| title | LongBanana: An Expert-Verified Benchmark for Long-Context Multi-Reference Image Synthesis |
|---|---|
| authors | Haoxiang Cao, Yuxuan Zhang, Penghui Du, Bo Li, Jiajiong Cao, Yingying Fan, Zimeng Wu, Yifu Luo, Mingzhe Zheng, Junwen Miao, Huaisong Zhang, Yang Yao, Yu Wang, Dongfu Jiang, Ping Nie, Wenhu Chen, Chaoqun Wang, Changqian Yu, Kelsey R. Allen |
| date | 2026-11-04 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2611.06856 |
| affiliation | Kolors Team, Kuaishou Technology |
| upvotes | 31 (Hugging Face Daily Papers, 2026-11-11) |
| links | abs pdf hf code |
| cite | |
| related | StableVQ: Practical Guidelines for Stable Vector-Quantized Tokenizer Training MetaView: Monocular Novel View Synthesis with Scale-Aware Implicit Geometry Priors AdvFD: Boosting Visual Generation via Adversarial Fr'echet Distance Loss Where, What, Why, and Importance: Structured Defect Grounding for Text-to-Image Feedback |
| abstract | We introduce LongBanana, to our knowledge the first expert-designed and expert-verified benchmark for long-context multi-reference image synthesis in the age of agent-driven image generation. The task requires retrieving dispersed visual evidence, rejecting distractors, binding entities and relations, and composing one coherent final image from multiple references. LongBanana contains 90 carefully designed tasks organized by a 3-by-3 taxonomy; each task includes about 32 reference images, 20 constraints, and 50 evaluation items, yielding 4,670 expert-specified atomic evaluation items in total. Its graph-grounded evaluation design instantiates each task with reference images and an entity-relation graph, then converts node, edge, and whole-graph obligations into checklist items for Content Fidelity, Relational Binding, and Global Constraint Satisfaction. Reviewed expert labels include a 15% double-labeled subset with 89.4% exact agreement and 0.83 quadratic weighted kappa. We further introduce LongBanana-Gen, an agent-harness framework for long-horizon image generation; experiments show that minimal agent harnesses improve over direct generation, while a ReAct-Plan-Exec-Reflect (RPR) workflow yields harness-dependent gains by strengthening evidence grounding and repair. Together, LongBanana provides an auditable testbed for measuring dense visual evidence use and a shared target for developing stronger agent-driven image-generation systems. Code and data are available at the project page: https://long-banana.com/. |
bibtex
@misc{cao2026longbanana,
title={LongBanana: An Expert-Verified Benchmark for Long-Context Multi-Reference Image Synthesis},
author={Haoxiang Cao and Yuxuan Zhang and Penghui Du and Bo Li and Jiajiong Cao and Yingying Fan and Zimeng Wu and Yifu Luo and Mingzhe Zheng and Junwen Miao and Huaisong Zhang and Yang Yao and Yu Wang and Dongfu Jiang and Ping Nie and Wenhu Chen and Chaoqun Wang and Changqian Yu and Kelsey R. Allen},
year={2026},
eprint={2611.06856},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2611.06856}
}