ilta222

portfolio about links shop journal tutorials contact
protect your pixel art from ai
      when more and more social medias started training AI on unconsenting artist's work, i went into a panic. i frantically researched all the different ways i could protect my own work and started glazing and nightshading everything i had ever made, until i had a revelation.

      pixel art is a very distinct medium and we've been approaching this the wrong way.

      i'm going to explain a few things about ai art models and downscaling, and then i'll explain how i've been protecting my work.





      ai art model differences



      standard ai art models produce results like this, with no pixel grid consistancy. they look objectively bad, are completely unusable in professional game dev settings. considering how unappealing this 'pixel art' output by regular art models is, protection methods like nightshade and glaze are seemingly useless to pixel artists. using those protections, you would essentially be protecting against people using your art to produce results like above, which people don't want anyways.


      ai pixel art models produce results at original resolution, and generally capture the 'essence' of pixel art and use common techniques pixel artists use, like anti aliasing and sel-out. to train ai pixel art models, original resolution, 'downscaled' pixel art is used. this is the only way to produce results that employ usual pixel art techniques. so, because of this:

      the most important thing for pixel artists to protect is their grid.


      you need to make it as difficult as possible for someone to downscale your art, so that they cannot use the original resolution to train a model with your art.





      downscaling explained

      standard downscaling

      is when you take pixel art that's been 'cleanly upscaled' (intervals of 100%, shown above) and downscale it to the original size. you can figure the original size out pretty easily by cropping to one pixel on the image, counting the amount of pixels from left to right, then dividing the art's resolution by that number. algorithms could also be created to do this automatically.



      manual downscaling

      when the pixel art grid has been 'broken', the only way to downscale pixel art perfectly back to the original resolution is to manually go through the image and fix the grid inconsistancies, then downscale after you have fixed it. this is extremely time consuming and i doubt anyone using ai would have the patience to do this.





      how to break your grid


      so, we need to make it as hard as possible to get the original grid by intentionally breaking it, but we ideally, we want to do this in a way that isn't super noticeable.
      pixel artists have actually been using methods to do this for years in pixel asset packs, so that people can't use their packs without purachasing. here's the 3 main methods:

      badly resizing very good and easy grid break. just resize to percentages that aren't intervals of 100, so like 403%, 577%, 289%, etc. do not use this on animations! it 'animates' the resize and looks weird.

      rotating rotating very easily breaks the grid, and is very covert. though, the stronger rotation, the stronger protection. just merge your layers, select everything, and rotate at one of the corners. crop to avoid transparent edges.

      cropping cropping only protects the outer border, but it does make it more annoying to get the image back to original size. best used in tandem with other methods.

      blurring pretty effective, but can look pretty bad. i would use very sparingly if you decide to use this.

      artifacting also effective, looks a little better. use sparingly still. search 'add artifacting to image' to find online tools to add artifacting.



      my method


      you can develop your own protection methods, or use mine! if you make your own method, i will say combining multiple is always a good idea. also, my method will work with animations, since i select the layer. here's what i've been doing:

      1. merge everything
      2. select entire canvas using select tool
      3. click on the layer name
      4. grab a corner of your selection and rotate slightly
      5. crop so that there's no transparent corners
      6. do a 'clean resize' for animated art OR do a 'bad resize' if static art




      that's everything! feel free to DM me or use the contact form for any questions. i would also prefer you ONLY share this with artists you 100% know are anti ai art, as i don't want this getting out to the pixel art ai model creators.

      hope this helped you!
      keep fighting the good fight <3





      disclaimer


      although i have done as much research as i can on how pixel models work, i am not a data scientist so i don't fully understand the intricacies. this is all based on the research i have done myself, and i could very well be wrong or things could change in the future to train models on pixel art more easily. i can only confidently say that these methods will make it much more difficult for someone to manually train ai pixel art models on your work. this is in no way legal advice or counsel.