Midjourney v6-alpha

Midjourney released v6 alpha early this morning. This new model boasts more accurate prompt following and longer prompts, improved coherence and knowledge, minor text drawing ability, different and improved upscalers, and a few more features. Not every prompting feature is fully supported yet, and the devs have warned that we will have to relearn how to prompt this model. I explored these new features a little using some old prompts from previous posts and some new ideas.

Julius Caesar riding a mini motorcycle

anger watercolor

a robot dressed as a medieval knight drinking a goblet

offroad dirtbikes in the desert

gnome carrying fruit comic style illustration

An Empty Castle Landscape

a rainy day at the acropolis inscribed with the roman were here

oranges and persimmons

oranges and persimmons photorealistic 8k

a robot holding a neon sign with the words Will work for electrons

Creative vs Subtle upscaling

Creative

Subtle

Text drawing is definitely more of a best effort than an exact thing. Subtle upscaling keeps a lot more of the features of the original image than Creative, but both keep the composition well. The devs called photorealistic, 8k, and similar modifiers "junk", but they still have an effect on the prompt. I'll have to keep poking at this to figure out prompting differences.