mlsanityData to 7 Oct 2026

Latent Forcing: Reordering the Diffusion Trajectory for Pixel-Space Image Generation

arXiv 2602.11401 ↗First posted 11 Feb 2026cs.CV cs.LGcited by 33

On the map: Diffusion models → Diffusion foundations and sampling

Abstract (arXiv)

Latent diffusion models excel at generating high-quality images but lose the benefits of end-to-end modeling. They discard information during image encoding, require a separately trained decoder, and model an auxiliary distribution to the raw data. In this paper, we propose Latent Forcing, a simple modification to existing architectures that achieves the efficiency of latent diffusion while operating on raw natural images. Our approach orders the denoising trajectory by jointly processing latents and pixels with separately tuned noise schedules. This allows the latents to act as a scratchpad for intermediate computation before high-frequency pixel features are generated. We find that the order of conditioning signals is critical, and we analyze this to explain differences between REPA distillation in the tokenizer and the diffusion model, conditional versus unconditional generation, and how tokenizer reconstruction quality relates to diffusability. Applied to ImageNet, Latent Forcing achieves a new state-of-the-art for diffusion transformer-based pixel generation at our compute scale.

Key facts. Cited by 33 papers in the map (data through 7 Oct 2026). Placed in Diffusion foundations and sampling, one of 14 directions in Diffusion models. Builds on 38 of its 53 references that are in the map. First posted on arXiv 11 Feb 2026.

Builds on

38 of 53 references.

Closest ancestors · 21

Same direction, newest first

  1. 2025Diffuse and Disperse: Image Generation with Representation Regularization10 Jun
  2. 2025REPA Works Until It Doesn't: Early-Stopped, Holistic Alignment Supercharges Diffusion Training22 May
  3. 2025REPA-E: Unlocking VAE for End-to-End Tuning with Latent Diffusion Transformers14 Apr
  4. 2025Flow to the Mode: Mode-Seeking Diffusion Autoencoders for State-of-the-Art Image Tokenization14 Mar
  5. 2025Improving the Diffusability of Autoencoders20 Feb
  6. 2025Reconstruction vs. Generation: Taming Optimization Dilemma in Latent Diffusion Models2 Jan
  7. 2024Simpler Diffusion (SiD2): 1.5 FID on ImageNet512 with pixel-space diffusion25 Oct
  8. 2024Deep Compression Autoencoder for Efficient High-Resolution Diffusion Models14 Oct
  9. 2024Representation Alignment for Generation: Training Diffusion Transformers Is Easier Than You Think9 Oct
  10. 2024Diffusion Forcing: Next-token Prediction Meets Full-Sequence Diffusion1 Jul
  11. 2024Guiding a Diffusion Model with a Bad Version of Itself4 Jun
  12. 2024Applying Guidance in a Limited Interval Improves Sample and Distribution Quality in Diffusion Models11 Apr
9 more
  1. 2024Scaling Rectified Flow Transformers for High-Resolution Image Synthesis5 Mar
  2. 2023Efficient Diffusion Training via Min-SNR Weighting Strategy16 Mar
  3. 2023Simple diffusion: End-to-end diffusion for high resolution images26 Jan
  4. 2022Scalable Diffusion Models with Transformers19 Dec
  5. 2022Elucidating the Design Space of Diffusion-Based Generative Models1 Jun
  6. 2021High-Resolution Image Synthesis with Latent Diffusion Models20 Dec
  7. 2021Cascaded Diffusion Models for High Fidelity Image Generation30 May
  8. 2021Diffusion Models Beat GANs on Image Synthesis11 May
  9. 2020Denoising Diffusion Probabilistic Models19 Jun

Other references · 17

  1. 2015Deep Unsupervised Learning using Nonequilibrium Thermodynamicscited by 7,005
  2. 2021RoFormer: Enhanced Transformer with Rotary Position Embeddingcited by 4,581
  3. 2021Zero-Shot Text-to-Image Generationcited by 4,276
  4. 2022Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flowcited by 3,550
  5. 2020Taming Transformers for High-Resolution Image Synthesiscited by 3,229
  6. 2016Perceptual Losses for Real-Time Style Transfer and Super-Resolutioncited by 2,678
  7. 2018Large Scale GAN Training for High Fidelity Natural Image Synthesiscited by 2,609
  8. 2024Training Large Language Models to Reason in a Continuous Latent Spacecited by 763
  9. 2024Autoregressive Image Generation without Vector Quantizationcited by 644
  10. 2022DiffusionBERT: Improving Generative Masked Language Models with Diffusion Modelscited by 214
  11. 2022Efficient Self-supervised Learning with Contextualized Target Representations for Vision, Speech and Languagecited by 108
  12. 2023Return of Unconditional Generation: A Self-supervised Representation Generation Methodcited by 95
5 more
  1. 2025EQ-VAE: Equivariance Regularized Latent Space for Improved Generative Image Modelingcited by 85
  2. 2025Halton Scheduler For Masked Generative Image Transformercited by 30
  3. 2024Diffusion Posterior Sampling is Computationally Intractablecited by 20
  4. 2025Latent Denoising Makes Good Tokenizerscited by 18
  5. 2025Rethinking Generative Image Pretraining: How Far Are We From Scaling Up Next-Pixel Prediction?cited by 2

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 33.

33 papers.

Sort:

By area:

  1. 2026VGGT-World: Transforming VGGT into an Autoregressive Geometry World Modelcited by 11
  2. 2026What Matters for Diffusion-Friendly Latent Manifold? Prior-Aligned Autoencoders for Latent Diffusioncited by 7
  3. 2026End-to-End Training for Unified Tokenization and Latent Denoisingcited by 6
  4. 2026ActionPlan: Future-Aware Streaming Motion Synthesis via Frame-Level Action Planningcited by 4
  5. 2026Asymmetric Flow Modelscited by 4
  6. 2026Lens: Rethinking Training Efficiency for Foundational Text-to-Image Modelscited by 4
  7. 2026Representation Forcing for Bottleneck-Free Unified Multimodal Modelscited by 4
  8. 2026Modality Forcing for Scalable Spatial Generationcited by 3
  9. 2026An Empirical Study of Training Pixel-Space Text-to-Image Diffusion Modelscited by 3
  10. 2026V-Co: A Closer Look at Visual Representation Alignment via Co-Denoisingcited by 2
  11. 2026Registers Matter for Pixel-Space Diffusion Transformerscited by 2
  12. 2026Show the Signal, Hide the Noise: Spectral Forcing for Pixel-Space Diffusioncited by 2
21 more
  1. 2026Trajectory Forcing: Structure-First Generation with Controllable Semantic Trajectoriescited by 2
  2. 2026A Frozen Pixel-Space Diffusion Model Can Guide Itself with Its Own Samplescited by 2
  3. 2026Modality-Autoregressive World-Action Modelscited by 2
  4. 2026WiT: Waypoint Diffusion Transformers for Alleviating Trajectory Conflict in Pixel-Space Image Generationcited by 1
  5. 2026Know3D: Prompting 3D Generation with Knowledge from Vision-Language Modelscited by 1
  6. 2026ReWorld: Representation Learning for World Action Modelscited by 1
  7. 2026Understanding Latent Diffusability via Fisher Geometrycited by 0
  8. 2026Frequency-Forcing: From Scaling-as-Time to Soft Frequency Guidancecited by 0
  9. 2026Taming Outlier Tokens in Diffusion Transformerscited by 0
  10. 2026Learning When to Denoise: Optimizing Asynchronous Schedules for Latent Diffusioncited by 0
  11. 2026Pixel-Space Diffusion Transformerscited by 0
  12. 2026Where Does Generative Difficulty Reside? An Empirical Study of Target Representationscited by 0
  13. 2026When Latents Forget Pixels: Restoring Fidelity in Diffusion Transformer Super-Resolutioncited by 0
  14. 2026PixelDiT2: Representation-Grounded Pixel Diffusion Transformerscited by 0
  15. 2026Reasoning with Continuous Latent Diffusioncited by 0
  16. 2026Persistence Forcing: Exploiting Feature Specialization in Pixel-Space Diffusioncited by 0
  17. 2026DC-SAE: Deep Compression Semantic Autoencoder for Faster Diffusion Convergencecited by 0
  18. 2026Discrete Forcing: Infusing Discrete Guidance into Continuous Denoising for Few-Step Action Expertscited by 0
  19. 2026PixelDense: Dense Prediction as Representation Alignment for Pixel Diffusioncited by 0
  20. 2026Just Align bmx: Aligning Predictions, Not Representationscited by 0
  21. 2026Position Forcing: Self-Conditioning 3D Generationcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2024Pixel-Space Post-Training of Latent Diffusion Models
  2. 2025DeCo: Frequency-Decoupled Pixel Diffusion for End-to-End Image Generation
  3. 2025PixelDiT: Pixel Diffusion Transformers for Image Generation
  4. 2025OSCAR: One-Step Diffusion Codec Across Multiple Bit-rates
  5. 2026LDM-is-AE: Latent Diffusion Model is an Auto-Encoder for End-to-End Image Generation
  6. 2026FrequencyBooster: Full-Frequency Modeling for High-Fidelity Pixel Diffusion
  7. 2022Unifying Diffusion Models' Latent Space, with Applications to CycleDiffusion and Guidance
  8. 2026One Model, Many Budgets: Elastic Latent Interfaces for Diffusion Transformers