Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..

HiFlow: Training-free High-Resolution Image Generation with Flow-Aligned Guidance

Authors: Jiazi Bu, Pengyang Ling, Yujie Zhou, Pan Zhang, Tong Wu, Xiaoyi Dong, Yuhang Zang, Yuhang Cao, Dahua Lin, Jiaqi Wang

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments validate Hi Flow s capability in achieving superior high-resolution image quality over state-of-the-art methods. Our code is available at Hi Flow Repo.
Researcher Affiliation Collaboration Jiazi Bu1,5 Pengyang Ling2,5 Yujie Zhou1,5 Pan Zhang5 Tong Wu4 Xiaoyi Dong3,5 Yuhang Zang5 Yuhang Cao5 Dahua Lin3,5,7 Jiaqi Wang5,6 1Shanghai Jiao Tong University 2University of Science and Technology of China 3The Chinese University of Hong Kong 4Stanford University 5Shanghai AI Laboratory 6Shanghai Innovation Institute 7CPII under Inno HK
Pseudocode Yes F Pseudo code of flow-aligned guidance In this section, we present the pseudo code for each component of the proposed flow-aligned guidance, as detailed in Algorithm 1.
Open Source Code Yes Our code is available at Hi Flow Repo. Justification: We have included the source code of Hi Flow in the supplementary material. We will make the code publicly available after review.
Open Datasets Yes FID is calculated between generated images and 10K real high-quality images (with at least 1024 1024 resolution) sourced from LAION-High-Resolution [48].
Dataset Splits No The proposed Hi Flow is a training-free method that requires no training or fine-tuning; thus, there are no training details. For evaluation: We collect 1K high-quality captions across various scenarios for diverse image generation. The CLIP [43] score is used to assess the prompt-following capability, Frechet Inception Distance [18] (FID) and Inception Score [47] (IS) are reported to measure image quality, in which FID is calculated between generated images and 10K real high-quality images (with at least 1024 1024 resolution) sourced from LAION-High-Resolution [48].
Hardware Specification Yes All experiments are conducted on a single NVIDIA A100 GPU.
Software Dependencies No If not specified, the generated images are based on Flux.1.0-dev [28], an advanced open-sourced Rectified Flow model based on Di T architecture. The paper also mentions using the 4-bit version of Flux (quantized by SVDQuant [31]). However, it does not list multiple key software components with their specific version numbers (e.g., Python, PyTorch, CUDA versions).
Experiment Setup Yes The sampling steps are set as 30, the noise-adding ratio τ in initialization alignment is set as [0.6, 0.3, 0.3] for 1K 2K 3K 4K cascade generation, and the normalized cutoff frequency is set as D = 0.4. The guidance weights in direction/acceleration alignment are set as αt = βt = t/τ for gradually weakening control.