# Exploring Representation-Aligned Latent Space for Better Generation

arXiv: https://arxiv.org/abs/2502.00359 · HTML: https://mlsanity.com/paper/2502.00359/

## Key facts

- Cited by 22 papers in the map (data through 14 Oct 2026).
- Placed in Diffusion foundations and sampling, one of 14 directions in Diffusion models.
- First posted on arXiv 1 Feb 2025.

## Placed in

- Diffusion models / [Diffusion foundations and sampling](https://mlsanity.com/diffusion/diffusion-foundations-and-sampling/)

## Built on by (most cited first, 22 of 22)

- [Aether: Geometric-Aware Unified World Modeling](https://mlsanity.com/paper/2503.18945/) (2025)
- [GigaTok: Scaling Visual Tokenizers to 3 Billion Parameters for Autoregressive Image Generation](https://mlsanity.com/paper/2504.08736/) (2025)
- [Transition Models: Rethinking the Generative Learning Objective](https://mlsanity.com/paper/2509.04394/) (2025)
- [Improved Baselines with Representation Autoencoders](https://mlsanity.com/paper/2605.18324/) (2026)
- [USP: Unified Self-Supervised Pretraining for Image Generation and Understanding](https://mlsanity.com/paper/2503.06132/) (2025)
- [Both Semantics and Reconstruction Matter: Making Representation Encoders Ready for Text-to-Image Generation and Editing](https://mlsanity.com/paper/2512.17909/) (2025)
- [Native-Resolution Image Synthesis](https://mlsanity.com/paper/2506.03131/) (2025)
- [Audio Super-Resolution with Latent Bridge Models](https://mlsanity.com/paper/2509.17609/) (2025)
- [Toward Diffusible High-Dimensional Latent Spaces: A Frequency Perspective](https://mlsanity.com/paper/2511.22249/) (2025)
- [SRA 2: Variational Autoencoder Self-Representation Alignment for Efficient Diffusion Training](https://mlsanity.com/paper/2601.17830/) (2026)
- [Decoupling High and Low Frequencies for Faithful Image Generation with Fine Details](https://mlsanity.com/paper/2509.05441/) (2025)
- [Skywork UniPic 3.0: Unified Multi-Image Composition via Sequence Modeling](https://mlsanity.com/paper/2601.15664/) (2026)
- [EVATok: Adaptive Length Video Tokenization for Efficient Visual Autoregressive Generation](https://mlsanity.com/paper/2603.12267/) (2026)
- [InfGen: A Resolution-Agnostic Paradigm for Scalable Image Synthesis](https://mlsanity.com/paper/2509.10441/) (2025)
- [UniDDT: Unifying Multimodal Understanding and Generation with Decoupled Diffusion Transformer](https://mlsanity.com/paper/2606.16255/) (2026)
- [Repulsor: Accelerating Generative Modeling with a Contrastive Memory Bank](https://mlsanity.com/paper/2512.08648/) (2025)
- [Latent Iterative Refinement Flow: A Geometric Constrained Approach for Few-Shot Generation](https://mlsanity.com/paper/2509.19903/) (2025)
- [Wave-GMS: Lightweight Multi-Scale Generative Model for Medical Image Segmentation](https://mlsanity.com/paper/2510.03216/) (2025)
- [Align & Invert: Solving Inverse Problems with Diffusion and Flow-based Models via Representation Alignment](https://mlsanity.com/paper/2511.16870/) (2025)
- [Distilling Latent Manifolds: Resolution Extrapolation by Variational Autoencoders](https://mlsanity.com/paper/2603.14536/) (2026)
- [Rethinking Generative Image Compression at Extremely Low Bitrates](https://mlsanity.com/paper/2609.39315/) (2026)
- [CoVA: Composing VAE Representations for Generative Learning](https://mlsanity.com/paper/2610.13322/) (2026)

## Abstract (arXiv)

Generative models serve as powerful tools for modeling the real world, with mainstream diffusion models, particularly those based on the latent diffusion model paradigm, achieving remarkable progress across various tasks, such as image and video synthesis. Latent diffusion models are typically trained using Variational Autoencoders (VAEs), interacting with VAE latents rather than the real samples. While this generative paradigm speeds up training and inference, the quality of the generated outputs is limited by the latents' quality. Traditional VAE latents are often seen as spatial compression in pixel space and lack explicit semantic representations, which are essential for modeling the real world. In this paper, we introduce ReaLS (Representation-Aligned Latent Space), which integrates semantic priors to improve generation performance. Extensive experiments show that fundamental DiT and SiT trained on ReaLS can achieve a 15% improvement in FID metric. Furthermore, the enhanced semantic latent space enables more perceptual downstream tasks, such as segmentation and depth estimation.