TerraVis: Towards Evaluation of World-Grounded Visual Consistency in Text-to-Image Generation via MLLM Workflows

Hugging Face Daily Papers Papers
text-to-image evaluation visual-consistency mllm benchmark world-knowledge

Summary

TerraVis 提出了一个基于 MLLM 工作流的框架，用于评估文生图模型生成图像的世界一致性（如畸形结构、物理上不可能的交互、空间关系不一致等），并定义了涵盖对象、交互和场景层级的 18 类一致性违规分类体系。实验表明该框架与人类判断相关性最强，且常规指标表现优秀的模型仍可能存在大量世界一致性缺陷。

Recent text-to-image models have made substantial progress in photorealism, aesthetics, and text-image alignment. Yet visually appealing images can still violate real-world plausibility, exhibiting malformed object structures, impossible anatomy, physically implausible interactions, or inconsistent spatial relationships. Such failures are not well captured by existing fidelity, aesthetics, preference, or alignment metrics. To address this gap, we introduce TerraVis, a framework for evaluating world-grounded visual consistency in generated images. TerraVis defines a structured taxonomy of world-consistency violations spanning object-, interaction-, and scene-level failures, and employs a multi-stage evaluation framework to identify and quantify them. Given an image, TerraVis first uses an MLLM to assess its eligibility for evaluation, then detects violations across 18 taxonomy-defined types and classifies them as minor or major to derive an overall world-consistency score. Across diverse open-source and proprietary text-to-image models on two widely used benchmarks, TerraVis achieves the strongest correlation with human judgments of world consistency among existing metrics. Our benchmark results further show that models that achieve strong performance on conventional metrics can still exhibit substantial world-consistency failures. These findings highlight world consistency as a complementary evaluation dimension and demonstrate that TerraVis enables systematic quantification, diagnosis, and comparison of such failures. Our code is publicly available at https://github.com/ShyFoo/TerraVis.
Original Article
View Cached Full Text

Cached at: 10/09/26, 06:24 AM

Paper page - TerraVis: Towards Evaluation of World-Grounded Visual Consistency in Text-to-Image Generation via MLLM Workflows

Source: https://huggingface.co/papers/2610.02959

Abstract

Recenttext-to-imagemodelshavemadesubstantialprogressinphotorealism,aesthetics,andtext-imagealignment.Yetvisuallyappealingimagescanstillviolatereal-worldplausibility,exhibitingmalformedobjectstructures,impossibleanatomy,physicallyimplausibleinteractions,orinconsistentspatialrelationships.Suchfailuresarenotwellcapturedbyexistingfidelity,aesthetics,preference,oralignmentmetrics.Toaddressthisgap,weintroduceTerraVis,aframeworkforevaluatingworld-groundedvisualconsistencyingeneratedimages.TerraVisdefinesastructuredtaxonomyofworld-consistencyviolationsspanningobject-,interaction-,andscene-levelfailures,andemploysamulti-stageevaluationframeworktoidentifyandquantifythem.Givenanimage,TerraVisfirstusesanMLLMtoassessitseligibilityforevaluation,thendetectsviolationsacross18taxonomy-definedtypesandclassifiesthemasminorormajortoderiveanoverallworld-consistencyscore.Acrossdiverseopen-sourceandproprietarytext-to-imagemodelsontwowidelyusedbenchmarks,TerraVisachievesthestrongestcorrelationwithhumanjudgmentsofworldconsistencyamongexistingmetrics.Ourbenchmarkresultsfurthershowthatmodelsthatachievestrongperformanceonconventionalmetricscanstillexhibitsubstantialworld-consistencyfailures.ThesefindingshighlightworldconsistencyasacomplementaryevaluationdimensionanddemonstratethatTerraVisenablessystematicquantification,diagnosis,andcomparisonofsuchfailures.Ourcodeispubliclyavailableathttps://github.com/ShyFoo/TerraVis.

View arXiv pageView PDFProject pageGitHub1Add to collection

Get this paper in your agent:

hf papers read 2610\.02959

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2610.02959 in a model README.md to link it from this page.

Datasets citing this paper1

#### ShyFoo/TerraVis-Annotations Viewer• Updatedabout 3 hours ago • 4.5k • 19

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2610.02959 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Words to World: Benchmarking LLMs in Interactive Physical Environments

Hugging Face Daily Papers

该论文提出了一个用于评估大语言模型在闭环物理交互中表现的基准，模型需通过连接机械臂、摄像头和传感器的系统在真实环境中执行任务并根据物理反馈调整行动。实验揭示了LLM在物理行动中的常见失败模式，如幻觉状态变化、糟糕的异常恢复和不稳定的长程规划。

Orbia: Do Generated Videos Form a Consistent 3D World?

Hugging Face Daily Papers

Orbia 提出了首个系统评估交互式视频世界模型 3D 一致性的基准，通过受控探索与回访协议，从输入保持、生成内容持久性和 3D 自洽性三个维度对 27 个模型进行评测。

LMBuild: Evaluating LLM Agents for Generating Buildable and Functional Structures

Hugging Face Daily Papers

该论文提出 LMBuild，一个用于评估 LLM 智能体生成可建造、功能性结构能力的基准，包含交互式装配环境、基于 CAD 数据集构建的评测集以及涵盖结构可靠性、功能可供性和物理可实现性的评估框架，在 30 个系统上的评测揭示了功能推理仍是前沿模型的主要瓶颈。