Hacker News Reader

Comment by AmbroseBierce | Hacker News Reader

I have tried to use ChatGPT and Google's Gemini to make SVG from simple logos bitmaps but its still a daunting task for them, so I guess tools like this one will still be needed for a while.

View on HN

If you search for ‘vectorization AI’ there are a handful of specialized tools and apis that can do it. It worked well for a handful of logos I wanted to convert. Nano banana generated the raster logos, and these other tools vectorized them

Even inkscape can do this

But only gives useful results some of the time. But I don't know if "vectorization AI" is already better.

I haven't seen one that worked properly—can you list a couple examples? Some of the ones that say they're "AI" are just VTracer / Potrace and don't give nice control points.

I liked the results of vectorizer.ai and recraft.ai

Input image is important too. When working with the generalist LLM on the raster art, give it context that you are making a logo, direct it to use strokes and fills and minimal color palette, readable at small sizes, etc.

vectorizer.ai is amazing. It's worked great for like over 10 years (back when it had a name like vector magic or something). I'm super curious how it's implemented

Others have mentioned SVG AI tools... I've tried 3-4 over the previous days and eventually ended up with svgai.org (after I've used Google Gemini for bitmap).

You can instruct it to make edits, or say "Use SVG gradients for the windows" and so on and you can further iterate on the SVG.

It can be frustrating at times, but the end result was worth it for me.

Though for some images I've done 2-3 roundtrips manual editing, Nano Banana, svgai.org ...

The advantage is that it produces sane output paths that I can edit easily for final manual touches in Inkscape.

Some of the other "AI" tools are often just simply algorithms for bitmap->vector and the paths/curves they produce are harder to work with, and also give a specific feel to the vector art..

It seems like the problem of pushing pixels around in an exact way and iterating on visual design is a problem that needs very specialized tools, regardless whether there is LLM support.

Free idea: turn this into an MCP server. Give the agent the ability to virtually "hover" a path and see which part of the final render it corresponds to

If anyone sees this, I tried it and unfortunately am not getting better results on the pelican-on-bicycle test. I think the vision models just aren't good enough yet (I tried Claude and Gemini)

I can share the code if there's interest.

You can get pretty decent initial results if you explicitly tell them to first make a detailed description with exact coordinates and then feed the description back into them to build the SVG.

Try Claude code. I have built so many. Entire pitch decks for my startup. It is the best. Tell it to use animation libraries gsap framer motion etc to build svg.

SVG is often relatively complex and dense.

A dedicated or fine tuned model for just SVGs would be pretty wild.

The problem isn't really SVG but the more complex problem of looking at a, possibly noisy, image with continuous color variations and identifying the cutoff point where you contain one part in a border and a different part in another border. That can be judgement call that is made better if you actually understand what is represented but harder if you are working at the pixel level.

Reply on news.ycombinator.com