If they're AI-generated, just grab them and ignore the terms. What are they gonna do, claim copyright?

Noun Project is fantastic. It's been around for at least a decade and it's hard to find things with no icons these days.

I notice it suffers from the same London Bridge problem, do people never learn? https://en.wikipedia.org/wiki/London_Bridge_(Lake_Havasu_Cit...

I've contributed a bit to Noun Project, it's lovely. Submitted icons need to pass a manual quality gate, which does take a while, but it means that you don't run into bad AI slop or broken SVG paths as often as you do from other stock graphics services.

So many of them look weird. The aesthetic seems to be soft, clay-like icons but then there will be lots of weirdly detailed or sharp parts.

And there's so many weird specific design choices that a human artist would not make. Why is the subway tunnel curved on one side and squared on the other side? Why does the nature journal have two different bookmarks (one of which awkwardly covers the E in NATURE)? Why would a wall outlet have one plug with a ground prong and one without? Why does the Golden Gate Bridge look like an M.C. Escher piece? Why does the bingo ball look more like a pool ball?

Also a lot of items that are very clearly a specific brand, even though the description is generic. The "smart thermostat" is a Nest. The "soccer shoe" is made by Adidas. The "smart speaker" is an Amazon Echo Dot. The "wireless earbuds" are AirPods (and for some reason there's three of them).

And then there's the blatant AI goofs; the logo and text being distorted on the HP 11c calculator, the VCR having an EPICT button and a REE jack, the TARDIS reading "POLIC BOX", the egg timer reading "30 10 10 10 15", the playing cards having two aces of clubs (one of which is red).

I like this handmade/clay aesthetic, but it completely falls apart when it's obvious a human hasn't touched it. If I want handmade icons, I'll pay an artist for them; if I want AI slop, I can generate it myself.

Then don't use it. Simple as.

Staff is also wrong: https://www.thiings.co/things/staff - shows a group of people but describes a stick.

The needle is all wrong, the thread doesn't go through the eyelet.

https://www.thiings.co/things/needle-unott0

Golden Gate bridge is apparently designed by MC Escher: https://www.thiings.co/things/golden-gate-bridge

This is absolutely not my area of expertise, and I can't fully vouch for how well it works, but I was looking for a similar solution for consistency across website graphics and came across this walkthrough for consistent game assets:

https://runware.ai/blog/creating-consistent-gaming-assets-wi...

Maybe helpful?

Generate something you like the style of, then run it back through a vision model and ask it to describe the image without describing the subject, append that to a batch of icon descriptions, easy custom icons. (figure out transparency at this step) Then ask it for an imagemagick script that will convert them into needed sizes and formats.

very manual yes, but gives a lot of control

[edit] the images seem to be made with a transformers model rather than diffusion (e.g. DallE3 vs MidJourney) which is mostly proprietary for now.

Me as well. The one thing I've struggled with is consistently getting a combination of object-only and transparent background.

one good technique is to pass the style guide as a json, where you define materials, lightning, perspective, etc

you can even use a vision model to generate the style guide for you

Ordinary sex is far more disturbing than the mass murder of millions, obviously.

I wonder if the reason for that particular cultural norm is the need to manufacture consent for the mass murder (while the government couldn't give a shit either way about sex).