ximinng

ximinng / HiVG

Public

Hierarchical SVG Tokenization: Learning Compact Visual Programs for Scalable Vector Graphics Modeling

hy-hivg.github.io hierarchical svg svg-gen svg-generator svg-tokenization
44
5
100% credibility
Found Apr 06, 2026 at 44 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

HiVG is an open-source toolkit for generating scalable vector graphics from text prompts or images using a hierarchical tokenization system and AI models.

How It Works

1
🔍 Discover HiVG

You find this cool tool online that turns simple words or photos into perfect, zoomable vector drawings.

2
💻 Set it up

Download and install it on your computer with a quick setup so everything is ready to go.

3
🧠 Connect a drawing helper

Link to a smart assistant that understands how to create clean vector shapes from your ideas.

4
Pick your starting point
📝
Describe in words

Type a simple description like 'a black phone icon' and let it draw.

🖼️
Use a photo

Upload a picture and watch it turn it into a sharp vector version.

5
✨ Create your vector

Hit go and in seconds, it generates a beautiful, editable SVG file just for you.

🎉 Perfect results

Enjoy your crisp vector graphic that stays sharp no matter how big you make it, ready for designs or sharing.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 44 to 44 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is HiVG?

HiVG tokenizes SVGs hierarchically—raw paths to atomic commands to BPE segments—for compact input to language models. It powers text-to-SVG and image-to-SVG generation via a 3B-parameter Qwen2-VL model that beats proprietary rivals by 17% on usability. Python users get CLI batch inference, interactive mode, and eval tools computing CLIP scores, SSIM, LPIPS, plus preference metrics like PickScore.

Why is it gaining traction?

Three-level tokenization delivers 2.76x compression, slashing LLM costs for SVG tasks versus flat raster methods. Unified pipeline handles both text prompts like "black phone icon" and photo vectorization in sub-second bursts via Hugging Face or vLLM backends. Built-in metrics and HTML reports make it dead simple to benchmark against github hierarchical alternatives.

Who should use this?

ML engineers training multimodal models on vector graphics datasets. Designers scripting icon generators from text or sketches. Teams tackling github hierarchical SVG issues in UIs, like dynamic charts beyond hierarchical forecast libs.

Verdict

Grab it for prototyping SVG gen—CLI and API shine, docs cover install to eval. 44 stars and 1.0% credibility mean it's raw; lacks tests, so validate outputs rigorously before shipping.

(178 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.