# What matters for Representation Alignment: Global Information or Spatial Structure?

arXiv: https://arxiv.org/abs/2512.10794 · HTML: https://mlsanity.com/paper/2512.10794/

## Key facts

- Cited by 84 papers in the map (data through 14 Oct 2026).
- Builds on 50 of its 68 references that are in the map.
- First posted on arXiv 11 Dec 2025.

## Builds on (most cited first, 30 of 50)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [Very Deep Convolutional Networks for Large-Scale Image Recognition](https://mlsanity.com/paper/1409.1556/) (2014)
- [Diffusion Models Beat GANs on Image Synthesis](https://mlsanity.com/paper/2105.05233/) (2021)
- [DINOv2: Learning Robust Visual Features without Supervision](https://mlsanity.com/paper/2304.07193/) (2023)
- [Scalable Diffusion Models with Transformers](https://mlsanity.com/paper/2212.09748/) (2022)
- [Generative Adversarial Networks](https://mlsanity.com/paper/2203.00667/) (2022)
- [Scaling Rectified Flow Transformers for High-Resolution Image Synthesis](https://mlsanity.com/paper/2403.03206/) (2024)
- [Improved Techniques for Training GANs](https://mlsanity.com/paper/1606.03498/) (2016)
- [SAM 2: Segment Anything in Images and Videos](https://mlsanity.com/paper/2408.00714/) (2024)
- [An Empirical Study of Training Self-Supervised Vision Transformers](https://mlsanity.com/paper/2104.02057/) (2021)
- [Instance Normalization: The Missing Ingredient for Fast Stylization](https://mlsanity.com/paper/1607.08022/) (2016)
- [Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture](https://mlsanity.com/paper/2301.08243/) (2023)
- [Improved Precision and Recall Metric for Assessing Generative Models](https://mlsanity.com/paper/1904.06991/) (2019)
- [Adversarial Diffusion Distillation](https://mlsanity.com/paper/2311.17042/) (2023)
- [SiT: Exploring Flow and Diffusion-based Generative Models with Scalable Interpolant Transformers](https://mlsanity.com/paper/2401.08740/) (2024)
- [Representation Alignment for Generation: Training Diffusion Transformers Is Easier Than You Think](https://mlsanity.com/paper/2410.06940/) (2024)
- [Mean Flows for One-step Generative Modeling](https://mlsanity.com/paper/2505.13447/) (2025)
- [All are Worth Words: A ViT Backbone for Diffusion Models](https://mlsanity.com/paper/2209.12152/) (2022)
- [StyleGAN-XL: Scaling StyleGAN to Large Diverse Datasets](https://mlsanity.com/paper/2202.00273/) (2022)
- [Reconstruction vs. Generation: Taming Optimization Dilemma in Latent Diffusion Models](https://mlsanity.com/paper/2501.01423/) (2025)
- [Perception Encoder: The best visual embeddings are not at the output of the network](https://mlsanity.com/paper/2504.13181/) (2025)
- [Rethinking FID: Towards a Better Evaluation Metric for Image Generation](https://mlsanity.com/paper/2401.09603/) (2023)
- [Generating Images with Sparse Representations](https://mlsanity.com/paper/2103.03841/) (2021)
- [REPA-E: Unlocking VAE for End-to-End Tuning with Latent Diffusion Transformers](https://mlsanity.com/paper/2504.10483/) (2025)
- [The Role of ImageNet Classes in Fr'echet Inception Distance](https://mlsanity.com/paper/2203.06026/) (2022)
- [StyleGAN-T: Unlocking the Power of GANs for Fast Large-Scale Text-to-Image Synthesis](https://mlsanity.com/paper/2301.09515/) (2023)
- [Projected GANs Converge Faster](https://mlsanity.com/paper/2111.01007/) (2021)
- [Exposing flaws of generative model evaluation metrics and their unfair treatment of diffusion models](https://mlsanity.com/paper/2306.04675/) (2023)
- [JanusFlow: Harmonizing Autoregression and Rectified Flow for Unified Multimodal Understanding and Generation](https://mlsanity.com/paper/2411.07975/) (2024)
- [DDT: Decoupled Diffusion Transformer](https://mlsanity.com/paper/2504.05741/) (2025)

## Built on by (most cited first, 30 of 84)

- [Self-Supervised Flow Matching for Scalable Multi-Modal Synthesis](https://mlsanity.com/paper/2603.06507/) (2026)
- [Improved Baselines with Representation Autoencoders](https://mlsanity.com/paper/2605.18324/) (2026)
- [VISReg: Variance-Invariance-Sketching Regularization for JEPA training](https://mlsanity.com/paper/2606.02572/) (2026)
- [Stable Velocity: A Variance Perspective on Flow Matching](https://mlsanity.com/paper/2602.05435/) (2026)
- [Generalization of Diffusion Models Arises with a Balanced Representation Space](https://mlsanity.com/paper/2512.20963/) (2025)
- [Making Foresight Actionable: Repurposing Representation Alignment in World Action Models](https://mlsanity.com/paper/2606.12217/) (2026)
- [Olaf-World: Orienting Latent Actions for Video World Modeling](https://mlsanity.com/paper/2602.10104/) (2026)
- [What Matters for Diffusion-Friendly Latent Manifold? Prior-Aligned Autoencoders for Latent Diffusion](https://mlsanity.com/paper/2605.07915/) (2026)
- [PixelU: A U-Shaped Transformer for Efficient End-to-End Pixel Diffusion](https://mlsanity.com/paper/2606.27760/) (2026)
- [Laminating Representation Autoencoders for Efficient Diffusion](https://mlsanity.com/paper/2602.04873/) (2026)
- [Spectrum Matching: a Unified Perspective for Superior Diffusability in Latent Diffusion](https://mlsanity.com/paper/2603.14645/) (2026)
- [Don't Retrain, Align: Adapting Autoregressive LMs to Diffusion LMs via Representation Alignment](https://mlsanity.com/paper/2605.06885/) (2026)
- [Diffusing in the Right Space: A Systematic Study of Latent Diffusability](https://mlsanity.com/paper/2606.03578/) (2026)
- [DiffusionBench: On Holistic Evaluation of Diffusion Transformers](https://mlsanity.com/paper/2606.24888/) (2026)
- [TORA: Topological Representation Alignment for 3D Shape Assembly](https://mlsanity.com/paper/2604.04050/) (2026)
- [AHPA: Adaptive Hierarchical Prior Alignment for Diffusion Transformers](https://mlsanity.com/paper/2605.03317/) (2026)
- [Beyond Point-Wise Matching: Structural Representation Alignment for Accelerating Diffusion Transformers](https://mlsanity.com/paper/2605.16949/) (2026)
- [Adaptive 1D Video Diffusion Autoencoder](https://mlsanity.com/paper/2602.04220/) (2026)
- [Zero-Shot Cross-City Generalization in End-to-End Autonomous Driving: Self-Supervised versus Supervised Representations](https://mlsanity.com/paper/2603.11417/) (2026)
- [Teacher-Feature Drifting: One-Step Diffusion Distillation with Pretrained Diffusion Representations](https://mlsanity.com/paper/2605.07327/) (2026)
- [Test-Time Conditioning with Representation-Aligned Visual Features](https://mlsanity.com/paper/2602.03753/) (2026)
- [V-Co: A Closer Look at Visual Representation Alignment via Co-Denoising](https://mlsanity.com/paper/2603.16792/) (2026)
- [CrystalREPA: Transferring Physical Priors from Universal MLIPs to Crystal Generative Models](https://mlsanity.com/paper/2605.08960/) (2026)
- [Registers Matter for Pixel-Space Diffusion Transformers](https://mlsanity.com/paper/2605.16147/) (2026)
- [DecQ: Detail-Condensing Queries for Enhanced Reconstruction and Generation in Representation Autoencoders](https://mlsanity.com/paper/2605.22777/) (2026)
- [MaskAlign: Token-Subset Representation Alignment for Efficient Diffusion Training](https://mlsanity.com/paper/2606.08788/) (2026)
- [ReWorld: Representation Learning for World Action Models](https://mlsanity.com/paper/2606.27504/) (2026)
- [Persistence Forcing: Exploiting Feature Specialization in Pixel-Space Diffusion](https://mlsanity.com/paper/2609.36014/) (2026)
- [DiT-JSCC: Rethinking Deep JSCC with Diffusion Transformers and Semantic Representations](https://mlsanity.com/paper/2601.03112/) (2026)
- [Self-transcendence: Is External Feature Guidance Indispensable for Accelerating Diffusion Transformer Training?](https://mlsanity.com/paper/2601.07773/) (2026)

## Abstract (arXiv)

Representation alignment (REPA) guides generative training by distilling representations from a strong, pretrained vision encoder to intermediate diffusion features. We investigate a fundamental question: what aspect of the target representation matters for generation, its textitglobal revisionsemantic information (e.g., measured by ImageNet-1K accuracy) or its spatial structure (i.e. pairwise cosine similarity between patch tokens)? Prevalent wisdom holds that stronger global semantic performance leads to better generation as a target representation. To study this, we first perform a large-scale empirical analysis across 27 different vision encoders and different model scales. The results are surprising; spatial structure, rather than global performance, drives the generation performance of a target representation. To further study this, we introduce two straightforward modifications, which specifically accentuate the transfer of spatial information. We replace the standard MLP projection layer in REPA with a simple convolution layer and introduce a spatial normalization layer for the external representation. Surprisingly, our simple method (implemented in <4 lines of code), termed iREPA, consistently improves convergence speed of REPA, across a diverse set of vision encoders, model sizes, and training variants (such as REPA, REPA-E, Meanflow, JiT etc). %, etc. Our work motivates revisiting the fundamental working mechanism of representational alignment and how it can be leveraged for improved training of generative models. The code and project page are available at https://end2end-diffusion.github.io/irepa