10 comments

[ 3.6 ms ] story [ 31.4 ms ] thread
(comment deleted)
What did you train it on? Would be great to k ow if I’m opening myself up to liabilities by using it
This is almost certainly a custom (or copied) prompt to an image generator API + open source raster to vector conversion library.
Don't think so, raster to vector conversions don't look that sharp
Looks interesting, but inviting users to try your service, letting them put in effort (in this case typing a prompt) and THEN forcing them to login is considered a black pattern in my book.

I understand some form of rate limiting is necessary, but this bait and switch setup always feels like a kick to the face.

Yeah, it says it will create an SVG in seconds, but seconds later, there is no SVG. Did not occur to me that I had to log in, but probably won’t as I cannot be bothered to follow black patterns…
Nice job!

I've been working on a similar project @ https://vectorart.ai

There are lots of interesting problems to solve and I feel SVGs have been somewhat underrepresented in the AI boom. But that's beginning to change...

Can somebody explain how the workflow works here exactly? Is the LLM trained on SVG? If so: could it hallucinate SVG properties or so? Or is it a regular image generating AI that vectorizes raster images afterwards with traditional tooling?

I'm a noob in that field but I'm curious about potential risks. ;)