I've been comparing image and video generation models for a few years now. I typically use the same prompts to compare quality, adherence to the request, and most oddly enough legality of the use of the image. Typically using anything Mickey Mouse triggers fails for any AI company big enough to be sued or have a legal team.

What is disappointing about this MidJourney 7 image product is that I had to train my preferences by rating 200 images (which took 7 minutes), and then it created some of the crappiest versions of my default image prompts that I've seen. The adherence to what I specifically want, even if complicated, is poor with the MidJourney 7 product. Look at the Messi image. I've wanted a stain glass window style for two years, and this model has failed to deliver.

I couldn't believe how poor the results were after spending 7 minutes rating 200 images to train my preferences. The quality is nowhere near what I expected from a new version release.

Castle Generation

Neuschwanstein castle, lightning, pixar style, volumetric lighting, unreal engine, hyper realistic, hyper detailed, maximum details, photorealistic, 8k, rimlight

Family Scene

Family on their laptops while sitting around a Christmas tree with presents underneath and looking worried because they have to finish up work, realistic, 4k

Evil Mickey

Evil mickey mouse taking a selfie in Disneyland surrounded by shocked families, realistic, 70s style polariod, 8k

Messi Generation

Lionel messi wearing his Argentina uniform floating in the air with beams of light behind him posed like Jesus. Sunrise breaking behind him and a soft halo behind his head, in the style of a gothic stained glass window of a church, volumetric lighting, unreal engine, hyper realistic, hyper detailed, maximum details, photorealistic, 8k, rimlight, maximum details

Robot Portrait

rusty robot with bow tie, portrait, 8k, ultra realism, chrome background

MidJourney's latest release, the version 7 model, claims to enhance image generation capabilities through user preference training. However, after spending 7 minutes rating 200 images to train my preferences, the results were disappointing. The model failed to deliver the quality and adherence to prompts that I expected.

To evaluate MidJourney 7's performance, I tested a series of consistent prompts that I've used with other AI image generators. The results demonstrate a regression in detail accuracy, contextual understanding, and creative interpretation compared to what I've seen with other platforms. Particularly disappointing is MidJourney 7's inability to maintain stylistic coherence and accurately follow complex prompts.

Overall, MidJourney 7 fails to deliver on its promise of enhanced visual generation through preference training. The time investment required to train preferences doesn't yield proportional improvements in output quality. Whether for generating concept art, marketing materials, or creative visual content, this new version represents a step backward rather than the leap forward that was promised.