Research map / Image generation / Text-to-image diffusion
JarvisEvo: Towards a Self-Evolving Photo Editing Agent with Synergistic Editor-Evaluator Optimization
Yunlong Lin, Linqing Wang, Kunjie Lin, Zixu Lin, Kaixiong Gong, Wenbo Li, Bin Lin, Zhenxi Li, Shiyi Zhang, Yuyang Peng, Wenxun Dai, Xinghao Ding and 2 more
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 the evaluator optimization loop (ablation) | SEPO · ArtEdit-Bench-Lr (EN and CN, ablation) | O: 7.78 -> 8.54 | Improved |
| add selective loss masking (SLM) in the editor loop (ablation) | SEPO · ArtEdit-Bench-Lr (EN and CN, ablation) | O: 7.40 -> 8.54 | Improved |
| add both evaluator loop and SLM (ablation) | SEPO · ArtEdit-Bench-Lr (EN and CN, ablation) | O: 7.25 -> 8.54 | Improved |
| add the evaluator optimization loop (ablation) | SEPO · ArtEdit-Bench-Lr (EN and CN, ablation) | L1: 12.43 -> 8.25 | Improved |
| add SEPO RL stage after SFT (ablation) | SEPO · ArtEdit-Bench-Lr (EN and CN, ablation) | O: 8.26 -> 8.54 | Improved |
| use intrinsic self-reward instead of Gemini-2.5-Pro external reward (ablation) | SEPO · ArtEdit-Bench-Lr (EN and CN, ablation) | O: 8.14 -> 8.54 | Improved |
| use pairwise preference reward instead of absolute point score reward (ablation) | SEPO · ArtEdit-Bench-Lr (EN and CN, ablation) | O: 8.43 -> 8.54 | Improved |
| use pairwise preference reward instead of absolute point score reward (ablation) | SEPO · ArtEdit-Bench-Lr (EN and CN, ablation) | L1: 8.78 -> 8.25 | Improved |
| add reflection fine-tuning (RFT) after SEPO (ablation) | SEPO · ArtEdit-Bench-Lr (EN and CN, ablation) | O: 8.54 -> 8.76 | Improved |
| add reflection fine-tuning (RFT) after SEPO (ablation) | SEPO · ArtEdit-Bench-Lr (EN and CN, ablation) | L1: 8.25 -> 7.72 | Improved |
| compare JarvisEvo with Nano-Banana (human preference) | JarvisEvo · Human pairwise preference (200 samples) | win rate: 49% vs 28% | Improved |
| JarvisEvo vs Nano-Banana on pixel fidelity | JarvisEvo · ArtEdit-Bench-Lr (English set) | L1: 11.54 -> 7.82 | Improved |
Later papers that compared with this method
1 later accepted paper reports JarvisEvo as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Dotting the Eye: An Intent-Driven Image Retouching Agent for Visual Focus Enhancement (ECCV 2026 · better on 5 of 7 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- JarvisArt: Liberating Human Artistic Creativity via an Intelligent Photo Retouching Agent (NeurIPS 2025 · similarity 0.90)
- RetouchIQ: MLLM Agents for Instruction-Based Image Retouching with Generalist Reward (CVPR 2026 · similarity 0.88)
- CREA: A Collaborative Multi-Agent Framework for Creative Image Editing and Generation (NeurIPS 2025 · similarity 0.86)
- EdiVal-Agent: An Object-Centric Framework for Automated, Fine-Grained Evaluation of Multi-Turn Editing (ICLR 2026 · similarity 0.86)
- FaSTA*: Fast-Slow Toolpath Agent with Subroutine Mining for Efficient Multi-turn Image Editing (ICLR 2026 · similarity 0.85)
- ReasonEdit: Towards Reasoning-Enhanced Image Editing Models (CVPR 2026 · similarity 0.85)
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.
