¡Presentamos nuestro modelo LoRA de sprites para Ragnarok Online en Citivai! 🎮✨ Con más de 190 imágenes de alta calidad, es perfecto para los fans y creadores que buscan llevar su creatividad al siguiente nivel. ⚔️
Únete y colabora con otros apasionados de Ragnarok Online en Citivai. ¡Juntos podemos hacer crecer esta colección épica!
Description
FAQ
Comments (13)
Oh HELL yeah! I started pixel arting a little over a decade ago, and while I learned a lot...it was just one of those things I was too slow at. Tools like this are valuable. Not because they directly create game-ready assets, but because they give you a good starting point. The 'workflow' I'm thinking of is as follows:
- Conceptualize character design
- Generate images of how character would look as a sprite
- Pixelize, use a custom color palette that's consistent with the world aesthetic you're going for, and correct little erroneous pixels along the way
- Adjust for inevitable height and proportion differences that are bound to occur.
YES! It is my idea to create a tool that helps artists really to optimize work times, with a correct workflow I am sure it will do a lot! Some things that you mention could be corrected from the training stage of the model, since These are the first tests I do. I will have to improve the dataset in several aspects, thanks for the feedback! I hope you enjoy this model
@Leem0nStudio So the one thing is, I don't know how many of those things could be corrected on the model level. These generations are made from noise and averages, so even if you only use one very specific artist's style and have 100% consistency between color palettes...it'll always be 'blurry pixel art'. A good pixel artist (unlike myself) can do the things necessary to make it good pixel art afterward.
But the reason this tool is so valuable even now, is because it can create very good references that the aspiring artist can look at and say "OHH, that's how I would draw this outfit at different angles!" I semi-gave up pixel art a few years ago and opted to render chibi character models using toon shaders, to try and get that "3d characters in 2d" look (like FFT, Saga Frontier and the like). Once you have a character designed, you don't have to worry about drawing it differently for each direction...you just rotate it.
By all means, continue to improve this model if you have a good handle on things. I'm eager to see how far it gets.
@bh_miscreant in fact I also tried to create 3d characters with blender and DAZ STUDIO, deforming the bodies to make them look like ragnarok bodies, I also know that the creation of consistent characters for conceptual art is possible by combining models that are trained with different perspectives, other tools like controlnet and openpose, I think this journey is just beginning. I will continue working to achieve my goal which is to be able to create the 3 views of the character consistently.
Publishing lora without description and 2-4 examples is considered as dick move.
And dick moves make babies o///o
Awesome, illustrious nails pixel art styles perfectly
Nope. Illustrious don't know what to do with pixels at all. It's not trained to do them at all.
Great work!
It's strange, but v3 and v4 loras constantly produces NaN errors in Unet. First time I see such errors on SDXL.
I try a different parameters for original size sprite with alpha... This is my mistake, recently update a new datasets, sorry 😔
Ok. I used Comfy and even in it v4 just output garbage, but v3 output picture.
But with v3 picture basically same as without any lora. Nearly zero impact. Only adds some noise to variety, like bg color everything else same as without any lora.
@Leonmitchelli , sometimes experiments can fail jeje, try the last version (2.1a), later delete the other version. ( 4 and 3), thx very much for your feedback