Full search page →

Esser et al., “Scaling Rectified Flow Transformers for High-Resolution Image Synthesis” (ICML 2024)

Optional · paper guide 2 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

Stable Diffusion 3’s research paper [20] asks which combination of training objective, time sampling, representation and backbone scales well for text-to-image generation. Changing only the sampler name would not answer that question. The paper studies rectified-flow training and a joint text–image transformer while also changing autoencoder capacity and caption data.

Method

The left panel assembles text, noisy image and time information; the right panel shows separate modality transformations participating in joint attention.
Original paper figure 2 — The left panel assembles text, noisy image and time information; the right panel shows separate modality transformations participating in joint attention.

Figure source: [20], PDF page 5. Original PDF.

Regress a velocity along noisy latent paths. Using the paper’s data-to-noise clock,

zt=(1−t)z0+tϵ,L=Ez0,ϵ,t,c∥vθ(zt,t,c)−(ϵ−z0)∥22.z_t=(1-t)z_0+t\epsilon,\qquad \mathcal L=\mathbb E_{z_0,\epsilon,t,c} \|v_\theta(z_t,t,c)-(\epsilon-z_0)\|_2^2.

z0z_0 is a clean image latent, ϵ\epsilon is independent Gaussian noise, t∈[0,1]t\in[0,1] is time, cc denotes text/global conditioning, and vθv_\theta predicts velocity. Sampling integrates from noise at t=1t=1 toward data at t=0t=0, then decodes. A training pair follows a straight interpolation; the learned marginal flow trajectory need not be exactly straight.

The paper compares distributions for drawing tt, including logit-normal sampling: draw u∼N(m,s2)u\sim\mathcal N(m,s^2) and set t=1/(1+e−u)t=1/(1+e^{-u}). Here m,sm,s control emphasis on different noise levels. Concentrating training near intermediate times changes error allocation across the path; it is not a declaration that inference must use the same random time distribution.

Let text and image representations interact. MM-DiT projects latent image patches and text representations into compatible token widths. The modalities use separate transformation weights, then concatenate their query/key/value sequences for joint attention. Both streams can be updated using information from the other. This is more specific than adding ordinary one-way text cross-attention to an image-only U-Net.

Conditioning combines two CLIP encoders and T5-XXL. Pooled text features and timestep embeddings modulate blocks, while sequence features participate in attention. Encoder dropout during training makes subsets of text encoders usable at inference; dropping an encoder is still a change to the conditioning setting.

Scale the surrounding system too. The paper increases latent channels, studies model depths up to its 8B configuration, and mixes original captions with generated captions. A stronger autoencoder, richer captions and a larger transformer are separate interventions. The reference guide concerns the ICML 2024 research system, not every later product carrying an SD3-family name.

Read the figure and choose a baseline. Follow text and image tokens into joint attention while keeping their separate transformations in view. Then locate the time condition: it specifies where the noisy latent lies along the training path, not the number of words or an image-quality target. To interpret the final results, make a small ledger of changes to the objective, time distribution, caption mixture, latent channels, model size and preference optimization. A gain from a row that changes several entries cannot isolate MM-DiT alone. The autoencoder reconstruction table and text-to-image alignment table also evaluate different parts of the system.

Results

Table 5 reports overall GenEval 0.68 for the depth-38 model at 512×512512\times512, 0.71 with preference optimization at that resolution, and 0.74 for the preference-optimized 1024×10241024\times1024 configuration. These rows change optimization and/or resolution and must not be attributed to architecture alone.

Table 3’s autoencoder reconstruction PSNR rises from 25.12 to 28.62 dB when latent channels increase from four to sixteen at the same spatial factor eight. That is a reconstruction result, not a text-to-image PSNR benchmark.

Ablations

Tables 1–2 compare objective/time-sampling variants using controlled smaller experiments and multiple sampling settings. Rectified flow with uniform time sampling does not universally win; tuned time distributions are central to the result, and rankings change with the number of sampling steps.

Table 4 raises overall GenEval from 43.27% to 49.78% when replacing the original-caption-only setup with a 50/50 original/synthetic-caption mixture. Data descriptions can improve alignment independently of the final backbone.

The T5 omission experiment reports approximately 50% aesthetic, 46% prompt-adherence and 38% typography win rates against the full setting. Similar aesthetics therefore do not mean equivalent text rendering. Figure 7 further shows that a larger latent representation needs sufficient prior capacity; low-capacity conclusions need not survive scaling unchanged.

Conclusion

SD3 is a documented combination of flow training, time weighting, multimodal attention and data/representation choices. Its results are best understood by tracking which of those components changed in each experiment.

Problems

Large-scale training data and compute restrict full reproduction. Small-scale ablations do not isolate every interaction in the largest model, and benchmark detectors cover only part of prompt understanding. Typography remains difficult, larger latents increase modeling cost, and human-preference results concern particular prompts and contemporary comparison systems.

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