This model is outdated, look at the newer one I've trained on PonyDiffusion. It's considerably better.
You can find it here.
A new and improved sprite model trained on the battler sprites from LISA: The Pointless SOTW! This model is a LOT better than my previous LISA sprite model, which was trained on Painful sprites.
This is a lora trained on a whopping 176 sprites of characters from LISA: The Pointless SOTW! It was trained at a huge 896x896 resolution, which helps with the 'pixel art' style a lot. I training it mostly on the human fighters, so few animals and no joy mutants or shadow monsters. I'm VERY satisfied with the results of this LORA, I wasn't sure if it'd work out since I was trying a lot of new training techniques, but it came out great compared to the last one. It was trained on danbooru-style tags, so use those when trying to prompt - I've included the tags in a .txt file in the "Training Images" download below.
IMPORTANT: This was trained on the Fluffyrock 896 multi-resolution model. It won't work with pretty much any other model, so please make sure to use this model, or any of the newer higher res Fluffyrock models when generating anything.
NOTES:
-You must generate with the Fluffyrock model, and ideally, you should generate at 832x832 resolution (lora was trained at 896x896, but Fluffyrock only supports up to 832x832). Although I've had success at different resolutions, lower ones like 576x576 might get a little weird.
-I specifically tagged infinity franchise members as "jerseyhead", so use that if you want a character wearing that funny 88 jersey
-Changing the aspect ratio from a square will also work, although all trained images are square, so keep that in mind.
-You can actually have detailed background, if you put "purple background, simple background" in negative prompt, and describe the background you want in the main prompt. Keep in mind all trained images have a purple background.
-It can do concepts that weren't trained, such as women, or vehicles or what-not, so try it out in those regards. In fact, I guess this could be a 'LISA-styled' lora, since it can do landscapes and backgrounds.
-It wasn't trained on any joy mutants, however it can still do monsters and gross creatures pretty well if you prompt it good enough.
CREDITS:
Austin Jorgensen - Creator of LISA
Edvinas Kandrotas - Creator of Pointless
All the devs on SOTW - They made most of these sprites!
MrFood - He made this model, I'm just uploading it in his place :P
Example prompt: (See example images, all images should still have their metadata. Just drag the images onto the PNG info tab and import it.)
MU never ever...
Description
FAQ
Comments (9)
This is fantastic. Did you upscale the dataset with nearest neighbour or bilinear?
Nearest Neighbor, of course. I also made sure to upscale by a percentage instead of an arbitrary amount, which helped prevent some artifacts that can happen when upscaling pixel art.
What exactly is the process to use this? Can I use it without paying for things? Do I need python etc.
Otherwise this looks scarily cool!
I got InvokeAI going with the files set up but it keeps giving me server errors.
EDIT: Got the pointless lora set up and fluffyrock as the model and set the settings to the same as your prompt but it still gives me server error. Never used this stuff before so im not sure if I forgot to set up anything
@Celulamp Gonna be honest, I've never used anything except local stable diffusion, so I have no idea how you would set it up on your end.
All I can say is that I would try to import the metadata from one of the example images, and just generate at 832x832. Do note that it has to be the exact model I linked, or it might not work (even the newer versions of the fluffyrock model don't seem to work well it with it.)
Edit: Here's a link to the checkpoint as Zooch's huggingface URL is down.
But the his Pony lora is recommended over this one.
This should have a download:
https://gitgud.io/WithoutOrdinary/fluffyrock-model-safetensors/-/blob/main/fluffyrock-2.1-832-multires/fluffyrock-576-704-832-lion-2.1-low-lr-e20.safetensors?ref_type=heads
THAT BEING SAID... You should really try and use the much better version I trained a month ago, it's trained on PonyDiffusion. You can find it here.
@ZoochMan you know i am SHOCKED i somehow missed that in the above description.
Big up for the fast response and yeah I'm gonna be checking the Pony lora out.
Edit: I asked a question about the fluffyrock models because I found the naming convention confusing. Ultimately I'm wondering how you managed to find this and use it in the first place but that's none of my business.
@2buckchuck You missed it because I added it after I responded to your comment lol. I didn't realize it was down, so thanks for bringing that to attention. I fixed the descriptions URL, too.
I found it when I found it in development after looking for a decent furry model, or anything that wasn't 'anime-based' since everything felt so samey with those models. I caught that model early in training and made this lora before they started making vpred models (which are a lot harder for me to train on). It doesn't really make much sense for me to have trained this on a furry model, but the results were great in my opinion.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.