Scientific Image Synthesis: Benchmarking, Methodologies, and Downstream Utility

TL;DR

This paper systematically evaluates scientific image synthesis methods, introduces a logic-driven framework and benchmark, and demonstrates that fine-tuning multimodal models on verified images enhances scientific reasoning capabilities.

Contribution

It presents ImgCoder, a new structured synthesis framework, and SciGenBench, a benchmark for scientific image correctness, advancing the fidelity and utility of synthetic scientific images.

Findings

Pixel-based models have systematic failure modes.

There is a fundamental trade-off between expressiveness and precision.

Fine-tuning LMMs on verified images improves reasoning performance.

Abstract

While synthetic data has proven effective for improving scientific reasoning in the text domain, multimodal reasoning remains constrained by the difficulty of synthesizing scientifically rigorous images. Existing Text-to-Image (T2I) models often produce outputs that are visually plausible yet scientifically incorrect, resulting in a persistent visual-logic divergence that limits their value for downstream reasoning. Motivated by recent advances in next-generation T2I models, we conduct a systematic study of scientific image synthesis across generation paradigms, evaluation, and downstream use. We analyze both direct pixel-based generation and programmatic synthesis, and propose ImgCoder, a logic-driven framework that follows an explicit "understand - plan - code" workflow to improve structural precision. To rigorously assess scientific correctness, we introduce SciGenBench, which evaluates generated images based on information utility and logical validity. Our evaluation reveals systematic failure modes in pixel-based models and highlights a fundamental expressiveness-precision trade-off. Finally, we show that fine-tuning Large Multimodal Models (LMMs) on rigorously verified synthetic scientific images yields consistent reasoning gains, with potential scaling trends analogous to the text domain, validating high-fidelity scientific synthesis as a viable path to unlocking massive multimodal reasoning capabilities.