Series 01 - Planet Exploration
This gallery of generative art was made in MidJourney. So let's dig into our imaginations and see what we find. I like to experiment with various prompt styles. Sometimes less is more. One method I have is what I call "micro" prompting. It's when I use the smallest prompt -usually just one word. I then move onto another similar word and start to build a theme.
The prompt output also depends on the library (LLM data-set) that was chosen in the settings and any variables. In this case I used a 1:1 aspect ratio. Remember the AI will try to fill the canvas whether its wide or tall - so a square aspect ratio tends to deliver uniform size results. If I want the trees taller, I can just make the canvas taller by adjusting the aspect ratio variable like so --ar 9:16.
If I want to try to render a wide tree or maybe get a forest I can use a wide desktop format. The aspect ratio parameter follows Width x Height format. So simply invert the values to --ar 16:9 to make the canvas wide instead of tall.
The micro prompt started with the word "ball"
The next prompt was "planet"
Then "tree-house"
Then "tree-palace"
The AI likes to add people sometimes, even though my prompt did not ask for any. And they are rendered in the same style as the other elements. You will notice this person is rendered inside a ball.
So why do this? data science! Because every word carries a definition or 2 or more, and the AI has it's own interpretation of those definitions measured against the selected LLM and model drift over time. So what I'm really doing is giving the AI room to be creative. I don't want to tell it what to do - I want to see what it has learned.
In some cases I used the zoom out feature. Other cases I drifted on variant images like going down a rabbit hole.
For all selected images to be exported I used the 4x upscale feature which essentially renders the image with more detail - a second generative pass that adds more detail and upscales the image to 4k. The filesize is upscaled too btw.
After all the images are rendered I apply my watermark in another graphics application. I create an action to make the process faster by automating it.
Lastly, I have discovered if you want a gallery in Linked-in you must use a pdf. So either in Canva or Acrobat pro you can assemble your pdf.
Then it becomes just a matter of uploading. Currently I am side tracked in this post when I meant to make a gallery. Happy accident.
领英推荐
You can see that its possible to acheive amazing results with very simple prompting. It may not happen the first prompt, that's why exploring is crucial.
The fun for me is discovering the variations. The collection for this excersize is about 400 images. I am showing less than 10% of the exploration.
If you made it this far - thank you!
Please feel free to leave feedback or comments or suggest a theme that fits into fantasy or sci-fi, and I will consider it.
Have a great day or night.
Follow me on LinkedIn: www.dhirubhai.net/comm/mynetwork/discovery-see-all?usecase=PEOPLE_FOLLOWS&followMember=brian-shurtleff-ux
Sr. UX/UI Designer | User Experience Designer | User Researcher
11 个月Very nicely done Brian!
?? IQaaS ? AI Alchemist ? Fortune 100 Enterprise ? 225k EDM YouTube Subscribers ??
11 个月Just added a bunch of images to the post, feel free to use them as you like... enjoy!