Can Nano Banana 2 Replace Traditional Image Restoration Models? An Evaluation of Its Performance on Image Restoration Tasks

arXiv cs.CV / 4/6/2026

💬 OpinionSignals & Early TrendsModels & Research

共有:

Key Points

The paper reports a systematic evaluation of Nano Banana 2 as a general-purpose generative image editing model for multiple image restoration tasks across diverse scenes and degradation types.
It finds that prompt design is pivotal, with concise prompts that include explicit fidelity constraints delivering the best balance between reconstruction accuracy and perceptual quality.
Compared with state-of-the-art restoration models, Nano Banana 2 achieves stronger performance on full-reference metrics while remaining competitive on perceptual quality, supported by both experiments and user studies.
The model shows strong generalization in difficult cases such as small faces, dense crowds, and severe degradations, indicating potential for broader “unified solver” use.
Despite its promise, Nano Banana 2 is sensitive to how prompts are formulated and may need iterative prompt refinement to achieve optimal results; the authors publish test results on GitHub.

Abstract

Recent advances in generative AI raise the question of whether general-purpose image editing models can serve as unified solutions for image restoration. In this work, we conduct a systematic evaluation of Nano Banana 2 for image restoration across diverse scenes and degradation types. Our results show that prompt design plays a critical role, where concise prompts with explicit fidelity constraints achieve the best trade-off between reconstruction accuracy and perceptual quality. Compared with state-of-the-art restoration models, Nano Banana 2 achieves superior performance in full-reference metrics while remaining competitive in perceptual quality, which is further supported by user studies. We also observe strong generalization in challenging scenarios, such as small faces, dense crowds, and severe degradations. However, the model remains sensitive to prompt formulation and may require iterative refinement for optimal results. Overall, our findings suggest that general-purpose generative models hold strong potential as unified image restoration solvers, while highlighting the importance of controllability and robustness. All test results are available on https://github.com/yxyuanxiao/NanoBanana2TestOnIR.