AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

LongBanana: An Expert-Verified Benchmark for Long-Context Multi-Reference Image Synthesis

Haoxiang Cao, Yuxuan Zhang, Penghui Du, Bo Li, Jiajiong Cao, Yingying Fan, Zimeng Wu, Yifu Luo, Mingzhe Zheng, Junwen Miao, Huaisong Zhang, YANG YAO and 7 more

NeurIPS 2026 · Image generation › Autoregressive and latent generation · approach: Compositional scene generation · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Introduces LongBanana, an expert-designed and expert-verified benchmark for long-context multi-reference image synthesis under agent-controlled image generation.

Problem

Evaluating image synthesis that must retrieve dispersed visual evidence from long multi-reference contexts, reject distractors, and bind entities and relations into one coherent image.

Approach

A benchmark of expert-designed, expert-verified tasks for long-context multi-reference image synthesis in the setting of agent-controlled image generation.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
decompose prompt with privileged graph (PD-Direct) instead of full-prompt direct generation (ablation)Nano Banana 2 · LongBananaoverall average: 68.4 -> 74.6Improved
use OpenClaw + RPR agent instead of PD-DirectNano Banana 2 · LongBananaP@70%: 35.6 -> 47.8Improved
add RPR workflow to OpenClaw harness (ablation)OpenClaw · LongBananaoverall average: 76.1 -> 78.4Improved
add RPR workflow to OpenCode harness (ablation)OpenCode · LongBananaP@70%: 47.8 -> 43.3Hurt
add RPR workflow to Hermes-Agent harness (ablation)Hermes-Agent · LongBananaP@70%: 43.3 -> 35.6Hurt
add RPR workflow to Hermes-Agent harness (ablation)Hermes-Agent · LongBananaoverall average: 73.9 -> 75.6Improved
add RPR to OpenClaw on temporal/state reasoning (ablation)OpenClaw · LongBanana TSRchecklist average: 78.7 -> 73.6Hurt
ReAct only, Plan-Exec only, Reflect only vs full RPR (ablation)OpenClaw · LongBananaoverall average and pass rateImproved
use Nano Banana 2 instead of Seedream 4 or GPT Image 1 backendOpenClaw + RPR · LongBananaoverall average: 63.6 -> 78.4Improved
use agentic checklist judge instead of single-pass VLM-as-judgeGemini 3 Flash judge · audit pool of 900 outputsPearson correlation with human: 0.8413 -> 0.8707Improved
use agentic checklist judge instead of VLM-as-judgeGemini 3 Flash judge · item levelopposite-end 0<->2 errors: 44.09% -> 24.81%Improved

Experimental setup

Benchmarks: LongBanana

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy