Research map / Image generation / Autoregressive and latent generation
Beyond Next-Token: Next-X Prediction for Autoregressive Visual Generation
Sucheng Ren, Qihang Yu, Ju He, Xiaohui Shen, Alan Yuille, Liang-Chieh Chen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| predict 8x8 cells instead of individual tokens (ablation) | xAR-B · ImageNet-256 (400 epochs) | FID: 3.51 -> 2.48 | Improved |
| predict cells instead of coarse-to-fine scales (ablation) | xAR-B · ImageNet-256 (400 epochs) | FID: 2.90 -> 2.48 | Improved |
| predict cells instead of the entire image (ablation) | xAR-B · ImageNet-256 (400 epochs) | FID: 3.13 -> 2.48 | Improved |
| increase cell size from 1x1 to 8x8 (ablation) | xAR-B · ImageNet-256 (400 epochs) | FID: 3.51 -> 2.48 | Improved |
| increase cell size beyond 8x8 to 16x16 (ablation) | xAR-B · ImageNet-256 (400 epochs) | FID: 2.48 -> 3.13 | Hurt |
| add Noisy Context Learning instead of clean-context teacher forcing (ablation) | xAR-B · ImageNet-256 (400 epochs) | FID: 3.45 -> 2.48 | Improved |
| sample noise time steps randomly rather than with increasing or decreasing order (ablation) | xAR-B Noisy Context Learning · ImageNet-256 (400 epochs) | FID: 2.78 -> 2.48 (decreasing vs random) | Improved |
| use rectangular cells instead of square cells (ablation) | xAR · ImageNet-256 | FID | No clear effect |
| use xAR-B with each entity type versus prior method of same entity type (ablation) | xAR-B vs LlamaGen-L, PAR-L, DiT-L/2, VAR-d16 · ImageNet-256 (400 epochs) | FID: 3.80 -> 3.51 (token), 5.02 -> 3.13 (image), 3.30 -> 2.90 (scale) | Improved |
| use xAR-L instead of REPA | ImageNet-256 sampling · ImageNet-256 throughput | images per second: 0.6 -> 3.2 | Improved |
Closest papers by meaning
- Hierarchical Masked Autoregressive Models with Low-Resolution Token Pivots (ICML 2025 · similarity 0.91)
- Understand Before You Generate: Self-Guided Training for Autoregressive Image Generation (NeurIPS 2025 · similarity 0.90)
- Vision Foundation Models as Effective Visual Tokenizers for Autoregressive Generation (NeurIPS 2025 · similarity 0.90)
- Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction (NeurIPS 2024 · similarity 0.90)
- Improving Autoregressive Visual Generation with Cluster-Oriented Token Prediction (CVPR 2025 · similarity 0.89)
- BitDance: Scaling Autoregressive Generative Models with Binary Tokens (NeurIPS 2026 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
