Research map / Image generation / Text-to-image diffusion
NextStep-1: Toward Autoregressive Image Generation with Continuous Tokens at Scale
Chunrui Han, Guopeng Li, Jingwei Wu, Quan Sun, Yan Cai, Yuang Peng, Zheng Ge, Deyu Zhou, Haomiao Tang, Hongyu Zhou, Kenkun Liu, Shu-Tao Xia and 4 more
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 |
|---|---|---|---|
| add Self-CoT reasoning prompt at inference | NextStep-1 · GenEval | overall: 0.63 -> 0.73 | Improved |
| add Self-CoT reasoning prompt at inference | NextStep-1 · WISE | overall: 0.54 -> 0.67 | Improved |
| add Self-CoT at inference | NextStep-1 · GenAI-Bench Advanced | score: 0.67 -> 0.74 | Improved |
| add prompt rewriting | NextStep-1 · WISE | overall: 0.54 -> 0.79 | Improved |
| vary flow-matching head size (small, base, large) (ablation) | NextStep-1 · image generation quality (qualitative) | image quality | No clear effect |
| add channel-wise normalization in tokenizer (ablation) | NextStep-1 tokenizer · image generation at CFG 3.0 | artifact/stability | Improved |
| raise noise intensity in tokenizer training (gamma=0.5) (ablation) | NextStep-1 tokenizer · generation quality | generation loss vs image quality | Improved |
| use higher guidance scale (CFG 3.0 vs 1.5) without per-token normalization (ablation) | NextStep-1 · image generation | per-token mean/variance | Hurt |
| SFT on small high-quality datasets | NextStep-1 · SFT | alignment stability | Hurt |
Later papers that compared with this method
2 later accepted papers report NextStep-1 as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- BitDance: Scaling Autoregressive Generative Models with Binary Tokens (NeurIPS 2026 · better on 2 of 2 shared results)
- Everything in Its Place: Benchmarking Spatial Intelligence of Text-to-Image Models (ICLR 2026 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- Infinity — Infinity: Scaling Bitwise AutoRegressive Modeling for High-Resolution Image Synthesis (CVPR 2025) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- EditAR: Unified Conditional Generation with Autoregressive Models (CVPR 2025 · similarity 0.90)
- BitDance: Scaling Autoregressive Generative Models with Binary Tokens (NeurIPS 2026 · similarity 0.90)
- FlowAR: Scale-wise Autoregressive Image Generation Meets Flow Matching (ICML 2025 · similarity 0.89)
- NEP: Autoregressive Image Editing via Next Editing Token Prediction (NeurIPS 2025 · similarity 0.89)
- Parallelized Autoregressive Visual Generation (CVPR 2025 · similarity 0.89)
- Autoregressive Image Generation without Vector Quantization (NeurIPS 2024 · 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.
