木叶飞舞之处，火亦生生不息

Not to downplay it or anything, cuz the outputs do look pretty decent; but, why are you 'seriously impressed' that they 'only trained it on one concept'? Wouldn't training multiple concepts and still getting a good output be more of an achievement? Again, not trying to offend anyone, just not too sure if I'm interpreting your input in the way you intended it.

Lazman This LoRA came out in 2023. Back then, reproduction of details like the leaf symbol on Naruto's headband were not common place in AI. It often came out jumbled. Advances in training techniques/algorithms eventually lead to LoRA like this that nailed details much better than was possible before.

I think to avoid being confused next time, you need to keep in mind the time period of the content. Especially when dealing with AI, don't forget how rapidly things have been developing. Go look at the outputs of older SD 1.5 checkpoints and you'll understand better.

ArtificialSweetener_ Oh, I know. Soon as you said "2023" I already knew where the rest of your comment was going, lol.. I got HEAVY into AI nearly a year ago now, and did deep dives into the inner workings of it, and learning how to get decent outputs, learning how to caption for training purposes, different techniques. Also learned different kinds of AI.

I currently have working programs for text, image, video, and even(to a lesser extent) speech AI on my computer.

Unfortunately, speech AI seems to have some way to go, at least last time I jiggered with it.. Well, for local, the corporate stuff is more refined. But, they got millions to spend on hardware, and often have proprietary software. But, I'll never do AI on corpo. I prefer not getting the sense that someone's looking over my shoulder every time I hit 'generate'.

Corps are creepy these days, cuz they lobby just enough so it's legal for them to be creepy..