mlsanityData to 14 Oct 2026

SSD: Spatially Speculative Decoding Accelerates Autoregressive Image Generation

arXiv 2606.20543 ↗First posted 18 Jun 2026cs.CVcited by 2

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

Abstract (arXiv)

Autoregressive image models treat images as 1D token sequences, inheriting the next-token factorization of language models. This flattening discards a useful property of images: visual tokens are organized and correlated in two-dimensional space. We introduce Spatially Speculative Decoding (SSD), an inference-time decoding framework that exploits this spatial structure for accelerated generation. Rather than speculating only along the flattened sequence, SSD predicts both the adjacent horizontal token and the token directly below it, allowing multiple spatial directions to advance in parallel. This reduces the number of backbone forward evaluations and alleviates the memory bottleneck of autoregressive decoding. SSD accelerates image generation by up to 11.03x in wall-clock time while maintaining generation quality on DPG-Bench and GenEval. These results show that spatial structure provides a simple and effective source of parallelism for faster autoregressive image generation.

Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Placed in Fast autoregressive visual generation, one of 12 directions in Efficient LLM inference and serving. Builds on 47 of its 51 references that are in the map. First posted on arXiv 18 Jun 2026.

Builds on

47 of 51 references.

Closest ancestors · 8

Same direction, newest first

  1. 2026Annealed Relaxation of Speculative Decoding for Faster Autoregressive Image Generation14 Jan
  2. 2026Multi-Scale Local Speculative Decoding for Image Generation8 Jan
  3. 2025Grouped Speculative Decoding for Autoregressive Image Generation11 Aug
  4. 2025Locality-aware Parallel Decoding for Efficient Autoregressive Image Generation2 Jul
  5. 2025Autoregressive Image Generation with Randomized Parallel Decoding13 Mar
  6. 2024ZipAR: Parallel Auto-regressive Image Generation through Spatial Locality5 Dec
  7. 2024LANTERN: Accelerating Visual Autoregressive Models with Relaxed Speculative Decoding4 Oct
  8. 2024Accelerating Auto-regressive Text-to-Image Generation with Training-free Speculative Jacobi Decoding2 Oct

Other references · 39

  1. 2017Attention Is All You Needcited by 60,361
  2. 2015Distilling the Knowledge in a Neural Networkcited by 10,324
  3. 2022Classifier-Free Diffusion Guidancecited by 5,847
  4. 2017Neural Discrete Representation Learningcited by 4,712
  5. 2024GPT-4o System Cardcited by 4,588
  6. 2021Zero-Shot Text-to-Image Generationcited by 4,288
  7. 2020Taming Transformers for High-Resolution Image Synthesiscited by 3,253
  8. 2015Fast R-CNNcited by 2,933
  9. 2020GLU Variants Improve Transformercited by 1,804
  10. 2022Fast Inference from Transformers via Speculative Decodingcited by 1,653
  11. 2019Root Mean Square Layer Normalizationcited by 1,480
  12. 2022Scaling Autoregressive Models for Content-Rich Text-to-Image Generationcited by 1,295
27 more
  1. 2023Accelerating Large Language Model Decoding with Speculative Samplingcited by 1,006
  2. 2023GenEval: An Object-Focused Framework for Evaluating Text-to-Image Alignmentcited by 972
  3. 2024Chameleon: Mixed-Modal Early-Fusion Foundation Modelscited by 927
  4. 2024Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Headscited by 843
  5. 2025Janus-Pro: Unified Multimodal Understanding and Generation with Data and Model Scalingcited by 812
  6. 2024Emu3: Next-Token Prediction is All You Needcited by 781
  7. 2024Autoregressive Model Beats Diffusion: Llama for Scalable Image Generationcited by 777
  8. 2023Multimodal Chain-of-Thought Reasoning in Language Modelscited by 752
  9. 2024Show-o: One Single Transformer to Unify Multimodal Understanding and Generationcited by 752
  10. 2023Language Model Beats Diffusion -- Tokenizer is Key to Visual Generationcited by 640
  11. 2024EAGLE: Speculative Sampling Requires Rethinking Feature Uncertaintycited by 578
  12. 2022Efficiently Scaling Transformer Inferencecited by 539
  13. 2024ELLA: Equip Diffusion Models with LLM for Enhanced Semantic Alignmentcited by 475
  14. 2024Janus: Decoupling Visual Encoding for Unified Multimodal Understanding and Generationcited by 457
  15. 2024EAGLE-2: Faster Inference of Language Models with Dynamic Draft Treescited by 370
  16. 2023SpecInfer: Accelerating Generative Large Language Model Serving with Tree-based Speculative Inference and Verificationcited by 365
  17. 2024Better & Faster Large Language Models via Multi-token Predictioncited by 360
  18. 2025EAGLE-3: Scaling up Inference Acceleration of Large Language Models via Training-Time Testcited by 331
  19. 2024Break the Sequential Dependency of LLM Inference Using Lookahead Decodingcited by 298
  20. 2024Lumina-mGPT: Illuminate Flexible Photorealistic Text-to-Image Generation with Multimodal Generative Pretrainingcited by 150
  21. 2023Accelerating Transformer Inference for Translation via Parallel Decodingcited by 124
  22. 2024CLLMs: Consistency Large Language Modelscited by 72
  23. 2024Parallelized Autoregressive Visual Generationcited by 60
  24. 2025Neighboring Autoregressive Modeling for Efficient Visual Generationcited by 29
  25. 2025L-MTP: Leap Multi-Token Prediction Beyond Adjacent Context for Large Language Modelscited by 15
  26. 2026FlashAR: Efficient Post-Training Acceleration for Autoregressive Image Generationcited by 3
  27. 2026Parallel Jacobi Decoding for Fast Autoregressive Image Generationcited by 2

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026Spectral Origins of the Self-Correction Blind Spot in Autoregressive Generationcited by 0
  2. 2026Is Multimodal Speculative Decoding Ready for Diffusion-Based Parallel Drafting? A Survey and Empirical Diagnosiscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025From Sequential to Spatial: Reordering Autoregression for Efficient Visual Generation
  2. 2025Speculative Coupled Decoding for Training-Free Lossless Acceleration of Autoregressive Visual Generation
  3. 2025Hawk: Leveraging Spatial Context for Faster Autoregressive Text-to-Image Generation
  4. 2025Towards Better & Faster Autoregressive Image Generation: From the Perspective of Entropy
  5. 2025SSD: Spatial-Semantic Head Decoupling for Efficient Autoregressive Image Generation
  6. 2025SJD++: Improved Speculative Jacobi Decoding for Training-free Acceleration of Discrete Auto-regressive Text-to-Image Generation
  7. 2026CSD: Content-aware Speculative Decoding for Efficient Image Generation
  8. 2024Continuous Speculative Decoding for Autoregressive Image Generation