Research map / Image generation / Text-to-image diffusion
FlowFixer: Towards Detail-Preserving Subject-Driven Generation
Jinyoung Jun, Won-Dong Jang, Wenbin Ouyang, Raghudeep Gadde, Jungbeom Lee
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 |
|---|---|---|---|
| use FlowFixer instead of OminiControl + F-Kontext | SDG refinement · FidelityBench-258K, FLUX.1-Kontext-Pro | AKI: 0.49 -> 66.5 | Improved |
| use FlowFixer instead of text-based editing | SDG refinement · FidelityBench-300 | keypoint matching gain: 45.9% -> 91.2% | Improved |
| use FlowFixer instead of text-based editing | SDG refinement · FidelityBench-300 | VLM preference: 41.3% -> 79.0% | Improved |
| use FlowFixer instead of OminiControl + F-Dev on Nano-Banana-Edit | SDG refinement · FidelityBench-258K, Nano-Banana-Edit | AKI: 108.0 -> 64.7 | Hurt |
| add FlowFixer refinement to SDG outputs | FLUX.1-Kontext-Pro · FidelityBench-258K | CLIP-I: 0.776 -> 0.778 | No clear effect |
| add FlowFixer refinement to Nano-Banana-Edit outputs | Nano-Banana-Edit · FidelityBench-258K | DINO: 0.711 -> 0.711 | No clear effect |
| add training at 0.5x and 0.25x distortion levels (ablation) | FlowFixer · qualitative, large-scale artifacts | robustness | Improved |
| use crop-based refinement instead of full-image pass (ablation) | FlowFixer · qualitative | fine detail recovery | Improved |
| use text prompt in editing refinement | text-based editing · human A/B | human preference: 64.9% vs 64.4% | No clear effect |
Closest papers by meaning
- MorphSIG: Subject-Driven Image Generation via Decoupled Anchoring and Feature Transport (NeurIPS 2026 · similarity 0.86)
- Fine-Tuning Visual Autogressive Models for Subject-Driven Generation (ICCV 2025 · similarity 0.86)
- Large-Scale Text-to-Image Model with Inpainting is a Zero-Shot Subject-Driven Image Generator (CVPR 2025 · similarity 0.85)
- OneActor: Consistent Subject Generation via Cluster-Conditioned Guidance (NeurIPS 2024 · similarity 0.84)
- LLM Blueprint: Enabling Text-to-Image Generation with Complex and Detailed Prompts (ICLR 2024 · similarity 0.84)
- Scaling Rectified Flow Transformers for High-Resolution Image Synthesis (ICML 2024 · similarity 0.84)
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.
