How to Create Concept Art with AI Part II (Mid 2023)

Using tools like Dall-E 2, Dream AI by Wombo, and Photoshop Essentials 2023

Have you ever been faced with creating some detailed concept art to convey ideas to an audience? Do you want to be able to complete multiple pieces in days instead of working on one piece for a week? Are you lacking the tools, skills, or time it takes to make your art and also have it look accurate and visually appealing? In this article I will teach you how to create concept art with AI, as well as provide detailed examples of my own work going through this process.

When it comes to creating concept art, you often need highly detailed images with quality lighting and backgrounds. For character concepts you can usually get by with a sketch, but most times this just isn’t good enough. Sometimes you wish you had the skills, or the necessary equipment to render on the canvas exactly what you see in your mind. For those that are not good at drawing, or don’t have the time, AI can work wonders in this regard.

Update to Previous Article

This article is a follow up to one that I wrote in October of 2022. Click the link below if you are curious.

In today’s article I will attempt to show you how much the technology has progressed since the last article was written. Please refer to the previous article if you would like to compare what was able to be created back then with what can be done now. I assure you that the progress in AI art is astounding, and just by comparing what I will show you today in How to Create Concept Art with AI Part II, you will see firsthand just how much of an improvement you could make to your own concepts if you followed my previous guide or if you are new to generating concept art in general.

Disclaimer:

I do not work for Adobe, Wombo, Dall-E, OpenAI, or any other image generation site. I am not being paid for any links. This post is not sponsored by any of the sites used herein. To that end, this is also not a review of the sites I am using. This article is for education purposes only. I am simply informing people how to use these tools like I have to get similar results. I am not affiliated in any way with these web-based or commercial software. However, I include links to some of these websites for two purposes. To help guide my readers to follow the instructions, and to give credit to these companies whos software I have used.

What you need

Note: This section will differ from the one in the previous article. I have changed it to reflect the software I currently use.

To accomplish this, all you need is your PC and some rudimentary image editing software. Since I already use Photoshop at work and am comfortable with it, I was able to purchase Adobe Photoshop Essentials 2023 for around $60 from Best Buy, but the same techniques can be accomplished using freeware such as GIMP and Inkscape. By the end of this article you will have the skills to create your own unique and detailed concept art for your audience.

Dream AI

Firstly, we will be using Dream AI by Wombo. This AI image generator will do the bulk of the work for you. All you have to do is have the ideas and put it into the right words for the AI to translate. Previously I used Craiyon for image generation. But Dream AI far surpasses it in it’s ability to create authentic and realistic looking images, as well as the fact that it contains several features and subject filters that most image generators lack. While it is not required to pay for the premium version, you will notice that I have done this to unlock some of the features it gives you. With the paid version you can generate 4 images at a time, and you can unlock different aspect ratios for your artwork instead of the stock vertical “trading card” size.

OpenAI

In the previous article I used Dall-E 2 project on OpenAI and while I did make use of it in some instances for my continued work, I feel that because of the credit system and paywall, that I used it so rarely and for minor touch ups that I don’t really need to go into much more detail about it here.

Once you have the desired images generated for you by the AI, you can then edit the image as you so choose with either Photoshop or a freeware substitute. After editing, you then use the image as a reference image. Run it through the AI once more while attempting to secure the desired result.

You’ll see that the steps are relatively simple. Generate an image you can work with. Edit the image. Use it as a reference and run it through the AI again until you get the desired result.

Before we get started, I need to post a spoiler warning. If any of you are currently reading my Web Series eBook, Star Stream then there are minor spoilers below. For any of you that have not read the series and would wish to check it out. Follow the link below.

Getting Started

If you remember from my previous article How to Create Concept Art with AI (Part I) we got started by finding a bunch of faces to add to our AI image outputs. Well I am happy to tell you that you don’t need to do this step anymore. In the rare case that you can’t get a good face, or you want to create your own, you can refer back to that article linked above to add faces. As it stands now, Dream AI is perfectly capable of producing high quality faces, and they can be edited and perfected with the same techniques I will describe in this article.

First Run

For the first piece, I had a particular look in mind. I wanted a sort of space jet that resembled a Leer Jet. For my particular concept, I wanted to go with a certain look and certain colors, Black with a Gold horizontal stripe.

The first thing I did was go to google and find an image of a Leer Jet that resembled the look I was going for.

I scrolled a bit but I eventually found this. It was perfect. It even had a Gold Stripe that I was going for. There were two problems that I’d need to account for. I wanted the Jet to be facing the other way, and I didn’t want a blue sky in my AI output. So I loaded the above photo into Photoshop Essentials (PSE), flipped the image, and changed the hue of the sky. Very minor tweaks that are easily accomplished.

Not bad. Very close to what I had envisioned, but now I need to run it through Dream.

The Mobile App Aspect Ratio Issue

Note: If you do not have a paid account, you will be locked into a vertical aspect ratio of 9:16 for your outputs. Furthermore, you must use the mobile app to unlock other aspect ratios.

This is my biggest complaint with the current state of Dream by Wombo. Because I have to send the photos to my phone, run them through AI, then send them back to my PC for editing. It is such a tedious process. I more often than not stick with the 9:16 aspect ratio on my PC unless I absolutely want something in a widescreen or other format.

Keeping this in mind, I will show you how to edit a reference image so that it fits that ratio. For the rest of the article I will assume you either adapted your reference image with this technique, or are able to access your chosen image size with your account.

Image Prep

The first thing you need to do in photoshop is change the canvas size of your photo to a 9:16 (or whatever ratio you choose in the Dream App) aspect ratio.

Once you have the photo altered, you will be left with blank white spaces depending on your chosen ratio. For cases such as this, we simply use one of two techniques in PSE (Photoshop Essentials 2023).

The first option is to simply use the “medicine dropper” to choose a color already in the background. For mine it was the pinkish color of the sky. Next, use the paint bucket tool to fill in the white area at the top. That I was I did in this particular edit.

The second technique is just a little bit more involved. In PSE you can use the clone brush and the healing brush to accomplish this. In the photo above, I cloned some of the pavement. Then I filled in the white area with it. And then I used the healing brush to smooth it out.

There is a reason for filling in the blank areas. If you don’t Dream will output an image with white at the top and bottom. If you have any empty space in a reference photo it will do this. It is usually best to try and fill that area or blend it.

Ok So lets head over to Dream.

Dream by Wombo

The basic steps

Getting familiar

There are three basic things to do once you get here. First, you’ll want to go ahead and scroll down and upload your input image.

Next, You’ll want to scroll back to the top and enter a prompt. Keep in mind that your prompt is basically what you want the output image to be, using your reference image. You cannot tell it to make changes to the image, that feature is currently in beta, so you want your prompt to say what the overall image should be, and not describing alterations to the reference image.

After typing in your prompt, select your Art Style. There are several of these to choose from, and they all significantly alter what your results will look like. I suggest playing around with each individual one and using trial and error to see which ones will ultimately suit the style you are going for. In my case, I chose VFX v2.

Once all of your variables are chosen, simply click the “create” button. As you can see above, I have the influence of my reference image set to “normal” but I want to get as close to a Leer Jet look as possible given the prompt. As you can see here, the outputs were great, but not close enough to the reference. What I will do in this case is adjust the slider under my reference image from “normal” to “strong” and I will also change the Art Style to Dreamland V2. I click Create.

In this instance I get pretty lucky. I immediately get results in the first 2 or 3 tries. I’m happy with the outputs, and they almost exactly match what I had envisioned in my mind for this particular artwork. Sometimes that is not the case. Sometimes you will need to make slight or significant alterations as you go along. In those cases, simply play around with more Art Styles and edit your prompt as you go along to add more detail. As you will soon see, sometimes you can keep going and going until you are finally satisfied with the output.

Evolving an Output

Something peculiar happened when I was working on this particular piece. I chose an Art Style to see what the outputs were, and while it didn’t match at all what I was going for, it happened to be similar to something else I wanted to make Concept Art for. When I chose the Art Style “Nightly v2” it generated the following:

The first thing you notice right off the bat is that it looks more like a space station than a fighter jet. But as soon as I saw this I knew I could work with it. I needed a space station as well, and now I have something to base it on. I saved the above image, then set it as my reference image in Dream with a normal influence setting. Then changed my prompt to “Silver Space Station” and clicked create.

Stage 2

Further Editing with PSE

Once I got the output I was satisfied with (the last one above) I knew it would need some tweaks in PSE before running it through Dream again as a reference. So I loaded up the image into Photoshop. The first thing I did was click the “selection tool” and then clicked “select subject” after that I made it into it’s own layer and deleted the background layer, leaving me with this.

The first Edit that I wanted to make was with color. All of the lines of orange accent lighting on the station needed to go. I wanted those to be blue. Once again I used the selection tool for all the orange parts and cut them to a new layer. Then, after I did that I changed the Hue from orange to the Blue color I was going for.

Once I make this change, there is an additional change that I’d like to make. I don’t like the position of the “saucer” on the top right. So I select the entire saucer and cut it to a layer, then move it.

Now that I am happy with the look of the station. I want to add a background to it so that Dream will interpret the station in space. For that I just go to Dream, and I enter a prompt for an outer space background. When I am satisfied with a result, I save it, then load it into PSE.

The last step is placing the station on top of the background in PSE and saving the image to load into Dream.

The Following Stages

The first stage was generating the appropriate reference image. The Second stage was altering the image and adding a background in PSE, and the 3rd stage will be the following results. For a couple of stages, I simply changed the reference image to a chosen previously generated output, without editing. As we move on I will denote those stages. In each stage I do things like alter the prompt, or alter the influence of the reference image, and several outputs will be from different chosen Art Styles. I’ve already gone over how to change these variables, so for the sake of moving on, I will omit those extra details in the following stages because they have already been explained, and because this is a guide, I don’t want to bore you with my particular choices. Yours may be different entirely.

Stage 3

Overall I was happy with the outputs, but there was still a lot of things I didn’t like in some of the images generated. At this point, I liked the final output above and decided to change my reference image to it and go from there.

Stage 4

Using the Final output of Stage 3 I generated these outputs for Stage 4.

Happy with the 4th output I decided it needed something more, so I saved it and brought it into Photoshop.

By using the selection tool, I enlarged the left portion of the station (Layer 2), then I also enlarged and adjusted the position of the right portion (Layer 1) of the station. I also wanted a large blue hangar door on Layer 1 to the right, so I used the clone stamp to clone some bright blue from the central region, and then I used the healing brush to erase the bright blue light from the central region when I was done using it. Now back to loading it into Dream.

Stage 5

The results were promising. I liked how the hangar doors came out, and the positions of the two sides, but I wasn’t happy with the left side coming to a point. Like how it looked like a fish. So I went back into Photoshop and edited the same project that I had loaded before, only this time I added a tubular effect to the “nose” of the left side.

Stage 6

The effort paid off, but I still wasn’t satisfied. The left side was looking more tubular, but I think I was wanting the effect to be a bit bigger. So I went back into photoshop and increased the size of the hole there.

Stage 7

And here we are. I ultimately saw the final output of Stage 7 and knew instantly that it corresponded with the vision in my mind. Through trial and error, on this particular piece, it only took me 7 stages to get the desired result.

Following this technique of editing and re-referencing, and changing factors such as Art Style and Influence, you can make just about anything that you envision in your mind.

Now that we’ve gone over the techniques used to achieve artwork that matches what is in your minds eye, I’d like to show you some end results for this process. For the following entries I will call back to Concept art from my previous Article on this subject. I will then show you end results that I was able to achieve with the above process, and you will certainly see that the results are far far better than what was achievable only around 6 months ago.

Keep in mind that for every end result you see, there may have been tens or hundreds of outputs that were “off” enough for me to discard before finally getting the desired output. Also remember that throughout the process, there are numerous times where I take an output, edit it a little bit, and then run it back through the AI generator. If you notice a drastic change to a color of an object, or overall style of the output, it is usually because I made edits and re-submitted reference art, or changed the Art Style, or Reference influence.

Even slightly tweaking words in the prompt can have sizeable effects. But, for the most part, the AI will follow very closely to what is presented in the reference image. To make it simple, sometimes the AI will not distribute colors or decorative items exactly like it should or how you want. In these cases it is far more beneficial to just edit those parts yourself and then re-submit the reference.

Starting with the Head

For my first order of business, I took a head that I had created for my original article. Back then, I simply added some details with Photoshop and then pasted the head on bodies that Craiyon had generated. Back then, the AI was terrible at making faces, so I had to do all the work myself with PSE. Even still, the faces sort of stood out as being “pasted on”. But now, I can use this face as a reference image. Then have the Dream AI generate outputs based on my prompts.

Results

Amazing isn’t it. Compared to the last effort, less than a year ago, now we can make faces as stunning as this. When the time comes, I will add this head, or a variation of it, to a perfect output body. And that is only if I choose to replace the head already on the output.

Also, lets say you like one of these outputs, but you need a different ethnicity. For my case, I had a character in mind that wore the same style armor as this guy, but he was of African decent. In that case, I simply loaded as a reference one of the images above and typed in a prompt indicating the desired ethnicity. I then loaded the output to photoshop, deleted the background and changed the hue of the lenses to the desired color, purple, then put it back in Dream as a reference.

It really was that simple.

Now that I’ve shown you what is possible with just the faces, I’d like to move on to some before and after comparisons of some of the characters I made for the last article. I’ll post pictures of what was in the original article and then show some of the final results next to it.

Note: If you are on Mobile, the images may appear above and below instead of left and right.

Side by Side

Old vs New

Right off the bat you can see the similarities but the overwhelming change to level of detail in the new and improved version. It went from being a dream like concept of a vision in my head in the old go round to an actual honest to goodness minds eye representation in the new.

Still we are seeing the level of detail just surpass the old in every way.

With this one you can see two of the final variations. They add a ton of realism. It’s also overall sharper and clearer instead of a rudimentary concept.

This shows a progression from an output, with a head added, to having the colors adjusted with PSE and ethnicity adjusted in Dream.

In this instance I decided to use the old reference image for an entirely new character. The dreamscape Art Style added the exact flair I was going for with this character.

Just by looking at the new output you can see the drastic improvement in the level of detail. Next, I took this one even further. I decided to make a custom face (Like the ones I made above) to some of the better results.

Taking it even further, I decided to keep some of the variants that did not have the mask. I simply kept the original output without changing the face.

Now look again at the new output vs the old concept art. If you compare these images to the ones in the previous article it is obvious they are superior.

Bonus

Additional Characters

The next section is just a bonus section showing variations of additional characters I have since created using Dream AI. I don’t have any previous concepts of these characters to compare them to.

Catra

A new character. I wanted to show how many variations I was able to put together in a relatively short time frame. I usually do this with most characters, while periodically eliminating ones that just don’t look right. That leaves me with about the number of concepts above for any particular character.

Rafe

The Space Pirate Rafe from the Magma Dredge. Read about him in the earlier chapters.

Haller

Captain Haller also appears in the same chapters as Catra and Rafe. Note the Silver and Blue of Cernova.

Drago V

One of the main antagonists. He is the brother of Catra (Contessa Caterina). Because of their Cernova family origin, they dress in chrome reflective silver and navy blue.

Count Dragomir IV

He is of the same family and Drago V’s father. That made it pretty easy to just feed Drago V concepts into Dream. Then go in and change the prompts to call for an old man. You’ll most definitely notice the similarities in attire.

Eckhart

This example shows how you can get an evolution of the same character using the same reference art. By changing the prompts for lighter armor, or a business like suit, you can go from fully armored to office worker. 3 separate characterizations simply by evolving a reference image with the techniques I’ve shown you.

Faction

You can even evolve a character reference into multiple characters of the same faction.

Bonus: Ships

Here I have some early concepts of some of the main ships in the Novel. These may change but I like the looks so far.

Conclusion

Keep in mind that final results, and most of the example results above are the sum of my efforts. You won’t likely achieve the perfect result with your first prompt and reference. However, if you use the techniques I’ve shown you and you keep at it, it shouldn’t take you any more than a couple of hours to perfect 2-3 pieces of artwork.

At the end of the day, someone lacking skill and time can produce satisfactory art. That will enable you to focus on the other tasks that you actually are good at. For example, writing the story or outlining the beats. It used to take many hours, even days, and you had to have skill. Now you can get a simple character concept in as easy as a few minutes of work.