Anybody have icon need about the apps?
The AI ​​must be trained to do this specific job. It is doubtful whether such AIs exist.
You can always try to train an AI to produce images but, to be useful the image should be in hvif or iom format. And if hvif is similar to svg, it doesn’t allow to embed images in another format.
One may use AI to generate a reference image, but avoiding converting it from raster to vector and importing the result into IOM
Or train the AI to use hey then everything is possible… 
I think it’s better to let some human with some graphically artistic prowess handle icons.
5 Likes
I don’t think we have to debate the general pros and cons of AI in multiple threads here on the forum.
Nothing stops you from experimenting with it. You can try if you can use AI to create an image (not just random, but the one you imagined first) and then see if you can have it converted (or generated in the first place) in our HVIF format (or at least SVG that is importable in Icon-O-Matic).
If you have figured out this process in a reliable and reproducable way you can look for projects that are in need of icons (my own app Kottan would be a good candidate).
I didn’t see anything about AI art prior, so my bad if I am duplicating efforts here. I just put my little two cents in, and don’t have much to offer to debate.
AI art doesn’t seem to me to be art at all. For something so small as an icon, it may seem irrelevant, but I think it isn’t good for creativity in Haiku.
One could say also, that it’s a creative project of its own, making Haiku icons with AI, so there’s that too. But since the AI thing is a whole big fad now, I feel it won’t be unique or as neat as anticipated. Of course, if Edouard wants to take a crack at it, I wish him well. 
1 Like
Possibly someone could train or finetune a LLM on SVG images, and use that as an alternative to other AI image generators? Has that been tried yet?
1 Like
I did something a few yrs ago, but looking at the original thread, you never answered to my proposal
2 Likes
9 Likes
I wonder how an LLM could ever match zuMi’s talent…
6 Likes
It could eventually if it was trained by him. And that’s exactly the problem of AI; the resources that you have to invest for a good result are considerable. In expert time always but, also in money if you’re hosting your AI. Add to that the cost of a professional model if you have special needs.
As with any apprentice, once the training done, if you’re starting to get things done quicker, you still have to check given results for errors.
At that point, a company will duplicate the AI, fire artists and hire a guy that will check multiple AIs work. They will save money at cost of real creativity because you can’t count on AI for innovation.
1 Like
Given the energy need of all the servers required to train and run the AI, I’m not even sure of that.
7 Likes
Oops, sorry, my bad. Must have somehow escaped my attention back then, don’t know why. I like the icon. Thanks a lot for making it.
Could you make a PR to the Kottan repo at GitHub - andimachovec/Kottan: an editor for archived BMessages on the Haiku operating system, submitting the hvif or rdef version?
1 Like
Automation will be the death of innovation.
1 Like
Automation is also an innovation.
Maybe we can do better than throwing oversimplified generalized one-liners at each other 
And at least roughly stick to our topic of Haiku, for that matter.
2 Likes
3 Likes
