# Open-Source Image Editing Models Are Zero-Shot Vision Learners

arXiv: https://arxiv.org/abs/2605.04566 · HTML: https://mlsanity.com/paper/2605.04566/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Builds on 8 of its 12 references that are in the map.
- First posted on arXiv 6 May 2026.

## Builds on (most cited first, 8 of 8)

- [The Cityscapes Dataset for Semantic Urban Scene Understanding](https://mlsanity.com/paper/1604.01685/) (2016)
- [Repurposing Diffusion-Based Image Generators for Monocular Depth Estimation](https://mlsanity.com/paper/2312.02145/) (2023)
- [DIODE: A Dense Indoor and Outdoor DEpth Dataset](https://mlsanity.com/paper/1908.00463/) (2019)
- [Video models are zero-shot learners and reasoners](https://mlsanity.com/paper/2509.20328/) (2025)
- [Lotus: Diffusion-based Visual Foundation Model for High-quality Dense Prediction](https://mlsanity.com/paper/2409.18124/) (2024)
- [LongCat-Image Technical Report](https://mlsanity.com/paper/2512.07584/) (2025)
- [FireRed-Image-Edit-1.0 Technical Report](https://mlsanity.com/paper/2602.13344/) (2026)
- [Is Nano Banana Pro a Low-Level Vision All-Rounder? A Comprehensive Evaluation on 14 Tasks and 40 Datasets](https://mlsanity.com/paper/2512.15110/) (2025)

## Built on by (most cited first, 3 of 3)

- [Marigold V2: Revisiting Diffusion Transformers for Monocular Depth Estimation](https://mlsanity.com/paper/2609.08084/) (2026)
- [The Camera Inside the Editor: Reading the Implicit Camera of Image Editors with Painted Calibration Patterns](https://mlsanity.com/paper/2609.37732/) (2026)
- [Let RGB Be the Language of Vision](https://mlsanity.com/paper/2607.12450/) (2026)

## Abstract (arXiv)

Recent studies have shown that large generative models can solve vision tasks they were not explicitly trained for. However, existing evidence relies on closed-source models~(Veo~3, Nano Banana Pro) or requires task-specific instruction tuning, leaving open whether publicly available image-editing models possess zero-shot vision abilities out of the box. We conduct a systematic evaluation of three open-source image-editing models -- Qwen-Image-Edit, FireRed-Image-Edit, and LongCat-Image-Edit -- on dense visual prediction tasks without any fine-tuning. We benchmark monocular depth estimation on NYUv2 and DIODE, surface normal estimation on NYUv2, and semantic segmentation on Cityscapes, covering both geometric and semantic scene understanding. Results show that open-source image-editing models exhibit non-trivial zero-shot visual understanding. On NYUv2 surface normals, FireRed-Image-Edit achieves a mean angular error of 17.69^circ, surpassing the fine-tuned Marigold (20.86^circ) and matching the instruction-tuned Vision Banana (17.78^circ) without any task-specific training. On NYUv2 depth estimation, LongCat-Image-Edit obtains δ1=0.822 with affine alignment, and Qwen-Image-Edit leads on DIODE Indoor (δ1=0.868). On Cityscapes semantic segmentation, Qwen-Image-Edit reaches 25.7 mIoU at the 19-class level and 49.5 mIoU at a coarser 7-category level. By comparing three independently trained editors, we test whether zero-shot vision ability is an emergent property of image-editing pretraining rather than a model-specific artifact. Code, evaluation scripts, and all results are publicly released to serve as a reproducible baseline for future work.