AI Generated NFT Content, & how to spot it
Nowadays it’s important as an NFT creator to be transparent about how your NFT was created. In this short article we want to talk about a couple of the types of AI assisted NFTs you might see, and a few of their identifiers.
We’ve got absolutely no issue with any piece of software or other method being used in this way for NFT creation, we’ve used high and low end AI tools in our own creation in the past. This content should always be appropriately labelled.
A basic example
AI can be a great tool for artists, it can quite literally change an image into something completely different. There are many different types of high and low end AI which can be applied to images. From a sculptors perspective powerful AI algorithms reveal the true intended shape and appearance of an object.
Here’s an example application applied to a 32×32 piece of pixel art (For us this is definitely one of the most interesting applications)
A little AI, and whoosh, our pixels are transformed and refined:
In this case we have first upscaled the pixel art and then applied several AI filters.
A general rule for identification – the more AI based filters and modifications that are applied, the more “corrupted” the image will become. We can see some examples of said corruption above. Most of them with AI will tend to blend into the image, the most pronounced sign is the banding (outline) around the moon, and the raven it’s self.
We can also see a lot of creasing here , these are generally produced when you put an area with static colour through the AI filter.
So, if an image looks a bit too abstract to you (which is also another sign), it’s good to check for these additional indicators.
Sometimes though they will not be present, there are methods of applying AI effects sequentially and to a degree which have the result of seriously minimising these indicators, however this is a skill that needs to be built over time and you won’t often see it on mis-labelled NFTs.
You might have heard of upscaling already, it’s got some great applications in emulation, and gaming in general.
Here’s an example with something mid-range, let’s use the Cosmo Canyon pixel background as an example (Credit to Square)
This is the FULL SIZE of the original image at 512×512 pixels, lets AI upscale this
We used an upscale to make the image 9999×9999 with basic AI smoothing included
The image is far above the upload limit for the blog, but here’s a screenshot:
Again we see quite pixelated imagery becoming crystal clear, this is the general effect of an upscale.
It’s harder to spot, but you will see similar signs. Creasing, banding, and also unnatural distortions
The steps are a great example, you can see that the last step is corrupted
We hope that this was a helpful read!