Graphic-Design-Bench: A Comprehensive Benchmark for Evaluating AI on Graphic Design Tasks
Paper • 2604.04192 • Published
sample_id stringlengths 20 20 | benchmark_id stringclasses 1
value | domain stringclasses 1
value | task_type stringclasses 1
value | benchmark_name stringclasses 1
value | prompt large_stringclasses 1
value | ground_truth large_stringlengths 873 65.7k | image stringclasses 1
value | media_path stringclasses 1
value | media_type stringclasses 1
value | metadata large_stringlengths 5.19k 248k |
|---|---|---|---|---|---|---|---|---|---|---|
style_completion_000 | template-4 | template | generation | Style Completion | {"components": [{"type": "GROUP", "left": "128.976px", "top": "416.015px", "width": "822.04665012px", "height": "247.969893px", "transform": "none", "background": "rgb(250, 250, 249)", "backgroundColor": "rgb(250, 250, 249)", "clipPath": "path(\"M5.702473848373801,5.702473848373801L2324.8197387405635,5.702473848373801L... | none | {"template_id": "010cc749-ae55-4eee-b748-cd1036ccb2f0", "context_layouts": [{"components": [{"type": "GROUP", "id": "0-0-0-0", "left": "576.8221px", "top": "149.64479999999998px", "width": "60.998144px", "height": "150.836258138px", "transform": "rotate(42.1613deg)", "background": "rgb(1, 1, 1)", "backgroundColor": "rg... | |||
style_completion_001 | template-4 | template | generation | Style Completion | "{\"components\": [{\"type\": \"GROUP\", \"left\": \"576.822px\", \"top\": \"149.645px\", \"width\":(...TRUNCATED) | none | "{\"template_id\": \"010cc749-ae55-4eee-b748-cd1036ccb2f0\", \"context_layouts\": [{\"components\": (...TRUNCATED) | |||
style_completion_002 | template-4 | template | generation | Style Completion | "{\"components\": [{\"type\": \"IMAGE\", \"src\": \"https://storage.googleapis.com/lica-video/471e69(...TRUNCATED) | none | "{\"template_id\": \"010f7dcf-7362-484d-ab21-e22230a75cd7\", \"context_layouts\": [{\"components\": (...TRUNCATED) | |||
style_completion_003 | template-4 | template | generation | Style Completion | "{\"components\": [{\"type\": \"IMAGE\", \"src\": \"https://storage.googleapis.com/lica-video/f1b3c7(...TRUNCATED) | none | "{\"template_id\": \"010f7dcf-7362-484d-ab21-e22230a75cd7\", \"context_layouts\": [{\"components\": (...TRUNCATED) | |||
style_completion_004 | template-4 | template | generation | Style Completion | "{\"components\": [{\"type\": \"GROUP\", \"left\": \"109.333px\", \"top\": \"186.313px\", \"width\":(...TRUNCATED) | none | "{\"template_id\": \"02ce2d2c-0fe8-42d5-b9a1-7e5356698a35\", \"context_layouts\": [{\"components\": (...TRUNCATED) | |||
style_completion_005 | template-4 | template | generation | Style Completion | "{\"components\": [{\"type\": \"GROUP\", \"id\": \"0-0-0\", \"left\": \"109.333px\", \"top\": \"181.(...TRUNCATED) | none | "{\"template_id\": \"02ce2d2c-0fe8-42d5-b9a1-7e5356698a35\", \"context_layouts\": [{\"components\": (...TRUNCATED) | |||
style_completion_006 | template-4 | template | generation | Style Completion | "{\"components\": [{\"type\": \"IMAGE\", \"src\": \"https://storage.googleapis.com/lica-video/d24190(...TRUNCATED) | none | "{\"template_id\": \"06d385b9-57f3-4ee2-ba52-44080ef8b5f3\", \"context_layouts\": [{\"components\": (...TRUNCATED) | |||
style_completion_007 | template-4 | template | generation | Style Completion | "{\"components\": [{\"type\": \"IMAGE\", \"src\": \"https://storage.googleapis.com/lica-video/f04e8a(...TRUNCATED) | none | "{\"template_id\": \"06d385b9-57f3-4ee2-ba52-44080ef8b5f3\", \"context_layouts\": [{\"components\": (...TRUNCATED) | |||
style_completion_008 | template-4 | template | generation | Style Completion | "{\"components\": [{\"type\": \"GROUP\", \"left\": \"5.68434e-13px\", \"width\": \"1920.00167171px\"(...TRUNCATED) | none | "{\"template_id\": \"08c5af24-935b-4864-806d-f34563decb10\", \"context_layouts\": [{\"components\": (...TRUNCATED) | |||
style_completion_009 | template-4 | template | generation | Style Completion | "{\"components\": [{\"type\": \"GROUP\", \"left\": \"5.68434e-13px\", \"width\": \"1920.00167171px\"(...TRUNCATED) | none | "{\"template_id\": \"08c5af24-935b-4864-806d-f34563decb10\", \"context_layouts\": [{\"components\": (...TRUNCATED) |
39 benchmarks for evaluating vision-language models on graphic design tasks — layout, typography, SVG, template matching, animation. Built on 1,148 real design layouts from the Lica dataset.
Paper: arXiv:2604.04192 | Code: github.com/lica-world/GDB | Blog: lica.world
from datasets import load_dataset
ds = load_dataset("lica-world/GDB", "svg-1")
| Field | Type | Description |
|---|---|---|
sample_id |
string | Sample identifier |
benchmark_id |
string | e.g. svg-1, typography-3 |
domain |
string | layout, typography, svg, template, temporal, category, lottie |
task_type |
string | understanding or generation |
prompt |
string | Evaluation prompt |
ground_truth |
string | Expected answer (JSON for complex types) |
image |
Image | Input image (when applicable) |
metadata |
string | Task-specific fields as JSON |
pip install git+https://github.com/lica-world/GDB.git
from gdb.registry import BenchmarkRegistry
registry = BenchmarkRegistry()
registry.discover()
bench = registry.get("svg-1")
scores = bench.evaluate(predictions, ground_truth)
@article{gdb2026,
title={GDB: A Real-World Benchmark for Graphic Design},
author={Deganutti, Adrienne and Hirsch, Elad and Zhu, Haonan and Seol, Jaejung and Mehta, Purvanshi},
journal={arXiv preprint arXiv:2604.04192},
year={2026}
}
Apache 2.0