Learn the best way to management the number of kinds generated with “random”
Today, I wish to proceed my collection on Midjourney’s Style Tuner by exploring --style random. This characteristic offers you one other alternative to play with completely different kinds, nevertheless it solely gives a restricted stage of management. Yet, I consider many of the customers aren’t conscious that they will roughly management the number of kinds generated with --style random.
So, I’ll begin with a short introduction to this new characteristic. Then, we’ll experiment with completely different prompts and parameters to find fascinating kinds, and eventually, we’ll see how our newly found kinds will play together with new prompts.
If you aren’t acquainted but with Midjourney’s Style Tuner, begin with my first publish on this characteristic as a result of, on this piece, I’ll assume you recognize the fundamentals.
So, for those who don’t wish to spend extra GPU credit on creating your Style Tuners or for those who simply wish to experiment with some new fascinating kinds, with none explicit aesthetics in thoughts, you possibly can have numerous enjoyable with --style random.
Basically, whenever you sort --style random on the finish of your immediate, Midjourney replaces “random” with a mode code. But this code doesn’t have to be completely random.
As you most likely bear in mind, whenever you create your Style Tuners, you select the variety of visible instructions to discover (i.e., 16, 32, 64, or 128). Then, you select a sure variety of favourite photos among the many generated samples. If you select just some choices, your model will likely be very daring and particular. If you select quite a lot of samples, the model will likely be much less distinct.
When utilizing --style random, you too can outline the variety of visible instructions to discover. The extra visible instructions in play, the bigger the pool of attainable kinds to discover. Furthermore, you possibly can outline the share of visible instructions to be activated, thus controlling how distinct your “random” model is. To management this stuff, you’ll have to outline the model within the following format: --style random-length-percentage.
For instance, for those who specify --style random-128-10, this implies that you really want 128 visible instructions to be explored…
