mlsanityData to 7 Oct 2026

RandAR: Decoder-only Autoregressive Visual Generation in Random Orders

arXiv 2412.01827 ↗First posted 2 Dec 2024cs.CV cs.AIcited by 82

On the map: Efficient LLM inference and serving → Fast autoregressive visual generation

Abstract (arXiv)

We introduce RandAR, a decoder-only visual autoregressive (AR) model capable of generating images in arbitrary token orders. Unlike previous decoder-only AR models that rely on a predefined generation order, RandAR removes this inductive bias, unlocking new capabilities in decoder-only generation. Our essential design enables random order by inserting a "position instruction token" before each image token to be predicted, representing the spatial location of the next image token. Trained on randomly permuted token sequences -- a more challenging task than fixed-order generation, RandAR achieves comparable performance to its conventional raster-order counterpart. More importantly, decoder-only transformers trained from random orders acquire new capabilities. For the efficiency bottleneck of AR models, RandAR adopts parallel decoding with KV-Cache at inference time, enjoying 2.5x acceleration without sacrificing generation quality. Additionally, RandAR supports inpainting, outpainting and resolution extrapolation in a zero-shot manner. We hope RandAR inspires new directions for decoder-only visual generation models and broadens their applications across diverse scenarios. Our project page is at https://rand-ar.github.io/.

Key facts. Cited by 82 papers in the map (data through 7 Oct 2026). Placed in Fast autoregressive visual generation, one of 12 directions in Efficient LLM inference and serving. Builds on 58 of its 67 references that are in the map. First posted on arXiv 2 Dec 2024.

Builds on

58 of 67 references.

References · 58

  1. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  2. 2014Adam: A Method for Stochastic Optimizationcited by 36,960
  3. 2020Language Models are Few-Shot Learnerscited by 32,199
  4. 2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,694
  5. 2019Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformercited by 14,491
  6. 2023LLaMA: Open and Efficient Foundation Language Modelscited by 13,958
  7. 2019RoBERTa: A Robustly Optimized BERT Pretraining Approachcited by 13,266
  8. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 12,865
  9. 2021Diffusion Models Beat GANs on Image Synthesiscited by 8,185
  10. 2021Masked Autoencoders Are Scalable Vision Learnerscited by 7,339
  11. 2022Scalable Diffusion Models with Transformerscited by 5,931
  12. 2022Classifier-Free Diffusion Guidancecited by 5,717
38 more
  1. 2021RoFormer: Enhanced Transformer with Rotary Position Embeddingcited by 4,581
  2. 2021Zero-Shot Text-to-Image Generationcited by 4,276
  3. 2020Taming Transformers for High-Resolution Image Synthesiscited by 3,229
  4. 2016Improved Techniques for Training GANscited by 2,938
  5. 2019XLNet: Generalized Autoregressive Pretraining for Language Understandingcited by 2,801
  6. 2018Large Scale GAN Training for High Fidelity Natural Image Synthesiscited by 2,609
  7. 2022InstructPix2Pix: Learning to Follow Image Editing Instructionscited by 2,591
  8. 2020GLU Variants Improve Transformercited by 1,724
  9. 2019Root Mean Square Layer Normalizationcited by 1,425
  10. 2022MaskGIT: Masked Generative Image Transformercited by 1,082
  11. 2024Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Predictioncited by 1,005
  12. 2019Improved Precision and Recall Metric for Assessing Generative Modelscited by 925
  13. 2024Chameleon: Mixed-Modal Early-Fusion Foundation Modelscited by 904
  14. 2016Conditional Image Generation with PixelCNN Decoderscited by 859
  15. 2024Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Headscited by 821
  16. 2024Emu3: Next-Token Prediction is All You Needcited by 761
  17. 2024Autoregressive Model Beats Diffusion: Llama for Scalable Image Generationcited by 760
  18. 2024Show-o: One Single Transformer to Unify Multimodal Understanding and Generationcited by 734
  19. 2024SiT: Exploring Flow and Diffusion-based Generative Models with Scalable Interpolant Transformerscited by 707
  20. 2022Autoregressive Image Generation using Residual Quantizationcited by 706
  21. 2019Fast Transformer Decoding: One Write-Head is All You Needcited by 700
  22. 2024Autoregressive Image Generation without Vector Quantizationcited by 644
  23. 2023Muse: Text-To-Image Generation via Masked Generative Transformerscited by 640
  24. 2023Language Model Beats Diffusion -- Tokenizer is Key to Visual Generationcited by 633
  25. 2022Compositional Visual Generation with Composable Diffusion Modelscited by 619
  26. 2024OLMo: Accelerating the Science of Language Modelscited by 599
  27. 2023Segment Everything Everywhere All at Oncecited by 546
  28. 2023Scaling up GANs for Text-to-Image Synthesiscited by 492
  29. 2022StyleGAN-XL: Scaling StyleGAN to Large Diverse Datasetscited by 489
  30. 2023Emergent Correspondence from Image Diffusioncited by 437
  31. 2022MAGVIT: Masked Generative Video Transformercited by 341
  32. 2024An Image is Worth 32 Tokens for Reconstruction and Generationcited by 295
  33. 2023SqueezeLLM: Dense-and-Sparse Quantizationcited by 285
  34. 2023MDTv2: Masked Diffusion Transformer is a Strong Image Synthesizercited by 260
  35. 2022MAGE: MAsked Generative Encoder to Unify Representation Learning and Image Synthesiscited by 222
  36. 2024Data Engineering for Scaling Language Models to 128K Contextcited by 183
  37. 2024Open-MAGVIT2: An Open-Source Project Toward Democratizing Auto-regressive Visual Generationcited by 153
  38. 2024Fluid: Scaling Autoregressive Text-to-image Generative Models with Continuous Tokenscited by 147

50 most cited of 58 All 58 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 2; 2025: 55; 2026: 25.

82 papers.

Sort:

By area:

  1. 2025Show-o2: Improved Native Unified Multimodal Modelscited by 242
  2. 2025Unified Multimodal Understanding and Generation Models: Advances, Challenges, and Opportunitiescited by 81
  3. 2024Parallelized Autoregressive Visual Generationcited by 58
  4. 2025Unified Autoregressive Visual Generation and Understanding with Continuous Tokenscited by 51
  5. 2025Transition Models: Rethinking the Generative Learning Objectivecited by 32
  6. 2025Neighboring Autoregressive Modeling for Efficient Visual Generationcited by 29
  7. 2025Bridging Continuous and Discrete Tokens for Autoregressive Visual Generationcited by 29
  8. 2025Frequency Autoregressive Image Generation with Continuous Tokenscited by 27
  9. 2025Next Block Prediction: Video Generation via Semi-Autoregressive Modelingcited by 26
  10. 2024Next Patch Prediction for Autoregressive Visual Generationcited by 25
  11. 2025Visual Autoregressive Modeling for Instruction-Guided Image Editingcited by 21
  12. 2025Next-Embedding Prediction Makes Strong Vision Learnerscited by 21
38 more
  1. 2025Towards Sequence Modeling Alignment between Tokenizer and Autoregressive Modelcited by 20
  2. 2025Autoregressive Image Generation with Randomized Parallel Decodingcited by 19
  3. 2025Macro-from-Micro Planning for High-Quality and Parallelized Autoregressive Long Video Generationcited by 19
  4. 2025Latent Denoising Makes Good Tokenizerscited by 18
  5. 2025Multi-Token Prediction Needs Registerscited by 16
  6. 2025Any-Order GPT as Masked Diffusion Model: Decoupling Formulation and Architecturecited by 16
  7. 2025Hyperspherical Latents Improve Continuous-Token Autoregressive Generationcited by 16
  8. 2025MergeVQ: A Unified Framework for Visual Generation and Representation with Disentangled Token Merging and Quantizationcited by 12
  9. 2025SoFlow: Solution Flow Models for One-Step Generative Modelingcited by 12
  10. 2025Self-Speculative Masked Diffusionscited by 11
  11. 2025Scalable Training for Vector-Quantized Networks with 100% Codebook Utilizationcited by 10
  12. 2025Image Tokenizer Needs Post-Trainingcited by 10
  13. 2025Robust Latent Matters: Boosting Image Generation with Sampling Error Synthesiscited by 9
  14. 2025D-AR: Diffusion via Autoregressive Modelscited by 9
  15. 2025Unleashing the Power of Chain-of-Prediction for Monocular 3D Object Detectioncited by 8
  16. 2025Locality-aware Parallel Decoding for Efficient Autoregressive Image Generationcited by 8
  17. 2025Improving Autoregressive Image Generation through Coarse-to-Fine Token Predictioncited by 7
  18. 2025Make It Efficient: Dynamic Sparse Attention for Autoregressive Image Generationcited by 7
  19. 2025DiCo: Revitalizing ConvNets for Scalable and Efficient Diffusion Modelingcited by 6
  20. 2025Rethinking Discrete Tokens: Treating Them as Conditions for Continuous Autoregressive Image Synthesiscited by 6
  21. 2025REAR: Rethinking Visual Autoregressive Models via Generator-Tokenizer Consistency Regularizationcited by 6
  22. 2025SimFlow: Simplified and End-to-End Training of Latent Normalizing Flowscited by 6
  23. 2025Spherical Leech Quantization for Visual Tokenization and Generationcited by 6
  24. 2025VA-π: Variational Policy Alignment for Pixel-Aware Autoregressive Generationcited by 6
  25. 2025Unified Cross-Scale 3D Generation and Understanding via Autoregressive Modelingcited by 5
  26. 2025NEP: Autoregressive Image Editing via Next Editing Token Predictioncited by 5
  27. 2025Next Visual Granularity Generationcited by 5
  28. 2026SparVAR: Exploring Sparsity in Visual AutoRegressive Modeling for Training-Free Accelerationcited by 4
  29. 2026BitDance: Scaling Autoregressive Generative Models with Binary Tokenscited by 4
  30. 2025From Prediction to Perfection: Introducing Refinement to Autoregressive Image Generationcited by 3
  31. 2025ReDDiT: Rehashing Noise for Discrete Visual Generationcited by 3
  32. 2026Generative Refinement Networks for Visual Synthesiscited by 3
  33. 2025NeoBabel: A Multilingual Open Tower for Visual Generationcited by 2
  34. 2025Exploiting Discriminative Codebook Prior for Autoregressive Image Generationcited by 2
  35. 2025VCE: Safe Autoregressive Image Generation via Visual Contrast Exploitationcited by 2
  36. 2025SoftCFG: Uncertainty-guided Stable Guidance for Visual Autoregressive Modelcited by 2
  37. 2025Visual Generation Tuningcited by 2
  38. 2025Autoregressive Image Generation Needs Only a Few Lines of Cached Tokenscited by 2

50 most cited of 82 All 82 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2025Progress by Pieces: Test-Time Scaling for Autoregressive Image Generation
  2. 2025From Sequential to Spatial: Reordering Autoregression for Efficient Visual Generation
  3. 2025VVS: Accelerating Speculative Decoding for Visual Autoregressive Generation via Partial Verification Skipping
  4. 2024LANTERN: Accelerating Visual Autoregressive Models with Relaxed Speculative Decoding
  5. 2025SimpleAR: Pushing the Frontier of Autoregressive Visual Generation through Pretraining, SFT, and RL
  6. 2024ZipAR: Parallel Auto-regressive Image Generation through Spatial Locality
  7. 2025Learning to Expand Images for Efficient Visual Autoregressive Modeling
  8. 2026SSD: Spatially Speculative Decoding Accelerates Autoregressive Image Generation