December 2023
Fictional Photography
Using the MidJourney’s Generative Adversarial Networks.
First tests and expirations with Midjourney V6 Alpha.
For this series of test, I used the idea of a common-based universal basic services urban community somewhere in Asia. With the currently limited tools available, I started to recalibrate how this version understands light, textures, moods, space and of course composition.
I am working in medium style settings and reusing some of my main prompts “bases” I rely on in order to get the same ambiences throughout my images, especially for the Codex. Here, I’m particularly impressed by the dust and the mineral surfaces.
I am mostly pleased with how it generates photographic visuals and depth of field. Also, the way it understand composition and contrast is very inspiring. Here as usual, hands can still be challenging for Midjourney. Same with small and distant faces, which often appear glitched.
One incredible improvement of this version seems to be the way the AI makes sense of different ambiances and spaces. The ventilation in the ceiling underlines particularly well the level of detail and realism it’s able to produce.
The bokeh and and different layers of depth is really pleasant. We do not have the regional variations mode yet. I would have fixed that lamp on the right.
Here as well, the composition, the light and the contrast are so interesting. The subject is in focus, while the foreground and background both have a progressive blur, nicely separating the subject from the background.
My prompts in this series rely mainly on white tiles. Here, the composition, the architectural symmetry and the asymmetry with the subject and the lamp is particularly pleasing.
I’m impressed how the generations are capable of producing these tensions between subjects in the background and foreground, using silhouettes.
One radical change with this version is the faces and expressions of people. They are a lot more “normal” and realistic than the fashion models usually generated by default by previous versions.
Again, a silhouette. This image particularly pleases me with the rubber/plastic surface on the table, almost like a patched tablecloth. Again the plates on the shelf below are so plausible. The generation of heterogeneous objects on the edge of the window is also particularly pleasing. I would have fixed that hanging device on the left wall. It looks like aa weird Covid hydroalcoholic gel dispenser.
The AI now understands light very well, here with various surfaces, like water, concrete and wood. I’m very impressed how Midjourney is able to break away from the sort of neat and perfect images found in advertising or architecture magazines to produce realistic level of roughness and materially correct scenery.
Again, people have this new sense of presence and realism. Hands are again better than previously. Yet, here I wonder where did the right hand go.? Not there yet at 100% but definitely improvements.
Composition, rhythm, light and a subject. Again, with the regional variation tool I would fix the plant on the left without the pot, but beside small details like this, images are usually quite usable out of the AI right away.
Playing with light again. Not sure about the shadows here, is it a tree in the foreground in bokeh? Probably. I would use the regional tool to test variations here. Anyway, the subject with the hat and the scarf is so plausible. You almost feel the chill in the room.
Back to matter. In this image I’m testing materials, textures and polygonal volumes. Again, beside the light and shadows, I’m particularly impressed by the mineral textures. I would use the regional variation tool to fix the levitating bowl on the right and the weird twig above the right shoulder of the subject.
Here again playing with the generations of large amounts of heterogenous objects on shelves, using pottery related prompts, which work much better than dishes related prompts.
On the left against the wall is a perfect example of the type of weird things the AI can produce. Is it a table, a shelf? How does it hold? Does it have legs? It’s an alpha and this will be easy to fix with the full set of tools available. I could load it into Dall-E and fix it. But as usual, my AI images are 100& unretouched.
This bar image with the coffee machine and the chopsticks is impressive. I’m not sure about what is on the ceiling on the right and if the hanging device is a webcam. I love the hooks below the bar. Otherwise, it shows how the AI understands space and light. There is something so plausible about this image, despite its imperfections. Maybe the imperfections have reached subtle a level that match the imperfections of the world.
And finally, another foreground/background play with silhouettes. This is a good example of how the AI understands light, reflections and how to compose a nice ambience shot with some symmetry and rhythm.
If I am currently more waiting for a beta or a full release, this alpha was a real Xmas gift to play with and enjoy the update. I have also tested the new upscale features. Quite impressive. Especially on skin and facial features, huge improvements. I might do a post on this soon.
W&G 2024