May 2025.

Builds on

18 of 21 references.

Closest ancestors · 5

Same direction, newest first

  1. 2024Scaling Rectified Flow Transformers for High-Resolution Image Synthesis5 Mar
  2. 2023SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis4 Jul
  3. 2022Scalable Diffusion Models with Transformers19 Dec
  4. 2021High-Resolution Image Synthesis with Latent Diffusion Models20 Dec
  5. 2020Denoising Diffusion Probabilistic Models19 Jun

Other references · 13

  1. 2020A Simple Framework for Contrastive Learning of Visual Representationscited by 11,979
  2. 2018Representation Learning with Contrastive Predictive Codingcited by 7,662
  3. 2014Explaining and Harnessing Adversarial Examplescited by 7,287
  4. 2022Hierarchical Text-Conditional Image Generation with CLIP Latentscited by 6,539
  5. 2022Flow Matching for Generative Modelingcited by 5,426
  6. 2023Attend-and-Excite: Attention-Based Semantic Guidance for Text-to-Image Diffusion Modelscited by 696
  7. 2022Compositional Visual Generation with Composable Diffusion Modelscited by 619
  8. 2023Mix-of-Show: Decentralized Low-Rank Adaptation for Multi-Concept Customization of Diffusion Modelscited by 261
  9. 2023CONFORM: Contrast is All You Need For High-Fidelity Text-to-Image Diffusion Modelscited by 54
  10. 2024Make It Count: Text-to-Image Generation with an Accurate Number of Objectscited by 47
  11. 2024Concept Weaver: Enabling Multi-Concept Fusion in Text-to-Image Modelscited by 23
  12. 2024LoRACLR: Contrastive Adaptation for Customization of Diffusion Modelscited by 20
1 more
  1. 2024Enhancing MMDiT-Based Text-to-Image Models for Similar Subject Generationcited by 11

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Training-Free Multi-Concept LoRA Composition with Prompt-Aware Weightingcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Boosting Latent Diffusion Models via Disentangled Representation Alignment
  2. 2025SARA: Structural and Adversarial Representation Alignment for Training-efficient Diffusion Models
  3. 2025AlignTok: Aligning Visual Foundation Encoders to Tokenizers for Diffusion Models
  4. 2026HyperAlign: Hypernetwork for Efficient Test-Time Alignment of Diffusion Models
  5. 2024Diffusion Model with Cross Attention as an Inductive Bias for Disentanglement
  6. 2026Guiding a Diffusion Model by Swapping Its Tokens
  7. 2025REGLUE Your Latents with Global and Local Semantics for Entangled Diffusion
  8. 2025Inference-Time Alignment of Diffusion Models via Evolutionary Algorithms