Research map / Image generation / Fast diffusion sampling
StreamFlow: Theory, Algorithm, and Implementation for High-Efficiency Rectified Flow Generation
Sen Fang, Hongbin Zhong, Yalin Feng, Yanxin Zhang, Dimitris N. Metaxas
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 xformer architecture and taesd decoder (ablation) | Vanilla pipeline · Ablation, FPS | FPS: 1.87 -> 2.3 | Improved |
| add VFB (batched velocity field) with taesd (ablation) | xformer+taesd · Ablation, FPS | FPS: 2.3 -> 8.0 | Improved |
| add TensorRT compilation (ablation) | xformer+taesd+VFB · Ablation, FPS | FPS: 8.0 -> 10.05 | Improved |
| add compatibility layer (ablation) | tensorrt+taesd+VFB · Ablation, FPS | FPS: 10.05 -> 9.8 | Hurt |
| use 8 inference steps instead of 4 (ablation) | Optimized pipeline · Ablation, FPS | FPS: 4.5 FPS vs 2.22 FPS 4-step baseline | Hurt |
| use taesd instead of original VAE (ablation) | xformer+VFB · Quality study, text-to-image | CLIP score: 96.74 -> 97.26 | Improved |
| use taesd instead of original VAE (ablation) | xformer+VFB · Quality study, text-to-image | FID: 31.82 -> 30.75 | Improved |
| use taesd instead of original VAE (ablation) | tensorrt+VFB · Quality study, text-to-image | CLIP score: 96.08 -> 96.25 | Improved |
| use taesd instead of original VAE (ablation) | tensorrt+VFB · Quality study, text-to-image | FID: 31.16 -> 31.28 | No clear effect |
| apply StreamFlow acceleration (xformer+taesd+VFB) | Baseline Vanilla · Quality study, text-to-image | FID: 29.44 -> 30.75 | No clear effect |
| apply tensorrt+taesd+VFB | Baseline Vanilla · Power/Memory comparison, image generation | power (W): 205.28 -> 110.51 | Improved |
| apply tensorrt+taesd+VFB | Baseline Vanilla · Power/Memory comparison, image generation | FPS: 1.87 -> 10.05 | Improved |
Closest papers by meaning
- FlowTurbo: Towards Real-time Flow-Based Image Generation with Velocity Refiner (NeurIPS 2024 · similarity 0.88)
- Align Your Flow: Scaling Continuous-Time Flow Map Distillation (NeurIPS 2025 · similarity 0.88)
- Rectified Diffusion: Straightness Is Not Your Need in Rectified Flow (ICLR 2025 · similarity 0.87)
- PeRFlow: Piecewise Rectified Flow as Universal Plug-and-Play Accelerator (NeurIPS 2024 · similarity 0.87)
- StreamFlow: Streamlined Multi-Frame Optical Flow Estimation for Video Sequences (NeurIPS 2024 · similarity 0.87)
- Mean Flows for One-step Generative Modeling (NeurIPS 2025 · similarity 0.87)
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.
