cs.CVFeb 24, 2026

FlowFixer: Towards Detail-Preserving Subject-Driven Generation

Authors: Jinyoung Jun, Won-Dong Jang, Wenbin Ouyang, Raghudeep Gadde, Jungbeom Lee

Organizations: Amazon · Korea University

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

We present FlowFixer, a refinement framework for subject-driven generation (SDG) that restores fine details lost during generation caused by changes in scale and perspective of a subject. FlowFixer proposes direct image-to-image translation from visual references, avoiding ambiguities in language prompts. To enable image-to-image training, we introduce a one-step denoising scheme to generate self-supervised training data, which automatically removes high-frequency details while preserving global structure, effectively simulating real-world SDG errors. We further propose a keypoint matching-based metric to properly assess fidelity in details beyond semantic similarities usually measured by CLIP or DINO. Experimental results demonstrate that FlowFixer outperforms state-of-the-art SDG methods in both qualitative and quantitative evaluations, setting a new benchmark for high-fidelity subject-driven generation.

Figures & tables

Figure 2: FlowFixer overview. FlowFixer enhances SDG images by restoring fine subject details, using the original subject image as reference.
Figure 3: FlowFixer inference pipeline. The model takes two conditional inputs: reference subject image Iref{\mathbf{I}}_{\text{ref}} and the generated image Igen{\mathbf{I}}_{\text{gen}} from any SDG model. Then the model produces a refined result I^gen\widehat{{\mathbf{I}}}_{\text{gen}} that preserves global layout. For faster inference, we optionally refine only a subject-centric crop of Igen{\mathbf{I}}_{\text{gen}} and blend it back using Poisson image blending.
Figure 4: Example of one-step denoising distortions. For each distortion level, pixel-wise variance maps are computed over 10 degraded samples. Insets show example outputs, with distortions concentrated in high-frequency regions.
MethodFLUX.1-Kontext-ProQwen-Image-EditNano-Banana-Edit
AKI ↑\uparrowKGain\mathcal{K}_{\text{Gain}} ↑\uparrowCLIP-I ↑\uparrowDINO ↑\uparrowAKI ↑\uparrowKGain\mathcal{K}_{\text{Gain}} ↑\uparrowCLIP-I ↑\uparrowDINO ↑\uparrowAKI ↑\uparrowKGain\mathcal{K}_{\text{Gain}} ↑\uparrowCLIP-I ↑\uparrowDINO ↑\uparrow
Baseline--0.7760.663--0.7770.668--0.7960.711
Text-based editing [ 19 ]7.552.7%0.7630.64711.154.1%0.7620.64761.777.0%0.7820.691
OminiControl [ 39 ] + F-Dev29.053.9%0.7240.5510.4843.8%0.7240.552108.070.7%0.7470.605
OminiControl [ 39 ] + F-Kontext0.4945.7%0.7660.649-3.3746.8%0.7650.64953.056.6%0.7860.696
FlowFixer (ours)66.577.9%0.7780.66854.074.8%0.7770.66864.779.2%0.7960.711
Table 1: Refinement performance on the FidelityBench-258K. For all metrics, higher numbers indicate better performance.
Figure 5: Qualitative comparison on Subject fidelity refinement on the FidelityBench-258K dataset. The insets in the full images show the reference subject images and the red and green boxes indicate the zoomed-in regions. The regions for zoomed-in views are found on the SDG baseline images and cropped the same area for all methods.
MethodAKI ↑\uparrowKGain\mathcal{K}_{\text{Gain}} ↑\uparrowVLM ↑\uparrow
Text-based editing [ 19 ]1.8745.9%41.3%
OminiControl [ 39 ] + F-Dev22.746.6%4.2%
OminiControl [ 39 ] + F-Kontext11.138.4%25.2%
FlowFixer (ours)67.391.2%79.0%
Table 2: Refinement performance compared to original SDG images on the FidelityBench-300. For all metrics, higher numbers indicate better performance.
Figure 6: Scatter plots of the number of keypoint matches on FidelityBench-300. Each dot represents a sample; points above the red dashed line indicate an increase in keypoint matches (positive AKI, green region), suggesting improved structural alignment and subject fidelity. Samples below the line show decreased correspondence after refinement.
Figure 7: A/B study results comparing the FlowFixer against four alternatives. FlowFixer is consistently preferred by human raters.
Figure 8: Impact of training distortion levels on refinement performance. Using a range of distortion levels during training enhances the model’s ability to handle diverse degradation at inference time, resulting in more robust restoration.
Figure 9: Efficacy of cropped refinement in comparison with full image refinement. While full image refinement moderately enhances subject fidelity, cropping further improves legibility.
More in this topic

Subject-Driven Image Generation

Subject-driven image generation uses visual references to create new images that preserve a chosen subject while following prompts.

See all 58 papers

Explore similar work

CardsList

  1. Decomposing Subject-Driven Image Generation via Intermediate Structural Prediction

    May 20, 2026Hanzhong Guo, Yizhou YuGeometry-Conditioned Image GenerationSubject-Driven Image Generation

    School of Computing and Data Science, The University of Hong Kong

  2. CopyCat: Improving Fine-Grained Subject Consistency in Subject-to-Image Models within Seconds

    Aug 1, 2026Peng Zheng, Ruiqi Liu, Rui Ma +1Diffusion TransformerSubject-Driven Image Generation

    School of Artificial Intelligence, Jilin University · 2Shanghai Innovation Institute · 4Engineering Research Center of Knowledge-Driven Human-Machine Intelligence, MOE, China +1

  3. Diagnose-to-Generate: Eliminating Structural Distortions in Flow Models via Patch-Level Grounded Alignment

    Oct 13, 2026Yifan Li, Lishuai Gao, Tong Zhang +8Image GenerationFlow-Based Generative Modeling

    University of Science and Technology of China, Hefei, AnHui Province, China · Meituan Inc., Beijing, China