mlsanityData to 14 Oct 2026

SVGenius: Benchmarking LLMs in SVG Understanding, Editing and Generation

arXiv 2506.03139 ↗First posted 3 Jun 2025cs.CV cs.AIcited by 16

On the map: Multimodal LLMs / vision-language models → Chart understanding and anomaly detection

Abstract (arXiv)

Large Language Models (LLMs) and Multimodal LLMs have shown promising capabilities for SVG processing, yet existing benchmarks suffer from limited real-world coverage, lack of complexity stratification, and fragmented evaluation paradigms. We introduce SVGenius, a comprehensive benchmark comprising 2,377 queries across three progressive dimensions: understanding, editing, and generation. Built on real-world data from 24 application domains with systematic complexity stratification, SVGenius evaluates models through 8 task categories and 18 metrics. We assess 22 mainstream models spanning different scales, architectures, training paradigms, and accessibility levels. Our analysis reveals that while proprietary models significantly outperform open-source counterparts, all models exhibit systematic performance degradation with increasing complexity, indicating fundamental limitations in current approaches; however, reasoning-enhanced training proves more effective than pure scaling for overcoming these limitations, though style transfer remains the most challenging capability across all model types. SVGenius establishes the first systematic evaluation framework for SVG processing, providing crucial insights for developing more capable vector graphics models and advancing automated graphic design applications. Appendix and supplementary materials (including all data and code) are available at https://zju-real.github.io/SVGenius.

Key facts. Cited by 16 papers in the map (data through 14 Oct 2026). Placed in Chart understanding and anomaly detection, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 38 of its 61 references that are in the map. First posted on arXiv 3 Jun 2025.

Builds on

38 of 61 references.

References · 38

  1. 2020Language Models are Few-Shot Learnerscited by 32,419
  2. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 13,028
  3. 2018The Unreasonable Effectiveness of Deep Features as a Perceptual Metriccited by 11,177
  4. 2025Qwen3 Technical Reportcited by 8,820
  5. 2021Evaluating Large Language Models Trained on Codecited by 8,320
  6. 2023DINOv2: Learning Robust Visual Features without Supervisioncited by 7,974
  7. 2022PaLM: Scaling Language Modeling with Pathwayscited by 5,326
  8. 2024Gemini 1.5: Unlocking multimodal understanding across millions of tokens of contextcited by 3,224
  9. 2025InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Modelscited by 1,834
  10. 2024Qwen2.5-Coder Technical Reportcited by 1,203
  11. 2021CodeXGLUE: A Machine Learning Benchmark Dataset for Code Understanding and Generationcited by 744
  12. 2023VisionLLM: Large Language Model is also an Open-Ended Decoder for Vision-Centric Taskscited by 501
26 more
  1. 2017A Neural Representation of Sketch Drawingscited by 313
  2. 2015A note on the evaluation of generative modelscited by 292
  3. 2023Human Preference Score: Better Aligning Text-to-Image Models with Human Preferencecited by 287
  4. 2020DeepSVG: A Hierarchical Generative Network for Vector Graphics Animationcited by 159
  5. 2021Im2Vec: Synthesizing Vector Graphics without Vector Supervisioncited by 116
  6. 2023SVGDreamer: Text Guided SVG Generation with Diffusion Modelcited by 76
  7. 2023IconShop: Text-Guided Vector Icon Synthesis with Autoregressive Transformerscited by 75
  8. 2025OmniSVG: A Unified Scalable Vector Graphics Generation Modelcited by 74
  9. 2024Empowering LLMs to Understand and Generate Complex Vector Graphicscited by 66
  10. 2022CLIPasso: Semantically-Aware Object Sketchingcited by 57
  11. 2019FIGR: Few-shot Image Generation with Reptilecited by 56
  12. 2023DiffSketcher: Text Guided Vector Sketch Synthesis through Latent Diffusion Modelscited by 54
  13. 2021Modern Evolution Strategies for Creativity: Fitting Concrete Images and Abstract Conceptscited by 40
  14. 2024Chat2SVG: Vector Graphics Generation with Large Language Models and Image Diffusion Modelscited by 38
  15. 2024Can Large Language Models Understand Symbolic Graphics Programs?cited by 35
  16. 2022CLIPVG: Text-Guided Image Manipulation Using Differentiable Vector Graphicscited by 34
  17. 2024Text-to-Vector Generation with Neural Path Representationcited by 33
  18. 2020P-Cloth: Interactive Complex Cloth Simulation on Multi-GPU Systems using Dynamic Matrix Assembly and Pipelined Implicit Integratorscited by 26
  19. 2024StrokeNUWA: Tokenizing Strokes for Vector Graphic Synthesiscited by 26
  20. 2024SVGEditBench: A Benchmark Dataset for Quantitative Assessment of LLM's SVG Editing Capabilitiescited by 26
  21. 2021ClipGen: A Deep Generative Model for Clipart Vectorization and Synthesiscited by 23
  22. 2024SuperSVG: Superpixel-based Scalable Vector Graphics Synthesiscited by 19
  23. 2024Exploring the Capability of LLMs in Performing Low-Level Visual Analytic Tasks on SVG Data Visualizationscited by 18
  24. 2024NIVeL: Neural Implicit Vector Layers for Text-to-Vector Generationcited by 17
  25. 2024SVGBuilder: Component-Based Colored SVG Generation with Text-Guided Autoregressive Transformerscited by 14
  26. 2025SVGEditBench V2: A Benchmark for Instruction-based SVG Editingcited by 11

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 3; 2026: 13.

16 papers.

Sort:

By area:

  1. 2025VinciCoder: Unifying Multimodal Code Generation via Coarse-to-fine Visual Reinforcement Learningcited by 21
  2. 2025InternSVG: Towards Unified SVG Tasks with Multimodal Large Language Modelscited by 14
  3. 2026VCG-Bench: Towards A Unified Visual-Centric Benchmark for Structured Generation and Editingcited by 9
  4. 2026VectorGym: A Multi-Task Benchmark for SVG Code Generation, Sketching and Editingcited by 5
  5. 2026LottieGPT: Tokenizing Vector Animation for Autoregressive Generationcited by 5
  6. 2026IntroSVG: Learning from Rendering Feedback for Text-to-SVG Generation via an Introspective Generator-Critic Frameworkcited by 4
  7. 2026Vision2Code: A Multi-Domain Benchmark for Evaluating Image-to-Code Generationcited by 4
  8. 2025RoboSVG: A Unified Framework for Interactive SVG Generation with Multi-modal Guidancecited by 2
  9. 2026ManiBench: A Benchmark for Testing Visual-Logic Drift and Syntactic Hallucinations in Manim Code Generationcited by 2
  10. 2026GeoSVG-RL: Geometry-Aware Reinforcement Learning for Layout-Constrained Text-to-SVG Diagram Generationcited by 2
  11. 2026Structural Evaluation Metrics for SVG Generation via Leave-One-Out Analysiscited by 1
  12. 2026mEOL: Training-Free Instruction-Guided Multimodal Embedder for Vector Graphics and Image Retrievalcited by 1
4 more
  1. 2026GVR-Coder: A Visual-Feedback Framework for Structured SVG Generation in Complex Document and Meeting Scenarioscited by 1
  2. 2026SVGEval: A Vision-Grounded Framework for Perceptual-Quality Benchmarking and Evaluation in Text-to-SVG Generationcited by 1
  3. 2026SVG-Score: Human-Aligned Evaluation of Text-to-SVG Generationcited by 0
  4. 2026MM-SVGEdit: A Multimodal-Driven SVG Editing for UI Designcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025SVGen: Interpretable Vector Graphics Generation with Large Language Models
  2. 2025SVGThinker: Instruction-Aligned and Reasoning-Driven Text-to-SVG Generation
  3. 2024VGBench: Evaluating Large Language Models on Vector Graphics Understanding and Generation
  4. 2025VectorEdits: A Dataset and Benchmark for Instruction-Based Editing of Vector Graphics
  5. 2025UniSVG: A Unified Dataset for Vector Graphic Understanding and Generation with Multimodal Large Language Models
  6. 2025SVGauge: Towards Human-Aligned Evaluation for SVG Generation
  7. 2023Beyond Pixels: Exploring Human-Readable SVG Generation for Simple Images with Vision Language Models
  8. 2023Leveraging Large Language Models for Scalable Vector Graphics-Driven Image Understanding