VQ-VA World: Towards High-Quality Visual Question-Visual Answering

TL;DR

This paper introduces VQ-VA World, a data-centric framework for training models to generate images in response to visual questions, significantly improving open-source VQ-VA performance and providing a new benchmark for evaluation.

Contribution

It presents a large-scale, targeted data construction pipeline and a human-curated benchmark, advancing open-source VQ-VA capabilities and evaluation methods.

Findings

Training with VQ-VA World data improves LightFusion's score to 53.06 on IntelligentBench.

The approach surpasses prior open-source baselines by a large margin.

Results narrow the gap toward proprietary systems like NanoBanana and GPT-Image.

Abstract

This paper studies Visual Question-Visual Answering (VQ-VA): generating an image, rather than text, in response to a visual question -- an ability that has recently emerged in proprietary systems such as NanoBanana and GPT-Image. To also bring this capability to open-source models, we introduce VQ-VA World, a data-centric framework built around an agentic pipeline for large-scale, targeted data construction. Leveraging web-scale deployment, this pipeline crawls a massive amount of ~1.8M high-quality, interleaved image-text samples for model training. For evaluation, we further release IntelligentBench, a human-curated benchmark that systematically assesses VQ-VA along the aspects of world knowledge, design knowledge, and reasoning. Training with VQ-VA World data yields strong empirical gains: it helps LightFusion attain 53.06 on IntelligentBench, substantially surpassing the best prior open-source baselines (i.e., 7.78 from vanilla LightFusion; 1.94 from UniWorld-V1), and significantly narrowing the gap toward leading proprietary systems (e.g., 81.67 from NanoBanana; 82.64 from GPT-Image). By releasing the full suite of model weights, datasets, and pipelines, we hope to stimulate future research on VQ-VA.