MIDJOURNEY Version Variations

I finally got access to MIDJOURNEY'S new "v 8.1" model, which is now the default model since June 11, 2026.  

I decided to do a test, by altering a photo of an ALEBRIJE sculpture in Albuquerque with ALL the model versions of MIDJOURNEY:

ORIGINAL IMAGE

MORE ALEBRIJES -- Real and Fake

The ORIGINAL image I altered

in all the versions of Midjourney

However the "v 8.1" loses a lot of architectural space -- so I prefer the older "v 7" and "v 6" models.

---

MIDJOURNEY VERSIONS

"v 8.1" back to "v 3"

Results from Models v 8.1 down to v 3,

going backwards in time

NOTE:  I used my Midjourney Moodboard -- "EZ ART#6" (--profile z9nizie) -- to generate the images in versions "v 8.1" through "v 6.1."

"EZ ART#6" Moodboard

(--profile z9nizie)

*****

v 8.1

(June 11, 2026)

Midjourney prompt:

"Low rider passing the Alebrije sculpture --profile z9nizie"

Animation of "v 8.1" variations

---

v 7

(June 17, 2025)

Midjourney prompt:

"Low rider passing the Alebrije sculpture --profile z9nizie"

Animation of "v 7" variations

---

v 6.1 

(July 30, 2024)

Midjourney prompt:

"Low rider passing the Alebrije sculpture --profile z9nizie"

Animation of "v 6.1" variations

---

v 6

(December 21, 2023)

Animation of "v 6" variations

---

v 5.2

(June 22, 2023)

Animation of "v 5.2" variations

---

v 5.1

(May 3, 2023)

Animation of "v 5.1" variations

---

v 5

(March 15, 2023)

Animation of "v 5" variations

---

v 4

(November 5, 2022)

Animation of "v 4" variations

---

v 3 

(July 25, 2022)

Animation of "v 3" variations

---

CONCLUSION

I prefer the older MIDJOURNEY model versions, as the new "v 8.1" model loses a lot of architectural space.

My favorite images below have a lot of SPATIAL qualities, and were generated with "v 6" and "v 7."

BEST IMAGES

 from PREVIOUS MODEL Versions

More -- EZ AI ART

"v 6" image from January 2024

More January AI

"v7" from January, 2025

SURREAL AI Slop

"v7" from March, 2026

Also MIDJOURNEY pushes too hard for me to "train" their model on my personal tastes.  However I have been avoiding this because I am afraid that Midjourney will "OVERFIT" to my tastes, and start generating the same image or flavor no matter what kind of input:

Midjourney pushes too hard

to get me to declare my preferences

Part of the adventure is to generate something I could not imagine.  I concur with what Brian Eno said (in a video that is no longer on YouTube):

"..so the thing becomes an exercise in trying to force the system to do something that is actually against it's nature...because it's nature (ie AI) is to suck up and take the most probable next word...no...I WANT THE LEAST PROBABLE NEXT WORD..."

KREA IMAGES

Last month KREA came out with their own answer to the Midjourney aesthetic with their KREA2 model.  It took me a while to figure out KREA's Moodboards, but eventually I did and generated some styles and images I liked:

KREA 2

ROPE CRITTERS

CLAYTON, New Mexico

Ultimately I think one has to Train models on their own images -- in my case, on my drawings -- in order to get their personal style and unique aesthetic.

MIDJOURNEY MEDICAL

Midjourney ultrasound body scanner:

• (YouTube):AI News: Fable Banned, New Open-Source Leader, Midjourney Shocker