Midjourney 5.1 has been launched, bringing one other important enchancment within the high quality of outcomes from the generative AI artwork service.
The corporate claims that model 5.1 of the engine is “extra opinionated”, bringing it nearer to the form of outcomes that you’d get with model 4 of Midjourney, however at a better high quality. There’s additionally a “uncooked” mode, for individuals who don’t need photographs which are as strongly opinionated.
Different claimed enhancements embrace higher accuracy, fewer undesirable borders or textual content artifacts in photographs, and improved sharpness.
Testing Midjourney 5.1
I put the brand new model of Midjourney to the take a look at through the use of the identical prompts in model 5 and model 5.1 and evaluating the outcomes. All different settings have been left the identical, with Midjourney set to base high quality and medium type.
In all the pictures under, the 4 photographs on the left are Midjourney 5 and the 4 photographs on the correct are model 5.1.
Right here the immediate was “girl strolling down a cobbled road into the sundown”:
There’s been a persistent downside with over-aggressive, artificial-looking lens flare in Midjourney 5, so it’s good to see that being addressed with model 5.1. The sundown within the model 5.1 photographs appears extra pure, with hotter, extra fascinating skies.
The second immediate I used was “explosion in a paint manufacturing unit, vector artwork”:
As soon as once more, the pictures in Midjourney 5.1 are a lot stronger than the model 5 equivalents, making higher use of shade and providing extra dynamic compositions. The highest left and backside proper photographs from model 5.1 are significantly punchy.
Faces have turn into an actual energy of Midjourney’s outcomes, so for the subsequent picture I requested the AI to create “shut up of a kid carrying swimming goggles”:
Right here we are able to see what Midjourney means about photographs being extra “opinionated” in model 5.1. The model 5 photographs on the left are high-quality, however very straight, smartphone camera-like snapshots of children carrying goggles. The 5.1 photographs on the correct have a a lot stronger editorial really feel, as in the event that they’ve been taken by knowledgeable photographer. The picture within the backside proper is especially putting.
Lastly, I wished to place the promise of fewer textual content artifacts to the take a look at by asking for “newspaper salesman on a road nook, Nineteen Fifties type, sepia”:
First, it’s notable that each variations misunderstood the immediate, delivering what appears like a salesman with a newspaper moderately than a newspaper salesman, however that would simply be refined with the immediate. Each have textual content within the background, moderately than the newspaper itself, however there’s not quite a lot of distinction between them. The textual content in each units of photographs appears OK from a distance, however once you zoom in shut it’s largely non-sensical. The model 5.1 photographs have stronger compositions, with the topic filling extra of the body.
Easy methods to use Midjourney 5.1
The most recent launch is at present in testing, however Midjourney clients can change to model 5.1 by default with the /settings command in Discord.
The corporate plans to make Midjourney 5.1 the default mode as quickly as the top of this week.