Throwing The Ball:
Getting AI text to image generation using Midjourney is like throwing a ball for a dog in a junkyard.
Just as a dog will come back with lots of different items when you throw a ball for it to fetch in a junkyard, AI text to image generation using Midjourney will also generate a wide variety of images based on the text input.
However, just as you have to keep throwing the ball until the dog brings the right item back, you may also need to keep adjusting and fine-tuning the text input in order to get the AI to generate the desired image. The key to success is understanding how the tool works and being willing to experiment with different inputs and settings. With a bit of patience and persistence, you can teach the AI to "fetch" the right image for you.
This is where the junkyard analogy comes in - just as a dog might bring back a random piece of junk instead of the ball, the AI may generate an image that is not quite what you wanted. But just like throwing the ball again, you can adjust the text input and try again.
Quick Midjourney Prompting Guide For Those Already Using Midjourney:
Prompt 1 - Create Your Base Prompts For The Image You Have In Mind:
This is the image I wanted for this blog post, so here was my initial prompt which generated one image I could use (top left), but I wanted to push it a little closer to a dog with a ball in a junkyard rather than a trailer park:
dog fetching the ball::2 junkyard::2 --ar 3:2
Note: I am adding ‘weights’ to keep the ‘levers’ or ‘elements’ of the image consistent, I can then play with these later if I need to.
Prompt 2 - That Didn’t Quite Work For What I Wanted
So I worked on the junkyard, but that just put too much emphasis on the junkyard and the ball and didn’t give me anything I could use:
dog fetching the ball::2 junkyard filled with computers and cars::2 --ar 3:2
Prompt 3 - Emphasising The Ball
I decided to give some more detail for the ball as the AI was struggling with this and added a ‘lever’ for the red ball I decide I wanted (you get ideas as AI throws stuff at you):
dog fetching the ball::2 red ball::2 junkyard filled with computers and cars::2 --ar 3:2
Prompt 4 - Controlling The Size Of The Ball
I wanted the ball to be small, so clarified that and it gave me one useable image (top left).
dog fetching the ball::2 small red ball on the ground::2 junkyard filled with computers and cars::2 --ar 3:2
Prompt 5 - Rolling Again
Now I rolled that prompt again and got another useful image (bottom right).
dog fetching the ball::2 small red ball on the ground::2 junkyard filled with computers and cars::2 --ar 3:2
So now I have the AI giving me some results I like, I would re-roll multiple times to give me more options to choose from.
I would like more computer junk in the background and could continue to work on that, but essentially I am creating an image for my blog, so don’t need to spend too much time nailing down the concept, I was happy I way 80% there, or was I?
Final Advanced Prompt:
Now I also wanted to try something a little more complex, a robot dog fetching the ball, which idea came at the end of the project.
To do this, I had to specify the scene first with a weight and then weight each element in the scene, I also weighted the overall image stronger than the individual levers and used ‘and/separate/in’ commands so that the AI knew to seperate those elements in the first part of the prompt. The AI still wanted to blend them, so rerolling helps it get to the right place sometimes (re-input the prompt to re-roll, don’t use the re-roll function as it doesn’t deliver quality re-rolls in my experience)...
Robot dog and separate ball in junkyard::3 robot dog::2 small red ball on the ground::2 junkyard filled with computers and cars::2 --ar 3:2
Prompt Guide PDF:
Here is a prompt guide for this exercise that you can download as a pdf and print:
Final Images:
And here are the final images (I edited), that I could use, one I will use now and two I will save for a different project.
Creating Stock Images:
These additional images are now a stock image in my own library and this is the power of this type of image generation, you can build a catalogue of stock images for your marketing / business / projects…
In conclusion, generating images from text using Midjourney is a powerful tool, but it can also be a bit like a game of fetch in a junkyard. With a bit of experimentation and persistence, you can teach the AI to "fetch" the right image for you.
Hope that helps!