A journey into midjourney (image generation)

I couldn't imagine a creative process to be so much fun – I usually have difficulty visualising my thoughts. Still, midjourney is an excellent example of how text can become visual in seconds.

So let's start prompting, shall we?

Of course, I started with our dog because what else is essential in life? (see input picture further below)

![[5a55f7f78a3a04551bafb48bcefdd1a7_MD5.jpg]]

jetsons cartoon

And since my girlfriend works at a tour guide company that goes through Berlin by bike...

![[13e2d16240da9215a5000101375b9d2e_MD5.jpg]]

a modern logo for a tourguide company in berlin called "Berlin on bike",comic style,fresh

And asked for more variants on option #4:

![[51a4768fed32d46daf70a951aeed99e8_MD5.jpg]]

I've also done a few variations on pictures from my girlfriend:

![[9259c80cb9da7e4649080fbdba4b9a3a_MD5.jpg]]

![[3b6b84c5ca51721f7c0b6109d9f0b5f2_MD5.jpg]]

...and found the outcomes mediocre at best, to be honest.

Finally, I had to put myself to the test, mainly using the images/avatar you might know from my site / social media to inspire midjourney to make some crazy attempts.

...to turn me into an icon

![[937f49353a46bd4a8c2c004f77523b33_MD5.jpg]]

as an icon,abstract,modern,tech

...or a logo...

![[741bc779ea4544e43795fc11d46c25f1_MD5.jpg]]

...or maybe a superhero?... (the model has some serious bias issues btw)

![[ae54463097cf097e0249738a105d3365_MD5.jpg]]

...but the cartoon/funny style made the best impression...

![[f4ebdfaa1938acf558c9843854220dc1_MD5.jpg]]

...which eventually led me into the dalli rabbit hole, ending up with this:

![[38187467a01cd864a8b8114b095f9f04_MD5.jpg]]

So one of the nicest results might be something like this:

![[e7347db9439d887f942b942996967152_MD5.jpg]]

group of people enjoying their bikeride in Berlin Germany, anime

(even though there's not much enjoyment to be seen)

Many things happen in a short time, which means there is v5 of midjourney.

[!NOTE] midjourney v5 This model has very high Coherency, excels at interpreting natural language prompts, is higher resolution, and supports advanced features like repeating patterns

And well let's start with the prompt result I ended with in v4:

![[8cb2cea2714923fcbe3df476f9871ef0_MD5.jpg]]

as you can already see, the difference is quite huge. It seems the model has a different understanding of 'anime', and I potentially also primed it less well. So I made it a bit more clear:

![[aeca9d6a027496b81fbb6dc698c33069_MD5.jpg]]

group of people enjoying their bikeride in Berlin Germany, anime cartoon style

and this already looks much more like a result that is usable (with some artifacts around the bikes) with some attention.

v5 creates some interesting (although fitting in a way) results when you lack decent prompts:

![[1fbb42e2bfc9bc8855d4a3868467c715_MD5.jpg]]

cartoon style -s 1000

(using a picture of our dog as input)

With a single adjustment in the prompt, the result is a lot cooler:

![[4df7af69a67d39b337a60011af4f1ee4_MD5.jpg]]

cartoon style dog --s 250

And adding some imagination is the strong suit of these models:

![[0fb3da52cdbc927b4230365e4304707a_MD5.jpg]]

cartoon style dog in the snow --s 500

For anyone asking, here's the original image I used to prompt the dog generation:

![[bdc41122cd0e7530326d02f4519f101a_MD5.jpg]]

I love science fiction, been a great fan of everything cyberpunk and for some reason gives me calm and some form of future perspective I both love and loathe.

So a great inspiration for some prompting!

![[bd29245c108c62538a09a2e353064efc_MD5.jpg]]

Berlin's brandenburger tor in cyberpunk style in 2222, late in the evening. It's raining, there's little people there. --s 1000 --ar 3:2 --v 5

![[a163e9a4b196862f8c6dfba1460d0515_MD5.jpg]]

Berlin in cyberpunk style in 2222, early in the morning. The brandenburger tor in the background. It's raining, there's little people there. --s 1000

![[948eb4b593c7381ce5e520d9882770c1_MD5.jpg]]

Berlin in cyberpunk style in 2222, early in the morning, in the distance a hovertrain is leaving the train station. It's raining, there's little people there. --s 1000

Sometimes less is more, since in the last prompt it had too much train and too little Berlin.

So one more without Berlin (that I really liked):

![[717e5c2cbd4bbc7456cec8392f035fea_MD5.jpg]]

a first person perspective from a cyberpunk city in 2222, late in the evening in neon lights, lots of people on the street photorealistic --s 1000

I still had to try this, since well – this is something a lot of people would want to do (and again used my regular avatar image to do so)

![[6ac07208ac6072acc8d69ab0ea3f8dc0_MD5.jpg]]

![[9584077b1a9437b3535b788d88fa112f_MD5.jpg]]

![[d101538d172ad5fd45fbae2c859ee213_MD5.jpg]]

as a cartoon logo, futuristic but subtle, imaginative but down to earth. Inquisitive --s 1000

the last one is quite funny!

So I'm looking for a good look for me in a suit. I found an image on instagram I loved, added my own and prompted for result.

![[666a9b2dab7f0f1e88ed761622aa61e9_MD5.jpg]]

![[c6b6725ab69a6c2dea0e567bff1ef294_MD5.jpg]]

combine these images so the red guy is in the grey suit, photorealistic --s 1000

the result is actually hilarious and could well be used for some kind of ad campaign for a suit (without a shoot!) but well, not completely what I was looking for 😆

So stepping away from reality / realism, I found a prompt like this to be fascinating:

a group of people being in venice in the 17th century, taking a picture on top of a bridge, selfie style, enjoying themselves, men and women combined watercolor painting van gogh style --no phone --s 1000

![[21bf653c7706e6f48039570c69ebcd2f_MD5.jpg]]

I will not go into detail too much, just watch this video:

https://www.youtube.com/watch?v=Asg1e_IYzR8

the results are quite stunning:

![[ef2ec19d00a84fb75505adb0b53542d8_MD5.jpg]]

![[bf23cbfb8480cde9324bb86cd6a9937e_MD5.jpg]]

![[d6f020029f804415c72249f1dde915bb_MD5.jpg]]

![[93f9ed91ec5d7effc6b916247ac59297_MD5.jpg]]

![[a44a0cef1100925999f6c7469eabc5bf_MD5.jpg]]

![[3cb7067a4a767dbdf211b155f9690e07_MD5.jpg]]

![[4df0015a30a2c5ad5f79ef484211fe10_MD5.jpg]]

![[107136ee8aba29fffc897fddc3686515_MD5.jpg]]

please note this is still the alpha version in use (release notes) By the end of December 2023, midjourney introduced the alpha version of V6. It's very impressive, stunning to say the least! Let's show you a few of my previous examples in this new iteration of midjourney so you (hopefully) see what I see as well.

Let's start with the bridge example of V5 that I used previously and see what happens:

a group of people being in venice in the 17th century, taking a picture on top of a bridge, selfie style, enjoying themselves, men and women combined watercolor painting van gogh style --no phone --s 1000

![[92e5acb947d7b0c8c4bf01d72d8a7297_MD5.jpg]]

this is pretty awesome right? The examples still need a bit of work, but I really like how the secon and third are looking right now.

The model is still in the works, and it shows - there are quite a few artifacts, but the results are so much nicer and better looking, with a lot less 'prompting' to be done, see the following:

a first person perspective from a cyberpunk city in 2222, late in the evening in neon lights, lots of people on the street photorealistic --s 1000

![[357aa7b2aa9e64fdebf27cc9322d1b7d_MD5.jpg]]

The results are simply way more photorealistic, it displays the imagination a bit more and it feels more 'real' if you catch my drift.

Now let's use the example of our dog that we used before to consider image 'reading' instead. You remember her from the V5 right?

cartoon style

![[04f52bfdcd0f1ea62e9c3a3eff1b41ed_MD5.jpg]]

even though the result looks more 'realistic', this is not what I wanted. So again we need a bit of finetuning of the prompt:

cartoon style dog --s 250

![[840a4380c2b874140b7dff341c7b3a4f_MD5.jpg]]

it's still far away from my 'original' image but I like the outcomes a lot better 🙂

Now for my favourite topic, combining Berlin and Cyberpunk:

Berlin's brandenburger tor in cyberpunk style in 2222, late in the evening. It's raining, there's little people there. --s 1000 --ar 3:2

![[dfcfdd36df0a5ebda618f89c58d7e239_MD5.jpg]]

Again there's so much more detail, sharpness and realism to the image that I almost believe this is the real Tor in 2222.

another variation brought me this:

![[f73aaec830f511115492101956e95f0c_MD5.jpg]]

Let's take a look at what MJ makes out of my own image when I try to make a logo out of it:

as a cartoon logo, futuristic but subtle, imaginative but down to earth. Inquisitive

![[941e8e46f3525f5d9800154e3e76e94d_MD5.jpg]]

What strikes me is that the images make more sense, there's just less nonesense on there. I choose the style raw on it, if I put it back to style 1000 it looks a bit like this:

![[638617c77d18ade199dc16f5ebdc8441_MD5.jpg]]

This just doesn't work that well, although I understand where the model gets the ideas from.

I was really stunned when I started using chatGPT with MJ v5 to create extensive prompts for midjourney, but guess what, that's no longer needed! You can create stunning, photorealistic results with very 'basic' prompts:

A female influencer from the 1930's

![[33a8dda045b9cfbdaf07e7796debb7ee_MD5.jpg]]

![[86fa49c715238ca521c0fe83d033bf84_MD5.jpg]]

One exceptional feature of V6 is the ability to 'vary' mild or strong based on one of the images generated. Let's do a strong variation on the 4th image for instance:

![[6c209296c8e9157c50027cfaa001d0fd_MD5.jpg]]

A red haired viking getting ready for action

![[b767a93b39eee17f85ca76a074c61b3b_MD5.jpg]]

A red haired cyberpunk guy wearing really awkward glasses

![[cc48fe9fc5ce75812c15419052ddc59b_MD5.jpg]]

With a new alpha released recently, performance of 'text' has improved somewhat. a

A funky cyberpunk logo for "Casey"

![[dd266dbf0db7da30258781a2fe2e32bf_MD5.jpg]]

My idea of 'image manipulation' isn't the critical thought behind it – midjourney is good in generating things, not adapting stuff you might have in your head.

Using it to improve your wedding pictures or turning things into a logo isn't the right way to look at it (yet).

It is there to create new things from already existing ones in their model. Of course, you could train the model, and I've seen successful blends of famous persons known to the model in new environments (which is a deep fake territory, and against the ToS).

I'll keep this post in progress while creating more stuff. I've done a few really cool logo's (from scratch) and hope to find more reasons to use it!

Why hands suck in AI (for now)

https://www.youtube.com/watch?v=24yjRbBah3w