Research map / Image generation / Autoregressive and latent generation
NAMI: Efficient Image Generation via Bridged Progressive Rectified Flow Transformers
Yuhang Ma, Bo Cheng, Shanyuan Liu, Hongyi Zhou, Liebucha Wu, Dawei Leng, Yuhui Yin
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 |
|---|---|---|---|
| split flow by resolution and partition the model into progressive stages | NAMI-2B vs FLUX-based 2B · 1024x1024 inference | inference time (s): 8.47 -> 2.98 | Improved |
| add piecewise flow only (ablation) | NAMI at 1024 · 1024 inference time | inference time reduction: 53.01% reduction | Improved |
| add model partitioning on top of piecewise flow (ablation) | NAMI at 1024 · 1024 inference time | inference time reduction: further 11.81% | Improved |
| use multi-resolution NAMI training instead of FLUX-based training | NAMI-0.6B · 256 res training | FID: 9.76 -> 8.9293 | Improved |
| add model partitioning on top of flow piecewise (ablation) | NAMI-0.6B · 256 and 512 res | FID / CLIP | No clear effect |
| replace Pyramid Flow renoising with BridgeFlow (ablation) | NAMI-0.6B · 256 res | FID: 9.8231 -> 8.9293 | Improved |
| replace BridgeFlow with Upsample + MLP (ablation) | NAMI-0.6B · 256 res | FID: 8.9293 -> 9.0842 | No clear effect |
| replace BridgeFlow with Pixel Shuffle + CNN (ablation) | NAMI-0.6B · 256 res | FID: 8.9293 -> 8.8975 | No clear effect |
| assign too few layers to the low-resolution stage (ablation) | NAMI-0.6B · 512 res | FID / CLIP | Hurt |
| add more layers to low-resolution stage beyond a point (ablation) | NAMI-0.6B · 512 res | FID / CLIP | No clear effect |
| skew time windows toward low or high stages instead of uniform (ablation) | NAMI-0.6B · flow window ratio ablation | FID / CLIP | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- Sana (1.6B) — SANA: Efficient High-Resolution Text-to-Image Synthesis with Linear Diffusion Transformers (ICLR 2025) · better on 1 of 3 shared results
- Infinity (2B) — Infinity: Scaling Bitwise AutoRegressive Modeling for High-Resolution Image Synthesis (CVPR 2025) · better on 1 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
- STARFlow: Scaling Latent Normalizing Flows for High-resolution Image Synthesis (NeurIPS 2025 · similarity 0.88)
- Scaling Rectified Flow Transformers for High-Resolution Image Synthesis (ICML 2024 · similarity 0.87)
- Exploring DCN-like architecture for fast image generation with arbitrary resolution (NeurIPS 2024 · similarity 0.86)
- Native-Resolution Image Synthesis (NeurIPS 2025 · similarity 0.86)
- D^2iT: Dynamic Diffusion Transformer for Accurate Image Generation (CVPR 2025 · similarity 0.86)
- Boosting Flow-based Generative Super-Resolution Models via Learned Prior (CVPR 2024 · similarity 0.86)
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.
