Laminating Representation Autoencoders for Efficient Diffusion

Recent work has shown that diffusion models can generate high-quality images by operating directly on SSL patch features rather than pixel-space latents. However, the dense patch grids from encoders like DINOv2 contain significant redundancy, making diffusion needlessly expensive. We introduce FlatDINO, a variational autoencoder that compresses this representation into a one-dimensional sequence of just 32 continuous tokens -an 8x reduction in sequence length and 48x compression in total dimensionality. On ImageNet 256x256, a DiT-XL trained on FlatDINO latents achieves a gFID of 1.80 with classifier-free guidance while requiring 8x fewer FLOPs per forward pass and up to 4.5x fewer FLOPs per training step compared to diffusion on uncompressed DINOv2 features. These are preliminary results and this work is in progress.

• Publication year

  2026

• Venue

  Unknown venue

• Publication date

  2026-02-04

• Fields of study

  Computer Science

• Identifiers
  arXiv 2602.04873
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• Democratizing Text-to-Image Masked Generative Models with Compact Text-Aware One-Dimensional Tokens

  2025cited by this paper
• Reconstruction vs. Generation: Taming Optimization Dilemma in Latent Diffusion Models

  2025influential reference
• PixelDiT: Pixel Diffusion Transformers for Image Generation

  2025cited by this paper
• Representation Entanglement for Generation:Training Diffusion Transformers Is Much Easier Than You Think

  2025cited by this paper
• What matters for Representation Alignment: Global Information or Spatial Structure?

  2025influential reference
• REPA-E: Unlocking VAE for End-to-End Tuning with Latent Diffusion Transformers

  2025cited by this paper
• PixelFlow: Pixel-Space Generative Models with Flow

  2025cited by this paper
• Beyond Next-Token: Next-X Prediction for Autoregressive Visual Generation

  2025cited by this paper
• FlexTok: Resampling Images into 1D Token Sequences of Flexible Length

  2025cited by this paper
• Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction

  2024cited by this paper
• A Review on Discriminative Self-supervised Learning Methods in Computer Vision

  2024cited by this paper
• Simpler Diffusion (SiD2): 1.5 FID on ImageNet512 with pixel-space diffusion

  2024cited by this paper
• Representation Alignment for Generation: Training Diffusion Transformers Is Easier Than You Think

  2024influential reference
• Autoregressive Image Generation without Vector Quantization

  2024cited by this paper
• An Image is Worth 32 Tokens for Reconstruction and Generation

  2024cited by this paper
• Applying Guidance in a Limited Interval Improves Sample and Distribution Quality in Diffusion Models

  2024cited by this paper
• MiniCPM: Unveiling the Potential of Small Language Models with Scalable Training Strategies

  2024cited by this paper
• Scaling Rectified Flow Transformers for High-Resolution Image Synthesis

  2024cited by this paper
• DINOv2: Learning Robust Visual Features without Supervision

  2023cited by this paper
• Vision Transformers Need Registers

  2023influential reference
• SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis

  2023cited by this paper
• Fast Training of Diffusion Models with Masked Transformers

  2023cited by this paper
• Diffusion policy: Visuomotor policy learning via action diffusion

  2023cited by this paper
• Splicing ViT Features for Semantic Appearance Transfer

  2022cited by this paper
• Scalable Diffusion Models with Transformers

  2022cited by this paper
• High-Resolution Image Synthesis with Latent Diffusion Models

  2021influential reference
• Diffusion Models Beat GANs on Image Synthesis

  2021cited by this paper
• Emerging Properties in Self-Supervised Vision Transformers

  2021cited by this paper
• An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

  2020cited by this paper
• Denoising Diffusion Implicit Models

  2020cited by this paper
• AUTO-ENCODING VARIATIONAL BAYES

  2020cited by this paper
• GLU Variants Improve Transformer

  2020influential reference
• A Simple Framework for Contrastive Learning of Visual Representations

  2020cited by this paper
• Taming Transformers for High-Resolution Image Synthesis

  2020cited by this paper
• Score-Based Generative Modeling through Stochastic Differential Equations

  2020cited by this paper
• Generative Modeling by Estimating Gradients of the Data Distribution

  2019cited by this paper
• Momentum Contrast for Unsupervised Visual Representation Learning

  2019cited by this paper
• The Unreasonable Effectiveness of Deep Features as a Perceptual Metric

  2018cited by this paper
• Neural Discrete Representation Learning

  2017cited by this paper
• Deep Image Prior

  2017cited by this paper
• Decoupled Weight Decay Regularization

  2017cited by this paper
• beta-VAE: Learning Basic Visual Concepts with a Constrained Variational Framework

  2016cited by this paper
• Deep Unsupervised Learning using Nonequilibrium Thermodynamics

  2015cited by this paper
• Food-101 - Mining Discriminative Components with Random Forests

  2014cited by this paper
• Describing Textures in the Wild

  2013cited by this paper
• Learning Multiple Layers of Features from Tiny Images

  2009cited by this paper
• ImageNet: A large-scale hierarchical image database

  2009cited by this paper
• Automated Flower Classification over a Large Number of Classes

  2008cited by this paper
• Learning Generative Visual Models from Few Training Examples: An Incremental Bayesian Approach Tested on 101 Object Categories

  2004cited by this paper
• cats and dogs

  2003cited by this paper

• No citing papers are available for this paper.