This has been an experiment to see if I could create everything using scripts and AI.
If AI couldn't do it I'd get it to create the code such as API calls and so on. This websvg.com site was completely created using these AI tools. Including the DNS being applied, the Cloudflare Pages were automatically set up and the the web site was a Svelte 5 application generated by v0.dev and Cursor. Every image was generated in Midjourney and converted to SVG.
I have now taken all of these scripts and can create a similar landing or directory site in less than a minute, provided I have the data.
Anyway it's been fun.
- Contact email in the footer is “hello@agentwebfactory.com”, which doesn’t seem right. Unless that’s also you?
- I didn’t read the part about this being AI-based before clicking through to the site, but those descriptions are unmistakable. Ditch the bullet points.
- You can’t make people credit you for AI-generated images. And vector tracing them is not copyrightable creative work either.
Yes, https://agentwebfactory.com is my project service that creates directories and websites.
Agreed about the content, it is too AI created, and can certainly fuel distain.
Mmm point taken about the crediting - my bigger goal with this is to actually build a PNG to SVG tool.
Thanks so much for the feedback.
This has been an experiment, and it does appear that people will use the tools more and more. Unfortunately we can't escape it - which I agree doesn't make it right.
Perhaps some kind of better licensing? The nature of these tools is that they exist and they have scraped all the "things". Maybe having your own private agent's is the way forward?
It’s only cheap until you or someone who uses one of these images gets sued for copyright infringement and loses. I, for one, will be rooting against you in that court case, pot-stirrer.
It seems that what happens is that the four mentioned thumbnails are there initially while the page is loading.
And that once the page finishes loading everything and running some JavaScript, it replaces the featured items.
If I reload the main page, it sometimes updates the thumbnails also but often it keeps the wrong thumbnails shown.
If I click on one of the featured thumbnails, and then navigate back to the main page, then it seems to consistently show the expected thumbnails that belong with the links in the featured row.
But doing a fresh load of the main page either by navigating to it from your HN submission link, or by reloading the page, ends up showing those initial thumbnails instead of the replaced ones most of the time.
Thanks so much for the feedback. I used Svelte 5, and I think there is bug that doesn't refresh the page correctly, I'll dig into it.
I have to say on the whole Svelte is a really nice framework - before this I'd never built a JS project before (plenty of Python and Ruby). It's interesting how the AI tools can support you in a language and framework you don't know well
I should also note that this buggy behaviour was observed in Brave browser (Chromium based) on desktop.
Looking at the site now in mobile Safari on iOS it seems to consistently show the expected thumbnails in the featured row both on first page load and on refresh.
Yep, point taken. I wanted to try the easiest thing to completely build a directory only using my own agents, and other AI tools.
Ultimately I'd like to build some kind of software based on feedback and traffic.
I looked at a bunch of Vectorising tools, and in the end used https://vectormagic.com/
It works really well, although I had to do all the conversions manually. They do have a bulk tool, but wanted to try this out first.
It does also produce .eps as well as .svg
I tested it with a bitmap image that ChatGPT 4o created for me based on the following prompt:
> Owl line art, stylized, scientific, illustration
I tested converting it to SVG using first “black and white” mode, which was unsatisfactory because it is literally only completely black and completely white shapes, which loses detail compared to the original image that had multiple shades of grey. The file size was around 291 KB which seems to be not too bad.
I then tried converting the original image with color mode instead, and the result looked good to me. The file size was pretty big, at 3.3 MB. Presumably because it has to use a lot more individual shapes to capture all of the shades in the original image.
I conjecture that if I fiddle a bit more with the prompt for generating the bitmap in ChatGPT, adding things like “unshaded”, “crisp lines”, etc, you might get a bitmap image that will be better as base for a black and white only SVG conversion that captures the original image well while keeping file size around the hundreds of KB.
I am on mobile so I have not tried editing the SVG in Inkscape. I assume that with the details in the image I generated, it consists of many small parts and is probably not well suited for manual editing.
I also know that Adobe Illustrator has a “Live Trace” feature since many years that might also work for this kind of “conversion” to vector graphics. It’s been a few years since last time I tried that feature though.
it would be super cool if there were a way for users to create their own SVGs using your flow -- these look great!
i was just playing around with SVGs yesterday and was wishing there were an easy-to-use tool out there for fiddling around. i couldn't get chatgpt 4.5 to get it quite right and ultimately abandoned it in favor of something premade
Thanks! I know it's controversial but it has taught me a lot about building with AI, as well as the legal and copyright parts.
I experimented with a number of prompt styles this was an example of one I used on Midjourney:
"Vector line art, icon of a flower, cute and minimalistic, made with bold outline, white background --stylize 400"
I felt I needed to add some kind of license, I also wanted to get attribution back to this website for traffic purposes.
Dare I admit that I just asked ChatGPT what to use... I then got it to create a Python script to apply the license to the SVG image itself and add the website details.
Its been a fascinating experience
The latter part of your sentence varies wildly by jurisdiction.
In the US at least, there have already been rulings considering it "fair use" (which is not a concept in almost any other country AFAIK)... I believe it is seen as basically doing an equivalent thing a human does i.e. drawing influence from other sources it has seen before to create something "new". The only legal difference in my mind is how different it actually is, and how obvious is it, from whatever the original influence(s) were, which also applies to humans.
If it's being served up to the U.S., files downloaded there are not copyrightable, hence not subject to restrictions imposed by a license which has its origin in copyright.
It's been fascinating experiment. The stuff you can now do, its like having scaffolding to help you.
You can create all the tools you need yourself and can host it for virtually nothing. I just upgraded Cloudflare to the $5 plan (btw I have 18 sites hosted atm and that price includes up to 100). I also paid for Fathom analytics $15pm and Cursor, ChatGPT, Midjourney (although cancelled now) and VectorMagic (also cancelled). So you could spend just the domain name and the AI tools, making it about $20. Totally bonkers
- I didn’t read the part about this being AI-based before clicking through to the site, but those descriptions are unmistakable. Ditch the bullet points.
- You can’t make people credit you for AI-generated images. And vector tracing them is not copyrightable creative work either.
reply