Full search page →

Latent Diffusion: compress perception before learning the image distribution

Optional · paper guide 1 of 3 for DLCV 12 · Image Generation: From GANs to Text-to-Image Systems. Citation numbers refer to the reference list at the bottom of this page.

Navigate: this page and this track: Navigate: this page and this track

On this page

Supplementary notes · 4

  1. Unified multimodal models, one by one
  2. ControlNet’s zero convolutions: why training starts from the original model
  3. Personalization objectives side by side
  4. Hard bindings, soft bindings and watermarks

Paper guides · 3

  1. Latent Diffusion: compress perception before learning the image distribution
  2. Esser et al., “Scaling Rectified Flow Transformers for High-Resolution Image Synthesis” (ICML 2024)
  3. Ghosh et al., “GenEval: An Object-Focused Framework for Evaluating Text-to-Image Alignment” (NeurIPS 2023)

Deep Learning for Computer Vision

  1. 02 · Deep Image Classification
View the whole course tree →

Challenge

Pixel-space diffusion repeatedly processes large spatial grids, even when much of their fine-scale information can be represented more compactly. Rombach et al.’s CVPR 2022 Latent Diffusion Models paper [5] separates learning a useful image representation from learning a distribution over that representation. The difficult choice is how much compression saves work without discarding the details the generator should produce.

Method

The learned image code is denoised in latent space; concatenation or cross-attention supplies conditioning before decoding.
Original paper figure 3 — The learned image code is denoised in latent space; concatenation or cross-attention supplies conditioning before decoding.

Figure source: [5], PDF page 4. Original PDF.

Train the image representation first. An encoder EE maps RGB image x∈RH×W×3x\in\mathbb R^{H\times W\times3} into z=E(x)∈Rh×w×cz=E(x)\in\mathbb R^{h\times w\times c}; decoder DD reconstructs x~=D(z)\widetilde x=D(z). The spatial compression factor is f=H/h=W/wf=H/h=W/w, while cc is latent channel count. Perceptual reconstruction and patch-adversarial training encourage an image-like reconstruction rather than only low pixel error.

The paper explores weak KL regularization and vector-quantized regularization. In the latter, the quantizer is absorbed into the decoder: the subsequent diffusion model still operates on a spatial latent representation. This is not the same as an autoregressive generator selecting discrete codebook indices.

Freeze the representation, then train the prior. A noisy latent and the conditional denoising objective can be written as

zt=αˉt z+1−αˉt ϵ,L=Ez,ϵ,t,y∥ϵ−ϵθ(zt,t,τθ(y))∥22.z_t=\sqrt{\bar\alpha_t}\,z+\sqrt{1-\bar\alpha_t}\,\epsilon,\qquad \mathcal L=\mathbb E_{z,\epsilon,t,y} \|\epsilon-\epsilon_\theta(z_t,t,\tau_\theta(y))\|_2^2.

tt is a diffusion timestep; αˉt\bar\alpha_t is the retained signal-power fraction; ϵ\epsilon is standard Gaussian noise; ϵθ\epsilon_\theta is the denoising U-Net; yy is a condition; and τθ\tau_\theta encodes it. The expectation averages training examples, noise and timesteps. The autoencoder is held fixed during this prior-training stage.

Choose conditioning to match the information. Spatially aligned information can be concatenated with the noisy latent. Cross-attention allows a feature grid to read a nonaligned sequence such as text: image features supply queries, condition features supply keys and values, and their attention-weighted combination is inserted into the U-Net. It does not impose a one-to-one correspondence between words and patches.

At inference, denoise a latent and decode once. The encoder/decoder and text encoder are part of the complete system cost, even though they are not run at every denoising step. Original text-to-image experiments use a learned language-conditioning component; the paper should not be described as specifying every later Stable Diffusion checkpoint or text encoder.

Read the figure and choose a baseline. Trace an image through the autoencoder before following diffusion in the smaller latent grid. This gives a useful diagnostic order: first check whether encoding and decoding preserve the detail of interest, then assess whether the latent prior can generate it. A failure already present in reconstruction cannot be diagnosed solely as a denoising-sampler problem. During generation, text enters as conditioning and the decoder returns an image after latent sampling; there is no original target image at inference. For restoration, add an explicit account of how observed measurements constrain that generative pathway.

Results

Original Table 1 reports FFHQ generation with LDM-4 and 200 DDIM steps at FID 4.98, precision 0.73 and recall 0.50. Table 2 reports text-conditioned MS-COCO FID 23.35 without classifier-free guidance and 12.61 with guidance at scale 1.5, using 250 DDIM steps. The changed condition strength is part of the experimental setting, not an interchangeable evaluation detail.

The paper also studies layout conditioning, super-resolution and inpainting. Its image-synthesis benchmark is evidence for learned image distributions, not proof that a generated detail matches an unobserved real scene.

Ablations

Section 4.1 varies compression factors 1, 2, 4, 8, 16 and 32, holding diffusion-model parameter count, training steps and the single-A100 compute setting comparable. Too little compression spends training effort on expensive pixel detail; too much first-stage compression loses information before diffusion begins. Figures 5–6 favor moderate compression, particularly factors 4 and 8 for the reported quality/throughput trade-off.

The first-stage reconstruction comparison and downstream generation comparison answer different questions. A better autoencoder reconstruction can help, but it does not guarantee that a finite-capacity prior can model the resulting latent distribution equally well. Likewise, reducing step count changes the sampler experiment rather than the autoencoder alone.

Conclusion

LDM establishes a two-stage design: learn a perceptually useful spatial code, then learn a conditional distribution in that code. Compression, conditioning, prior capacity and sampling each have a separate role.

Problems

The decoder cannot reliably recover exact evidence discarded by compression. Perceptual/adversarial reconstruction can prefer plausible appearance over pixel fidelity. Text conditioning does not guarantee spelling, counting or source attribution, and the training distribution limits which concepts the prior represents. For restoration, ordinary latent diffusion supplies a prior; measurement consistency needs an additional mechanism.

References

  1. T. Karras, S. Laine and T. Aila, “A Style-Based Generator Architecture for Generative Adversarial Networks,” CVPR, 2019. arXiv
  2. A. Brock, J. Donahue and K. Simonyan, “Large Scale GAN Training for High Fidelity Natural Image Synthesis,” ICLR, 2019. arXiv
  3. W. Peebles and S. Xie, “Scalable Diffusion Models with Transformers,” ICCV, 2023. CVF
  4. C. Saharia, W. Chan, S. Saxena et al., “Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding,” NeurIPS, 2022. arXiv · project page
  5. R. Rombach, A. Blattmann, D. Lorenz, P. Esser and B. Ommer, “High-Resolution Image Synthesis with Latent Diffusion Models,” CVPR, 2022. CVF
  6. C. Meng, Y. He, Y. Song et al., “SDEdit: Guided Image Synthesis and Editing with Stochastic Differential Equations,” ICLR, 2022. arXiv
  7. L. Zhang, A. Rao and M. Agrawala, “Adding Conditional Control to Text-to-Image Diffusion Models,” ICCV, 2023. CVF
  8. A. Hertz, R. Mokady, J. Tenenbaum et al., “Prompt-to-Prompt Image Editing with Cross-Attention Control,” ICLR, 2023. arXiv
  9. T. Brooks, A. Holynski and A. A. Efros, “InstructPix2Pix: Learning to Follow Image Editing Instructions,” CVPR, 2023. CVF
  10. A. Lugmayr, M. Danelljan, A. Romero et al., “RePaint: Inpainting using Denoising Diffusion Probabilistic Models,” CVPR, 2022. CVF
  11. J. Ho and T. Salimans, “Classifier-Free Diffusion Guidance,” NeurIPS 2021 Workshop on Deep Generative Models and Downstream Applications; arXiv:2207.12598, 2022. arXiv
  12. I. J. Goodfellow, J. Pouget-Abadie, M. Mirza et al., “Generative Adversarial Nets,” NeurIPS, 2014. arXiv
  13. D. P. Kingma and M. Welling, “Auto-Encoding Variational Bayes,” ICLR, 2014. arXiv
  14. A. van den Oord, O. Vinyals and K. Kavukcuoglu, “Neural Discrete Representation Learning,” NeurIPS, 2017. arXiv
  15. P. Esser, R. Rombach and B. Ommer, “Taming Transformers for High-Resolution Image Synthesis,” CVPR, 2021. CVF
  16. J. Ho, A. Jain and P. Abbeel, “Denoising Diffusion Probabilistic Models,” NeurIPS, 2020. arXiv
  17. P. Dhariwal and A. Nichol, “Diffusion Models Beat GANs on Image Synthesis,” NeurIPS, 2021. arXiv
  18. Y. Lipman, R. T. Q. Chen, H. Ben-Hamu, M. Nickel and M. Le, “Flow Matching for Generative Modeling,” ICLR, 2023. arXiv
  19. X. Liu, C. Gong and Q. Liu, “Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow,” ICLR, 2023. arXiv
  20. P. Esser, S. Kulal, A. Blattmann et al., “Scaling Rectified Flow Transformers for High-Resolution Image Synthesis,” ICML, PMLR 235, 2024. arXiv
  21. T. Li, Y. Tian, H. Li, M. Deng and K. He, “Autoregressive Image Generation without Vector Quantization,” NeurIPS, 2024. arXiv
  22. A. Sauer, D. Lorenz, A. Blattmann and R. Rombach, “Adversarial Diffusion Distillation,” ECCV, LNCS, pp. 87–103, 2024. doi · arXiv
  23. A. Radford, J. W. Kim, C. Hallacy et al., “Learning Transferable Visual Models From Natural Language Supervision,” ICML, PMLR 139, 2021. arXiv
  24. C. Raffel, N. Shazeer, A. Roberts et al., “Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer,” Journal of Machine Learning Research, vol. 21, 2020. arXiv
  25. D. Podell, Z. English, K. Lacey et al., “SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis,” ICLR, 2024. arXiv
  26. R. Liu, D. Garrette, C. Saharia et al., “Character-Aware Models Improve Visual Text Rendering,” ACL, 2023. ACL Anthology
  27. A. Ramesh, P. Dhariwal, A. Nichol, C. Chu and M. Chen, “Hierarchical Text-Conditional Image Generation with CLIP Latents,” arXiv:2204.06125, 2022. arXiv
  28. H. Chefer, Y. Alaluf, Y. Vinker, L. Wolf and D. Cohen-Or, “Attend-and-Excite: Attention-Based Semantic Guidance for Text-to-Image Diffusion Models,” ACM Transactions on Graphics, vol. 42, no. 4 (SIGGRAPH), 2023. doi · arXiv
  29. O. Ronneberger, P. Fischer and T. Brox, “U-Net: Convolutional Networks for Biomedical Image Segmentation,” MICCAI, 2015. arXiv
  30. J. Yu, Y. Xu, J. Y. Koh et al., “Scaling Autoregressive Models for Content-Rich Text-to-Image Generation,” Transactions on Machine Learning Research, 2022. arXiv
  31. H. Chang, H. Zhang, J. Barber et al., “Muse: Text-To-Image Generation via Masked Generative Transformers,” ICML, PMLR 202, 2023. arXiv
  32. R. Gal, Y. Alaluf, Y. Atzmon et al., “An Image is Worth One Word: Personalizing Text-to-Image Generation using Textual Inversion,” ICLR, 2023. arXiv
  33. N. Ruiz, Y. Li, V. Jampani, Y. Pritch, M. Rubinstein and K. Aberman, “DreamBooth: Fine Tuning Text-to-Image Diffusion Models for Subject-Driven Generation,” CVPR, 2023. CVF
  34. E. J. Hu, Y. Shen, P. Wallis et al., “LoRA: Low-Rank Adaptation of Large Language Models,” ICLR, 2022. arXiv
  35. Y. Huang, J. Huang, Y. Liu et al., “Diffusion Model-Based Image Editing: A Survey,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025. doi · arXiv
  36. J. Song, C. Meng and S. Ermon, “Denoising Diffusion Implicit Models,” ICLR, 2021. arXiv
  37. T. Salimans and J. Ho, “Progressive Distillation for Fast Sampling of Diffusion Models,” ICLR, 2022. arXiv
  38. Y. Song, P. Dhariwal, M. Chen and I. Sutskever, “Consistency Models,” ICML, PMLR 202, 2023. arXiv
  39. P. Sun, Y. Jiang, S. Chen et al., “Autoregressive Model Beats Diffusion: Llama for Scalable Image Generation,” arXiv:2406.06525, 2024. arXiv
  40. H. Chang, H. Zhang, L. Jiang, C. Liu and W. T. Freeman, “MaskGIT: Masked Generative Image Transformer,” CVPR, 2022. CVF
  41. K. Tian, Y. Jiang, Z. Yuan, B. Peng and L. Wang, “Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction,” NeurIPS, 2024. arXiv
  42. L. Yu, J. Lezama, N. B. Gundavarapu et al., “Language Model Beats Diffusion — Tokenizer is Key to Visual Generation,” ICLR, 2024. arXiv
  43. Chameleon Team, “Chameleon: Mixed-Modal Early-Fusion Foundation Models,” arXiv:2405.09818, 2024. arXiv
  44. X. Wang, Y. Cui, J. Wang et al., “Multimodal learning with next-token prediction for large multimodal models,” Nature, vol. 650, pp. 327–333, 2026 (preprint “Emu3: Next-Token Prediction is All You Need,” 2024). doi · arXiv
  45. C. Zhou, L. Yu, A. Babu et al., “Transfusion: Predict the Next Token and Diffuse Images with One Multi-Modal Model,” ICLR, 2025. arXiv
  46. J. Xie, W. Mao, Z. Bai et al., “Show-o: One Single Transformer to Unify Multimodal Understanding and Generation,” ICLR, 2025. arXiv
  47. C. Wu, X. Chen, Z. Wu et al., “Janus: Decoupling Visual Encoding for Unified Multimodal Understanding and Generation,” CVPR, 2025. CVF
  48. T.-Y. Lin, M. Maire, S. Belongie et al., “Microsoft COCO: Common Objects in Context,” ECCV, 2014. arXiv
  49. M. J. Chong and D. Forsyth, “Effectively Unbiased FID and Inception Score and Where to Find Them,” CVPR, 2020. arXiv
  50. G. Parmar, R. Zhang and J.-Y. Zhu, “On Aliased Resizing and Surprising Subtleties in GAN Evaluation,” CVPR, 2022. CVF
  51. T. Kynkäänniemi, T. Karras, M. Aittala, T. Aila and J. Lehtinen, “The Role of ImageNet Classes in Fréchet Inception Distance,” ICLR, 2023. arXiv
  52. G. Stein, J. C. Cresswell, R. Hosseinzadeh et al., “Exposing Flaws of Generative Model Evaluation Metrics and Their Unfair Treatment of Diffusion Models,” NeurIPS, 2023. arXiv
  53. S. Jayasumana, S. Ramalingam, A. Veit, D. Glasner, A. Chakrabarti and S. Kumar, “Rethinking FID: Towards a Better Evaluation Metric for Image Generation,” CVPR, 2024. CVF
  54. J. Hessel, A. Holtzman, M. Forbes, R. Le Bras and Y. Choi, “CLIPScore: A Reference-free Evaluation Metric for Image Captioning,” EMNLP, 2021. ACL Anthology
  55. D. Ghosh, H. Hajishirzi and L. Schmidt, “GenEval: An Object-Focused Framework for Evaluating Text-to-Image Alignment,” NeurIPS Datasets and Benchmarks Track, 2023. arXiv
  56. K. Huang, K. Sun, E. Xie, Z. Li and X. Liu, “T2I-CompBench: A Comprehensive Benchmark for Open-world Compositional Text-to-image Generation,” NeurIPS Datasets and Benchmarks Track, 2023. proceedings
  57. J. Xu, X. Liu, Y. Wu et al., “ImageReward: Learning and Evaluating Human Preferences for Text-to-Image Generation,” NeurIPS, 2023. arXiv
  58. Y. Kirstain, A. Polyak, U. Singer et al., “Pick-a-Pic: An Open Dataset of User Preferences for Text-to-Image Generation,” NeurIPS, 2023. arXiv
  59. T. Lee, M. Yasunaga, C. Meng et al., “Holistic Evaluation of Text-To-Image Models,” NeurIPS Datasets and Benchmarks Track, 2023. arXiv
  60. J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li and L. Fei-Fei, “ImageNet: A Large-Scale Hierarchical Image Database,” CVPR, 2009. doi
  61. C. Schuhmann, R. Beaumont, R. Vencu et al., “LAION-5B: An Open Large-Scale Dataset for Training Next Generation Image-Text Models,” NeurIPS Datasets and Benchmarks Track, 2022. arXiv
  62. A. Birhane, V. U. Prabhu and E. Kahembwe, “Multimodal datasets: misogyny, pornography, and malignant stereotypes,” arXiv:2110.01963, 2021. arXiv
  63. N. Carlini, J. Hayes, M. Nasr et al., “Extracting Training Data from Diffusion Models,” USENIX Security Symposium, 2023. USENIX · arXiv
  64. G. Somepalli, V. Singla, M. Goldblum, J. Geiping and T. Goldstein, “Diffusion Art or Digital Forgery? Investigating Data Replication in Diffusion Models,” CVPR, 2023. CVF
  65. F. Bianchi, P. Kalluri, E. Durmus et al., “Easily Accessible Text-to-Image Generation Amplifies Demographic Stereotypes at Large Scale,” ACM FAccT, 2023. doi
  66. A. S. Luccioni, C. Akiki, M. Mitchell and Y. Jernite, “Stable Bias: Analyzing Societal Representations in Diffusion Models,” NeurIPS Datasets and Benchmarks Track, 2023. arXiv
  67. Coalition for Content Provenance and Authenticity (C2PA), “C2PA Technical Specification,” version 2.1, 2024. specification
  68. Y. Wen, J. Kirchenbauer, J. Geiping and T. Goldstein, “Tree-Ring Watermarks: Fingerprints for Diffusion Images that are Invisible and Robust,” NeurIPS, 2023. arXiv
  69. P. Fernandez, G. Couairon, H. Jégou, M. Douze and T. Furon, “The Stable Signature: Rooting Watermarks in Latent Diffusion Models,” ICCV, 2023. CVF
  70. X. Zhao, K. Zhang, Z. Su et al., “Invisible Image Watermarks Are Provably Removable Using Generative AI,” NeurIPS, 2024. arXiv
  71. M. Saberi, V. S. Sadasivan, K. Rezaei et al., “Robustness of AI-Image Detectors: Fundamental Limits and Practical Attacks,” ICLR, 2024. arXiv
  72. W. Lu, Y. Xu, J. Zhang, C. Wang and D. Tao, “HandRefiner: Refining Malformed Hands in Generated Images by Diffusion-based Conditional Inpainting,” ACM Multimedia, 2024, pp. 7085–7093. doi
  73. C. Zhang, C. Zhang, M. Zhang, I. S. Kweon and J. Kim, “Text-to-image Diffusion Models in Generative AI: A Survey,” arXiv:2303.07909, 2023. arXiv
  74. F.-A. Croitoru, V. Hondru, R. T. Ionescu and M. Shah, “Diffusion Models in Vision: A Survey,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023. doi
  75. J. Xiong, G. Liu, L. Huang et al., “Autoregressive Models in Vision: A Survey,” Transactions on Machine Learning Research, 2025. arXiv