First, the basics. What do the developers themselves write about the new Midjourney mode?
- V5.1 is more opinionated, like V4
- V5.1 understands short prompts well
- V5.1 RAW is more neutral, unopinionated
- V5.1 follows text prompts more accurately, has fewer unwanted artifacts and improved sharpness
So, the creative power of V4 is available with the quality of V5? Let's see now. In the review, there are pictures that resulted from the first request, without rerolls.
1.1. Simple prompt
Let's start with a maximally laconic prompt: cat watching the aquarium
What V5 generates without any special clarifications on a regular simple prompt can pass for a simple photo on the phone.
It is immediately clear that even without a request like professional hq photography V5.1 gives a juicy, sometimes even too fabulous picture, especially in terms of colors. It seems that to get an imitation of a mundane photo on the phone, in the new versions you will have to forcibly write cheap camera phone photo or something like that.
RAW so far looks like V5, but with better sharpness and better color correction. The cats are kind of the same. And the aquariums are poorer. The first cat is looking into the abyss, and the abyss looks back at him.
1.2. Detailed prompt
Let's add
siamese cat is watching big beautiful backlit aquarium with jellyfishes, late evening, neon lighting, ultra wide shot, dreamy, modern interior, professional photography, Hasselblad X1D
Here between V5 and V5.1 I don't even see any special differences.
Except that the light is better in V5.1.
RAW, why are you scaring the cat?
There are no jellyfish anywhere, wide shot either. I'm not a fan of long requests, and here the result will definitely be better if you throw something out, but that's another story.
2.1. Surreal!
And now what we have gathered here for: capybara secret society cult performs a mysterious ritual
V5, as usual, tries to include realism. In this case, it's even charming (look at the cultists with ears, upper right) and a little creepy.
Oh yes. Without specifying, the opinionated V5.1 generates both a photo and an illustration. Looks great, although not as photorealistic as...
RAW. Not counting the bugs, in essence, this is exactly what many need: absurdity, not turning into illustrativeness.
2.2. Surreal x2
Capybara sect: too easy. We need to raise the degree of madness: photography of a mouse pilot piloting a plane made of cheese high in the sky
I specifically add photography, because V5 will draw an illustration poorly, but with the photorealism of such requests, everything is usually complicated.
To control the plane without a steering wheel is the highest piloting, as well as surfing on a cheese board, but we will leave the technical side without comments. The fourth mouse looks quite realistic and it has the best headgear.
Let's not notice this color correction, since I didn't specify lighting for the purity of the experiment, and look at the charming suit in the lower left. Nice for a short prompt.
The missing steering wheel emphasizes the closeness of RAW to V5. The construction of the plane is also in question. But cheese ears!!!
By the way, we caught MJ not knowing what kind of teeth rodents have.
3.1. Illustration
It's interesting what a simple request with an illustration will differ: waterfall, illustration
Four watermarks out of four! Dull result. But it's not surprising: without indicating artists/styles/details, V5 doesn't show much artistic ability.
Very nice! V5.1 fulfills the main dream of the customer and without specifying the request offers completely different styles. There is a mini-watermark, but it's nothing compared to the neighbors above and below.
RAW... It's not very visible but there are watermarks here on all pictures, except for the lower left one. Apparently, this is the fate of too simple requests for illustrative primitive themes. Even if I overlook the artifacts, I must admit that I'm not particularly fond of the styles either.
Time will tell If the results are not a coincidence. At least in V5.1 there are almost no watermarks: this is a success!
3.2. A slightly more detailed illustration
waterfall, tigers are chilling in the water, modern realistic illustration
Why the tigers are so bad today is a mystery. Maybe due to the implementation of changes, the quality of generations temporarily drops, and I won't even describe what's wrong here.
Interesting color palette, but the anatomy limps, like most of the tigers. As before, the fewer objects, even if they are the same the better.
Very strange things are happening with the general form of the left illustrations. And overall, not very expressive.
4. In the style...
What if in the style of Van Gogh? waterfall, by Van Gogh
Here, as they say, to each his own. V5 is most similar to painting.
On the verge of illustration, vibrant colors and epic scale. But there was no oil painting in the request, so no complaints.
Again something average. I liked the plasticity of the water most of all in RAW.
5. Photos of people
professional portrait photography
Hello, wax skin and close focal length.
Well, well! I see pores. And the photos are really professional without specifying lenses, etc.
The lower left is close to the level of V5.1, however compositionally sad. The rest are slightly less waxy than V5, and the upper right one was unexpectedly downgraded to the level of V4.
6. Macro
We were promised improved sharpness: let's check on macro shooting.
snowflake, macro photography
Macro V5 is quite authentic and without excessive color correction, but not everywhere symmetrical and cute shape.
V5.1 not very pleasantly surprises with a strange texture of snowflakes, and the color filters are fierce. An artifact in the form of a small watermark slips through.
But RAW pleased me: more even shapes, transparency, light (especially the upper right one).
7. Architecture
apple shaped house, professional architecture photography
Apple-like, but not appealing to the eye.
Better frame composition, pleasing shapes, but filters ruin it.
G
ood colors and angles, one bad thing: it's not an apple.
8. Food
With sushi generations, it has always been not smooth.
sushi rolls set, professional food photography
V5 shows us the classic picture of crooked rolls with generous filling and a bunch of red caviar. This time the surfaceis are not the cleanest either.
There's still a lot of caviar and filling, but the rolls themselves have become smoother. MJ is still far from understanding the principle of wrapping with fish. And the composition looks very cool (for minimalists).
The form of rolls from V5.1, and the style of photography and rich decoration from V5.
Quick conclusions:
V5.1 draws good light and shows itself equally well in the generation of photographs and illustrations, including those going beyond common sense. Loves to add excessive colorfulness, fabulousness, all this atmosphere. On photographs, it adds filters everywhere, so you need to pay attention to the description of lighting, or write --no color grading (be careful, it can give b/w).
V5.1 RAW is similar to the good old V5, the quality is better, the compositions are more successful, sometimes even more realistic in generating photographs than V5.1, agrees to generate nonsense. But artifacts slip through more often, not very creative on its own initiative.
How to enable:
Select in /settings "MJ version 5.1" and optionally "RAW mode", or manually write --v 5.1, --style raw