Website Images & Content ©2024 Robb Schmidt / All Rights Reserved

From the "X" Files


Creating A.I. artwork from your own Source Image.

Using 'visual prompts' when making A.I. artwork is a powerful tool to get what you want, especially if the source image was created by you to begin with!

For the most part, when I create an image in AI I do it from a written prompt. Sure, on occasion, I have uploaded a source image for the AI to work off of, but my main focus has always been on the prompts I write.

After all, writing a good prompt is really what creating with AI is all about, right? It's our written prompts that are the true 'art', right?

Well, I have (finally) figured out that you can get far more 'artistic' when adding images to your written prompts.

Visual Prompts

That's what I call them, because that's what they actually are! When you upload an image to the AI program, you are showing the AI what you want it to create. Just like a written prompt that tells the AI your instructions, a visual prompt shows the AI your instructions. Together, written and visual prompts can make some amazing images.

It's taken me more than a year of creating with AI to come up with what I'm about to show you. I have no idea WHY this idea didn't come to me sooner, as I know other folks use their own images as source material, (especially their photos), and I have been editing my AI artwork for several months now as well. But, as the old saying goes, better late than never.

I'm sure other people are already doing this on their own, but I thought I'd document it here for your viewing pleasure, in case like me, you hadn't thought of it yet either.

It suddenly dawned on me that I could create a crude collage of an ENTIRE picture and then use it as a source image in an AI model. Then I could have the AI totally clean it up for me! I could just use images I find on Google, throw them all together in a cut and paste composition, then have the AI tidy it all up for me, while retaining my original composition.

This way the AI would be making a picture that I created from scratch, (well, if you don't count that I didn't actually make the images I used to create the collage), and I'd get a picture back that I had envisioned from my mind.

Why didn't I think of this sooner?!

Make a Collage

I was laying in bed last night when I thought of this, and I could hardly wait to get up to try it! I decided to make my collage on a solid black background, as I knew it would be easy to find images on Google with black backgrounds, and that I could easily cut and paste them into my 'home made' creation. I made a blank black canvas in the dimensions I was going to create my AI piece in.

Then I went image hunting to see what I could find that interested me. Here's what I found...

A dark night time scene of a waterside downtown landscape filled with many lit skyscrapers. The sun can be seen coming up at the top of the picture.

A dark moonlit cloudy sky

A dark night time scene of a waterside downtown landscape filled with many lit skyscrapers. The sun can be seen coming up at the top of the picture.

A dimly lit silhouette of a person

A dark night time scene of a waterside downtown landscape filled with many lit skyscrapers. The sun can be seen coming up at the top of the picture.

An empty picture frame

A dark night time scene of a waterside downtown landscape filled with many lit skyscrapers. The sun can be seen coming up at the top of the picture.

A picture of some daisies

A dark night time scene of a waterside downtown landscape filled with many lit skyscrapers. The sun can be seen coming up at the top of the picture.

A picture of a single daisy

I wasn't planning on making some sort of 'masterpiece', or anything, I just wanted to create a quick collage so I could see what the AI made from it.

I crudely threw all 5 of the images together on my canvas, and here is the collage I created from them...

A dark night time scene of a waterside downtown landscape filled with many lit skyscrapers. The sun can be seen coming up at the top of the picture.

A collage of 'borrowed' images

Using your Collage with A.I.

So now the real fun begins, or maybe aggravation, I'll let you decide should you find yourself here.

After uploading your image to the AI, you need to tweak the buttons there a bit to get something you want. This is usually a slide scale or percentage number that you can fill in to tell the AI how much it should follow your image when it creates from it.

You need to play with this a bit to find the percentage point you're looking for. A high percentage will tell the AI to stick closely to the image you've uploaded into it, while low percentages will allow the AI to stray further away from the original image and make more of its own decisions with the composition.

I actually quite liked my collage, so I wanted the main themes of it to remain the same, but I also wanted the AI to have a bit of its own 'creative freedom' to play with it a bit. My goal was to keep my picture pretty much the same, but have the AI turn it into an 'original' for me, so it wasn't obvious that I had created it from random online images. I used very high values in the percentages I told the A.I. to create with. Depending on the specific render, I was always in the mid 60's to 80's.

I would do a render at one percentage point, then try another with a different input and then compare them. When the AI made a render I really liked, I'd download that image and then use it as the source image, replacing my original collage.

It took a bit of time, and probably about 15 renders or so, but I finally ended up with the following image...

A dark night time scene of a waterside downtown landscape filled with many lit skyscrapers. The sun can be seen coming up at the top of the picture.

"My Daisy Obsession"

Side by side Comparison

I'm actually super pleased with the outcome I eventually arrived at. My composition from the borrowed images is still there, but those images I borrowed are not!

The AI did an amazing job of cleaning it all up for me and making the images I used go away....

A dark night time scene of a waterside downtown landscape filled with many lit skyscrapers. The sun can be seen coming up at the top of the picture.

Original collage

A dark night time scene of a waterside downtown landscape filled with many lit skyscrapers. The sun can be seen coming up at the top of the picture.

Finished A.I. creation

The AI gave me a brand new sky, improved it as far as I am concerned. I have a better and more detailed sky with hints of blue, and a new tiny moon which I prefer.

Another moon, (or maybe a sun?), now appears in my fancy new picture frame of daisies, which blend nicely into a new daisy field that the A.I. was kind enough to create for me!

And although the male figure is in the same position as the collage figure is, he is now very detailed and far less cartoon-like, he has hair, even on his arms, and there is moonlight shining through his ears! He is very much new and improved, and even that big ol' daisy on his face has been enhanced nicely.

In Conclusion

Visual prompts are an AMAZING way of communicating your wishes to the AI, and as you can see from this simple demonstration, you can get some pretty good results when using your own creations.

Whether you want to make your own personalized image, like I did here, or just want to take control of the composition, (a person here and another there, positive and negative spaces, etc.), uploading images can give you what you want.

I will certainly be utilizing this 'image to image' method much more often now that I know how well it worX!

Imagine if I had really worked on my collage and made something with colors and more meaning. Well, you probably won't have to imagine for too long, as I suspect I'll be doing that SOON!

Another great thing about making your own visual prompts/images is that you really feel like the artwork is yours, much more so than just worded prompts can provide to you.

Anyway, I hope you found this interesting, and if you haven't done so already, I really hope you give this a try. It's fun, and ohhhh so satisfying!

Good luck & have fun,

"X"

February 4th, 2024

(This was PART ONE of a two part series that shows how you can get the A.I. to follow your image closely. PART TWO shows what can happen when you give the A.I. a LOT of artistic freedom. You can see that HERE.)