@kftiger what was your dataset? I know theres a fan made ppttyd uhd texture pack you could use thats very faithful to the art style. Its in chunks so it would be much easier to tag things as mario's arm (pmttyd) etc. I think theres sprite assets from the sprite-resources of the latest paper mario character sprites. Which should help make a larger dataset.

@FoxMccloud2022 It's just that have to build them yourself, if there's no one who can build them. For example, Rayman Origins sprite was the same situation before it could be used to make training.

Now, someone who could build them, with the time and effort, can be the biggest help to the lora maker.

For example, with GIMP, if you don't have Photoshop or something similar, it's easier to create a layer by separating sprites from one entire sprite collection image.

@GuugeliBox Wouldnt it be easier just to dump the sprites to save the haste of editing the images than just tag the files.

@FoxMccloud2022 Nope. Likes if want custom poses what also good Lora dataset.

@GuugeliBox A lora wouldn't contain pose data. What you need is a good checkpoint model. Than add every single prompt you can during the tagging phase to associate a image to the pose. A lora itself wont help with poses since that's dependent on the checkpoint file while has all the non-character trained information.

@FoxMccloud2022 If you say so. Once one of these had a request and it was also helpful. Note that not always from the front, back or side. But from every angle. Because if it comes from one angle, it makes one angle an excellent art style.

So one had problems with Lora and fanmade didn't help at that point because her art style also varied despite the character.
Now, if you want to make a great artstyle, use sprites and maybe even add custom poses. Not just one and the same. Also with other characters and objects. :)