| Representation Fréchet Loss for Visual Generation | ✓ Link | 0.78 | | | pMF-H + FD-loss (SIM) | 2026-04-30 |
| MixFlow Training: Alleviating Exposure Bias with Slowed Interpolation Mixture | | 1.10 | | 256.9 | MixFlow + RAE | 2025-12-22 |
| Diffusion Transformers with Representation Autoencoders | ✓ Link | 1.13 | | | DiT^DH-XL (DINOv2-B) | 2025-10-13 |
| Direct Discriminative Optimization: Your Likelihood-Based Visual Generative Model is Secretly a GAN Discriminator | ✓ Link | 1.21 | 50 | | EDM2-L + DDO (SD-VAE, 25 steps, DPM-Solver-v3) | 2025-03-03 |
| GenFirst: Generation Before Reconstruction for Stable End-to-End Latent Generative Modeling | | 1.23 | | 286.1 | GenFirst EiT (675M, 400 epochs) | 2026-08-29 |
| Unified Continuous Generative Models | ✓ Link | 1.24 | 300 | | DDT-XL/2 + UCGM-S (SD-VAE + 150 sampling steps + CFG) | 2025-05-12 |
| Guiding a Diffusion Model with a Bad Version of Itself | ✓ Link | 1.25 | | | EDM2-XXL Autoguidance | 2024-06-04 |
| Unified Continuous Generative Models | ✓ Link | 1.25 | 200 | | DDT-XL/2 + UCGM-S (SD-VAE + 100 sampling steps + CFG) | 2025-05-12 |
| Hi-DiT: Hybrid Latent-Pixel Diffusion Transformer for Image Generation | | 1.26 | | 295.0 | Hi-DiT (CFG=4.6) | 2026-09-08 |
| DDT: Decoupled Diffusion Transformer | ✓ Link | 1.28 | 500 | 305 | DDT-XL/2(22en6de 675M + guidance interval ) | 2025-04-08 |
| Guiding a Diffusion Model with a Bad Version of Itself | ✓ Link | 1.34 | | | EDM2- S Autoguidance (XS, T /16) | 2024-06-04 |
| Adversarial Score identity Distillation: Rapidly Surpassing the Teacher in One Step | ✓ Link | 1.366 | 1 | | SiDA-EDM2-XXL (1.5B) | 2024-10-19 |
| Adversarial Score identity Distillation: Rapidly Surpassing the Teacher in One Step | ✓ Link | 1.379 | 1 | | SiDA-EDM2-XL (1.1B) | 2024-10-19 |
| Applying Guidance in a Limited Interval Improves Sample and Distribution Quality in Diffusion Models | ✓ Link | 1.40 | | | EDM2-XXL w/ guidance interval | 2024-04-11 |
| Adversarial Score identity Distillation: Rapidly Surpassing the Teacher in One Step | ✓ Link | 1.413 | 1 | | SiDA-EDM2-L (777M) | 2024-10-19 |
| Simpler Diffusion (SiD2): 1.5 FID on ImageNet512 with pixel-space diffusion | | 1.48 | | | SiD2 | 2024-10-25 |
| PixelDiT2: Representation-Grounded Pixel Diffusion Transformers | ✓ Link | 1.48 | | 295.7 | PixelDiT2-H/16 | 2026-09-21 |
| Adversarial Score identity Distillation: Rapidly Surpassing the Teacher in One Step | ✓ Link | 1.488 | 1 | | SiDA-EDM2-M (498M) | 2024-10-19 |
| SpeeDiff: Scalable Pixel-Anchored End-to-End Latent Diffusion Model | | 1.53 | | 258.47 | SpeeDiff-XL (w/ REPA++) | 2026-06-03 |
| Condition Errors Refinement in Autoregressive Image Generation with Diffusion Loss | | 1.58 | | 302.3 | Autoregressive Condition Optimization (ACO) (≈481M) | 2026-02-02 |
| Energy-Guided Flow Matching | ✓ Link | 1.58 | | | HyperDiT-H + EG-FM | 2026-08-06 |
| Beyond the Estimate: Wiener-Informed Decoding for Easier Training of Pixel-Space Flow Models | ✓ Link | 1.64 | | | PixelDiT-XL+WiD | 2026-10-26 |
| Adversarial Score identity Distillation: Rapidly Surpassing the Teacher in One Step | ✓ Link | 1.669 | 1 | | SiDA-EDM2-S (280M) | 2024-10-19 |
| Hyperspherical Latents Improve Continuous-Token Autoregressive Generation | ✓ Link | 1.67 | | 292.5 | SphereAR-L | 2025-09-29 |
| Applying Guidance in a Limited Interval Improves Sample and Distribution Quality in Diffusion Models | ✓ Link | 1.68 | | | EDM2-S w/ guidance interval | 2024-04-11 |
| Generative Modeling with Explicit Memory | ✓ Link | 1.71 | | | GMem | 2024-12-11 |
| Deep Compression Autoencoder for Efficient High-Resolution Diffusion Models | ✓ Link | 1.72 | | | DC-AE-f32 + USiT-2B | 2024-10-14 |
| Autoregressive Image Generation without Vector Quantization | ✓ Link | 1.73 | | | MAR-L, Diff Loss | 2024-06-17 |
| Self-Improving Diffusion Models with Synthetic Data | | 1.73 | | | SIMS | 2024-08-29 |
| Learning Representations to Guide Pixel Diffusion Transformers | ✓ Link | 1.74 | | 287.5 | RG-XL/16 | 2026-10-09 |
| Persistence Forcing: Exploiting Feature Specialization in Pixel-Space Diffusion | | 1.76 | | 335.3 | PerF-H/32 | 2026-09-28 |
| Back to Basics: Let Denoising Generative Models Denoise | ✓ Link | 1.78 | | 306.8 | JiT-G/32 | 2025-11-17 |
| PaGoDA: Progressive Growing of a One-Step Generator from a Low-Resolution Diffusion Teacher | ✓ Link | 1.80 | | | PaGoDA | 2024-05-23 |
| Analyzing and Improving the Training Dynamics of Diffusion Models | ✓ Link | 1.81 | 126 | | EDM2-XXL | 2023-12-05 |
| Exposure Bias Can Alleviate Itself via Directional and Frequency Rectification in Flow Matching | | 1.82 | | 285.42 | DEFAR (DDT-XL/2) | 2026-06-26 |
| A Frozen Pixel-Space Diffusion Model Can Guide Itself with Its Own Samples | ✓ Link | 1.84 | | 305.2 | JiT-H/32 + SSG | 2026-07-31 |
| Analyzing and Improving the Training Dynamics of Diffusion Models | ✓ Link | 1.85 | 126 | | EDM2-XL | 2023-12-05 |
| Registers Matter for Pixel-Space Diffusion Transformers | ✓ Link | 1.87 | | | JiT-H/32 with Register Guidance + CFG | 2026-07-06 |
| Analyzing and Improving the Training Dynamics of Diffusion Models | ✓ Link | 1.88 | 126 | | EDM2-L | 2023-12-05 |
| Adversarial Score identity Distillation: Rapidly Surpassing the Teacher in One Step | ✓ Link | 1.888 | 1 | | SiD-EDM2-XL (1.1B) | 2024-10-19 |
| LDM-is-AE: Latent Diffusion Model is an Auto-Encoder for End-to-End Image Generation | ✓ Link | 1.90 | | 320 | LDM-is-AE | 2026-09-29 |
| Adversarial Score identity Distillation: Rapidly Surpassing the Teacher in One Step | ✓ Link | 1.907 | 1 | | SiD-EDM2-L (777M) | 2024-10-19 |
| Language Model Beats Diffusion -- Tokenizer is Key to Visual Generation | ✓ Link | 1.91 | | 324.3 | MAGVIT-v2 | 2023-10-09 |
| PixelU: A U-Shaped Transformer for Efficient End-to-End Pixel Diffusion | ✓ Link | 1.92 | | 322.1 | PixelU-H/32 (extra data) | 2026-06-26 |
| Adversarial Score identity Distillation: Rapidly Surpassing the Teacher in One Step | ✓ Link | 1.969 | 1 | | SiD-EDM2-XXL (1.5B) | 2024-10-19 |
| Analyzing and Improving the Training Dynamics of Diffusion Models | ✓ Link | 2.01 | 126 | | EDM2-M | 2023-12-05 |
| Frequency-Aware Flow Matching for High-Quality Image Generation | ✓ Link | 2.02 | | 285.3 | FreqFlow-L | 2026-04-16 |
| Adversarial Score identity Distillation: Rapidly Surpassing the Teacher in One Step | ✓ Link | 2.06 | 1 | | SiD-EDM2-M (498M) | 2024-10-19 |
| SRC-Flow: Compact Semantic Representations Enable Normalizing Flows for Image Generation | | 2.07 | | 305.7 | SRC-Flow | 2026-05-18 |
| Representation Alignment for Generation: Training Diffusion Transformers Is Easier Than You Think | ✓ Link | 2.08 | | 274.6 | SiT-XL/2 + REPA (200 epochs, w=1.35) (extra data) | 2024-10-09 |
| From SRA to Self-Flow: Data Augmentation or Self-Supervision? | | 2.08 | | 282.7 | SiT (XL/2) with self-representation alignment + dual-timestep scheduling + Attention Separation | 2026-07-02 |
| Locality-aware Parallel Decoding for Efficient Autoregressive Image Generation | ✓ Link | 2.10 | | 326.0 | LPD-XL | 2025-07-02 |
| An Image is Worth 32 Tokens for Reconstruction and Generation | ✓ Link | 2.13 | | | TiTok-B-128 | 2024-06-11 |
| Adversarial Score identity Distillation: Rapidly Surpassing the Teacher in One Step | ✓ Link | 2.156 | 1 | | SiDA-EDM2-XS (125M) | 2024-10-19 |
| Analyzing and Improving the Training Dynamics of Diffusion Models | ✓ Link | 2.23 | 126 | | EDM2-S | 2023-12-05 |
| CADS: Unleashing the Diversity of Diffusion Models through Condition-Annealed Sampling | | 2.31 | | | DiT-XL/2 with CADS | 2023-10-26 |
| FREPix: Frequency-Heterogeneous Flow Matching for Pixel-Space Image Generation | | 2.38 | | 334.7 | FREPix-XL | 2026-05-07 |
| SSG: Scaled Spatial Guidance for Multi-Scale Visual Autoregressive Generation | ✓ Link | 2.39 | | 320.6 | VAR-d36 + SSG | 2026-02-05 |
| StyleGAN-XL: Scaling StyleGAN to Large Diverse Datasets | ✓ Link | 2.40 | | | StyleGAN-XL | 2022-02-01 |
| MPDiT: Multi-Patch Global-to-Local Transformer Architecture For Efficient Flow Matching and Diffusion Model | ✓ Link | 2.47 | | | MPDiT-XL | 2026-03-27 |
| An Image is Worth 32 Tokens for Reconstruction and Generation | ✓ Link | 2.49 | | | TiTok-L-64 | 2024-06-11 |
| Optimizing Visual Generative Models via Distribution-wise Rewards | | 2.52 | | | EDM2-S (post-hoc model merging with RL) | 2026-07-02 |
| Locality-aware Parallel Decoding for Efficient Autoregressive Image Generation | ✓ Link | 2.54 | | 292.2 | LPD-L | 2025-07-02 |
| SiT: Exploring Flow and Diffusion-based Generative Models with Scalable Interpolant Transformers | ✓ Link | 2.62 | | 252.21 | SiT-XL/2 (cfg=1.5, SDE) (extra data) | 2024-01-16 |
| Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction | ✓ Link | 2.63 | | 303.2 | VAR-d36-s (extra data) | 2024-04-03 |
| DiffiT: Diffusion Vision Transformers for Image Generation | ✓ Link | 2.67 | | 252.12 | DiffiT | 2023-12-04 |
| Adversarial Score identity Distillation: Rapidly Surpassing the Teacher in One Step | ✓ Link | 2.707 | 1 | | SiD-EDM2-S (280M) | 2024-10-19 |
| QuadTok: Quadtree Visual Tokenizer for Autoregressive Image Generation | ✓ Link | 2.734 | | 268.01 | QuadTok | 2026-10-07 |
| SA-Solver: Stochastic Adams Solver for Fast Sampling of Diffusion Models | ✓ Link | 2.80 | | | DiT-XL/2 with SA-Solver | 2023-09-10 |
| Diffusion Model is Effectively Its Own Teacher | | 2.86 | | | DiT-XL/2 + iSSD | 2025-06-11 |
| Alleviating Distortion in Image Generation via Multi-Resolution Diffusion Models and Time-Dependent Layer Normalization | ✓ Link | 2.89 | | | DiMR-XL/3R | 2024-06-13 |
| Analyzing and Improving the Training Dynamics of Diffusion Models | ✓ Link | 2.91 | 126 | | EDM2-XS | 2023-12-05 |
| GIVT: Generative Infinite-Vocabulary Transformers | ✓ Link | 2.92 | | | GIVT-Causal-L+A | 2023-12-04 |
| Flow Straighter and Faster: Efficient One-Step Generative Modeling via MeanFlow on Rectified Trajectories | ✓ Link | 3.03 | | | Re-Meanflow (EDM2-S, 280.5M) | 2025-11-28 |
| Scalable Diffusion Models with Transformers | ✓ Link | 3.04 | | 240.82 | DiT-XL/2 | 2022-12-19 |
| Language Model Beats Diffusion -- Tokenizer is Key to Visual Generation | ✓ Link | 3.07 | | 213.1 | MAGVIT-v2 (w/o guidance) | 2023-10-09 |
| Dino Forcing Flow Models: Do not denoise what you can predict | | 3.11 | | 274.5 | DF-SiT-XL/2 | 2026-10-08 |
| MeanFlow Transformers with Representation Autoencoders | ✓ Link | 3.23 | | | MF-RAE (1 NFE) (extra data) | 2025-11-17 |
| Adversarial Score identity Distillation: Rapidly Surpassing the Teacher in One Step | ✓ Link | 3.353 | 1 | | SiD-EDM2-XS (125M) | 2024-10-19 |
| Discrete Predictor-Corrector Diffusion Models for Image Synthesis | | 3.54 | | 350.2 | DPC-U | 2022-09-29 |
| High-Resolution Image Synthesis with Latent Diffusion Models | ✓ Link | 3.60 | | 247.67 | Latent Diffusion (LDM-4-G) | 2021-12-20 |
| Revisiting Non-Autoregressive Transformers for Efficient Image Synthesis | ✓ Link | 3.74 | | 286.2 | AutoNAT-L | 2024-06-08 |
| Polynomial Implicit Neural Representations For Large Diverse Datasets | ✓ Link | 3.81 | | | Poly-INR | 2023-03-20 |
| Diffusion Models Beat GANs on Image Synthesis | ✓ Link | 3.85 | | 221.72 | ADM-G, ADM-U | 2021-05-11 |
| Revisiting Diffusion Model Predictions Through Dimensionality | | 4.03 | | 277.1 | k-Diff (JiT-B/32) | 2026-01-29 |
| Speedrunning ImageNet Diffusion | ✓ Link | 4.23 | | 192.34 | SR-DiT-B/1 | 2025-12-13 |
| Simple diffusion: End-to-end diffusion for high resolution images | ✓ Link | 4.28 | | 171 | simple diffusion (U-Net) | 2023-01-26 |
| MaskGIT: Masked Generative Image Transformer | ✓ Link | 4.46 | | 342.0 | MaskGIT (a=0.05) | 2022-02-08 |
| Simple diffusion: End-to-end diffusion for high resolution images | ✓ Link | 4.53 | | 205.3 | simple diffusion (U-ViT, L) | 2023-01-26 |
| MEPA: Multi-Scale Representation Alignment for Visual Autoregressive Modeling with Mixture of Experts | | 7.29 | | 199.2 | MEPA-d20 | 2026-07-01 |
| MaskGIT: Masked Generative Image Transformer | ✓ Link | 7.32 | | 156.0 | MaskGIT | 2022-02-08 |
| Residual Learning in Diffusion Models | | 7.65 | | 180.62 | Residual Learning in Diffusion Models (ADM-G) | 2024-06-17 |
| Diffusion Models Beat GANs on Image Synthesis | ✓ Link | 7.72 | | 172.71 | ADM-G | 2021-05-11 |
| Energy-Guided Flow Matching | ✓ Link | | | 295.5 | PixelDiT-XL + EG-FM | 2026-08-06 |