Hacker News new | ask | show | jobs
Show HN: I Built a Free Text-to-SVG Generator (text2s.vg)
2 points by reymaggle 519 days ago
Hey HN,

Last weekend, I got bored and decided to build something fun: https://text2s.vg – a simple tool that lets you type a prompt and generate a free vector SVG image.

Why I built this: I wanted to explore combining text-to-image generation with lightweight vector outputs, all while keeping it fast and accessible. It's a good way to progress at creating a project from A to Z, from backend workflows to frontend design. Plus, I thought it might be useful for developers, designers, or anyone working on creative projects.

How it works: Type a prompt, hit "Generate SVG," and you instantly get an SVG file ready to use. It’s completely free, no ads, no signup – just something I made for fun.

The tech stack: Backend: A custom ComfyUI workflow (using Flux, BiRefNet, and a custom node ComfyUI-ToSVG,) all running serverlessly on Runpod. Frontend: React with Tailwind CSS (no Next.js, no need for SSR). Database/Hosting: Firebase handles Firestore, storage, and hosting. What’s next:

I’m looking for feedback to improve it or explore new features. Maybe adding style controls or export options?

Would love to hear your thoughts, questions, or ideas!
1 comments
maalber 519 days ago
Cool! I appreciate that you made it openly available without signup!

If I understand correctly, it first generates a bitmap using a text-2-image model and then converts that to svg? It might be cool to see the image alongside the svg and then be able to adjust some settings such as the number of colors to include and so on.

Did you do any training/fine-tuning yourself? Are you looking to do so at some point?

link
reymaggle 519 days ago
Yes, exactly! The first step is generating a bitmap with a white background using Flux 1 Schnell (with a svg render Lora). The second step removes the background with BiRefNetUltraV2. Finally, the last step transforms it into an SVG using this node: https://github.com/Yanick112/ComfyUI-ToSVG.

Your suggestion is great! I'll definitely add a step-by-step display and some settings to adjust the results.

As for the last question, no, I don’t plan to train or fine-tune a model. I’m more interested in building final products and using the tools and resources provided by the open-source community.

link