RewardFlow: Generate Images by Optimizing What You Reward
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866910117250727936 |
|---|---|
| author | Susladkar, Onkar Jang, Dong-Hwan Prakash, Tushar Juvekar, Adheesh Shah, Vedant Barik, Ayush Bashir, Nabeel Wahed, Muntasir Shrirao, Ritish Lourentzou, Ismini |
| author_facet | Susladkar, Onkar Jang, Dong-Hwan Prakash, Tushar Juvekar, Adheesh Shah, Vedant Barik, Ayush Bashir, Nabeel Wahed, Muntasir Shrirao, Ritish Lourentzou, Ismini |
| contents | We introduce RewardFlow, an inversion-free framework that steers pretrained diffusion and flow-matching models at inference time through multi-reward Langevin dynamics. RewardFlow unifies complementary differentiable rewards for semantic alignment, perceptual fidelity, localized grounding, object consistency, and human preference, and further introduces a differentiable VQA-based reward that provides fine-grained semantic supervision through language-vision reasoning. To coordinate these heterogeneous objectives, we design a prompt-aware adaptive policy that extracts semantic primitives from the instruction, infers edit intent, and dynamically modulates reward weights and step sizes throughout sampling. Across several image editing and compositional generation benchmarks, RewardFlow delivers state-of-the-art edit fidelity and compositional alignment. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_08536 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | RewardFlow: Generate Images by Optimizing What You Reward Susladkar, Onkar Jang, Dong-Hwan Prakash, Tushar Juvekar, Adheesh Shah, Vedant Barik, Ayush Bashir, Nabeel Wahed, Muntasir Shrirao, Ritish Lourentzou, Ismini Computer Vision and Pattern Recognition Artificial Intelligence We introduce RewardFlow, an inversion-free framework that steers pretrained diffusion and flow-matching models at inference time through multi-reward Langevin dynamics. RewardFlow unifies complementary differentiable rewards for semantic alignment, perceptual fidelity, localized grounding, object consistency, and human preference, and further introduces a differentiable VQA-based reward that provides fine-grained semantic supervision through language-vision reasoning. To coordinate these heterogeneous objectives, we design a prompt-aware adaptive policy that extracts semantic primitives from the instruction, infers edit intent, and dynamically modulates reward weights and step sizes throughout sampling. Across several image editing and compositional generation benchmarks, RewardFlow delivers state-of-the-art edit fidelity and compositional alignment. |
| title | RewardFlow: Generate Images by Optimizing What You Reward |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence |
| url | https://arxiv.org/abs/2604.08536 |