Minimal Reflections

An Extendable World

We are entering a world of automated variation and extensibility of media, and this piece is a prediction of what might come.

The ability to extend and variate media is becoming easier, more accessible and nuanced, allowing for more complicated, specific instructions on just how to variate and extend.

“In the style of”

A great artists life’s work becomes a personality of sorts - a vibe, with patterns and data which genAI can understand and interpret.

Asset creation in mediums such as games will become much less time consuming. For example, uploading one sprite of a character could generate the entire sprite sheet.

If you are making a game where the player build their own guns, you could create one base gun sound and get variations for each attachment that the player can use. Or better yet, let some AI take into account all the different aspects of the gun and generate the sounds on the fly, creating a completely unique sound depending on the player’s current build, all while depending on some base-sound, or perhaps just some base-prompt, to get that cohesiveness.

You could create one piece of music for a part in the game, then have some model adjust its tempo, dynamics etc. depending on what’s happening in-game all while never deviating from the style of the original track.

You will be able to improvise on a keyboard, and a whole orchestra (or whatever style of music you want) will be generated in real time, reacting to your playing. Or, you could choose to improvise in the style of something more specific, like an artist, such as The Beatles, or in the style of some specific album even. You might have three sliders beside you, that controls the model’s parameters; stability, controlling how far the model deviates from the style of the source material; dynamics, controlling how intense the generated music gets; and density, which controls how thick or sparse the generated music feels.

Stealing People and Characters

You will also be able to perfectly steal someone likeness much better and easier than the current video and audio deepfakes. An AI model can be trained on all interviews of some person and generate videos that perfectly replicate that person’s mannerisms and likeness, in any setting. This also entails fictional characters. You will be able to use this person or characters “style”, just as you will be able to use the “style” of The Beatles, or Chopin, or Picasso, etc.

Pre-existing tv-shows and movies will be able to become interactive, with characters being generated perfectly in character. The data is all there, it’s just a matter of how good the models become.

A unique and distinct work now allows for infinite automated variations and extensions, where as previously it had to be done manually.