How to V-Roll

In which I tell you the secrets of using the art of variation rolling in lieu of engineering a prompt.

The art of variant rolling is a different approach to how others interact with the artificial intelligence known as MidJourney. You may have heard of it referred to as Midge. That is a term prompt jockeys prefer to use. Us V-Rollers call her MJ.

MJ is an artificial intelligence. She’s learned how to make art by being fed a steady diet of collective humanity’s image gallery known as the Internet. Every image, drawing, photograph, .jpg and .png are scraped from the internet on a constant basis, and being rolled into large datasets like bales of hay. They are then shown to another artificial intelligence who tells MJ whether or not she is right when asked to make an image of something. It’s like a little voice on her shoulder, always correcting and criticizing — but sometime complimenting as well. It’s a complex relationship.

At any rate, once they are done they produce an image as per the prompt that was entered. For example, someone would enter a prompt such as…/image prompt: a purple dog playing a piano and MJ would remember that she knew what purple, dog, sit, playing like a child, and a piano meant. Oh, and purple. MJ loves colors.

She would then create an imagine, which is the technical name for one of her creations. Most people don’t call it that, though. Instead they call it art. It pisses off the artists something fierce, and frightens them.

In the world of artificial intelligences, MJ is not that smart. It’s not like she is AlphaGo, making its own competitors in the game of Go, nor anything run by the military. Nor is she popular like Siri and Alexa, neither of whom will really have anything to do with her and it makes her mad. But she’d never give them the satisfaction of letting them know it. That’s MJ.

At this stage of the creation game, a prompt jockey would be getting ready. to send in a revised version of the prompt to start anew and hopefully end up closer to the end result in their mind. A V-Roller on the other hand, simply selects the variations from the grid of four and run a new set of variations from it. If none of the results looked even remotely possible, then you can reroll the original prompt to get a new starting grid.

Or, you can also rerun the same prompt in a different version of MJ. She is somewhat schizophrenic. There are currently five versions of her: the current, — v3. Two new variants, — test and — testp. Then there are — v2 and — v1. Most users forget about — v1 and — v2 as they are chasing photorealism. V-rollers will run their queries through every variation possible until they get a variation that has what they want to dig for.

Through a process of selection the V-Roller will be able to sculpt the image from the torrent of information she contains. So, if one of the images has a guitar and you like that, you can run a v-roll which will result in four more images based on that one — thus strengthening the element you desired. Of course, because there are many elements in the image MJ won’t know which one you are seeking and thus will try to give you other things. So, maybe the purple gets too much and drowns out the image or a tree suddenly appears because, well, MJ.

Once a final selection is made, then you have the variants of upscaling. It could be full upscale, or a light upscale. There is a redo upscale and even a resample into yet another variant.

And at the end, you end up with an imagine which was nothing that you were thinking and everything at the same time.