I still see people creating and using images that look like they were generated in 2023, which was ages ago.

This post will be less technical and more artsy, but I hope it’s still practical and useful. Generative models are like inference computers, each a little different, with their own personality, you might say (different training data, architecture, tuning, post-training RL, etc.), and we have to learn how to “push on them” in the right way to get interesting outputs. You could perhaps think of them as different musical instruments; even within the same family, the violin and viola will have their differences.

Our first task will be to identify at least three styles that we can blend in a way that matches an aesthetic that we like. With good reason, most models today won’t allow you to simply pick a single living artist and then create pictures that impersonate that style. However, what we can do, what any human would do in creating their own style, is draw inspiration from various sources.

I like to use a site called Midlibrary. This site is tailored to Midjourney users, but it’s okay for us because we’re just going to use it as a reference for artists and for art styles. I’m using ChatGPT Images 2.0 for this post, but the process is the same for any image generation model. The difference is that your output will vary quite a bit depending on what it’s trained on, but the trial and error of developing your specific style remains the same. The consistency should persist once you finalize your prompt.

I begin by exploring the site for styles and techniques that I believe will work with the idea I have in mind until I find at least three, maybe four, styles that have elements that not only match what I was imagining, but that I think will work well together.

Tip: you can also combine wildly different styles, but you’ll spend much more time in this first step doing trial and error. And there’s nothing wrong with that! I’d argue it can be quite fun, but less so if you’re on a timeline.

For this post I’ve picked: Aron Demetz, Paul Kaptein, Parquetry, Maquette.

Once you’ve picked a few styles, then I describe an image and feed it the styles to get an idea for what this particular model can do with those inputs. This was my first prompt:

The subject is a wooden figure standing in front of an easel, working on an oil painting, both represented as wooden carvings. The painter's clothing style is steampunk. The image on the canvas is a modern cityscape. The painter is standing in a natural setting like a field. There are no other objects other than the easel in the picture. Though the subject is a wooden figure, the image is actually drawn on rough canvas. And camera/viewer lens is wide angle. The aspect ratio is 16:9.

The style blends the influences of Aron Demetz and Paul Kaptein and further incorporates parquetry to create a maquette figure.

Normally my subject is quite a bit more simple than that, but when the subject you describe isn’t well covered by the styles you use, you might have to work harder. Of course, if it’s not a style you like, this is the first place to begin tweaking and experimenting by using other stylistic inputs. Don’t worry about making this perfect, because we’ll be doing further refinements in the next couple steps.

(An interesting experiment you can do is specify three styles that are very different. For example, maybe one is a vibrant illustration of characters, one is realistic architecture photography, and another is earthen pottery. Then, when asking for an image of a subject combining those styles, depending on the model, I’ve found in the past if the subject is, say, “A skyscraper,” I get realistic architecture photography with one of the other elements blended in. Which leads to both a very out-of-scope technical discussion and a fun philosophical conversation about whether these systems can be creative.)

A wooden steampunk figure painting a cityscape on an easel in a field

Once you get a style that you like, have the model describe the style to you. This requires a multimodal model, but that’s the norm at this point unless you’re using open-source models.

Describe the style, composition, and palette of this image.

Here’s what I got back from mine:

Style:
The image reads as a painterly digital artwork with a strong oil-on-rough-canvas texture applied across the whole scene. The central figure and easel are rendered as intricate wooden carvings, with visible grain, segmented panels, parquetry-like inlays, and maquette construction. The figure’s steampunk clothing is suggested through carved wooden coat panels, belts, buckles, cuffs, and subtle gear-like ornaments, giving it a handcrafted automaton quality. The cityscape on the easel contrasts with this sculptural wood aesthetic by appearing as a looser, impressionistic oil painting.

Composition:
The composition is wide and cinematic, using a 16:9 landscape frame with a slight wide-angle feel. The wooden painter stands left of center in profile, facing the easel on the right. The easel and canvas form the visual counterweight to the figure, creating a balanced two-subject arrangement. The open field and large sky give the scene a spacious, solitary atmosphere, while the low horizon emphasizes the figure’s vertical presence. There are no distracting props; the focus stays on the figure, easel, and painted cityscape.

Palette:
The palette is dominated by warm natural wood tones: honey, tan, amber, ochre, walnut brown, and darker carved accents. These are balanced by the muted blue-gray sky and soft cream clouds. The field uses subdued golds, olive browns, and dry grass tones, reinforcing the rustic natural setting. The cityscape introduces cooler grays, slate blues, and smoky blacks, adding urban contrast without overpowering the wooden figure. Overall, the color mood is warm, earthy, and softly atmospheric.

We want to use this in a new prompt now. For example, we want our prompt to be subject-agnostic, so we’ll remove those references. This is also the second place where you can make some adjustments to your taste. I reduced it to this:

Style:
The image reads as a painterly digital artwork with a strong oil-on-rough-canvas texture applied across the whole scene. The central figure is rendered as intricate wooden carvings, with visible grain, segmented panels, parquetry-like inlays, and maquette construction.

Composition:
The composition uses a 16:9 frame with a slight wide-angle feel and a spacious, solitary atmosphere. There are no distracting props.

Palette:
The palette is dominated by warm natural wood tones: honey, tan, amber, ochre, walnut brown, and darker carved accents. These are balanced by the muted blue-gray, green, and soft cream highlights. The lower tones are cooler grays, slate blues, and smoky blacks, adding urban contrast without overpowering the figure. Overall, the color mood is warm, earthy, and softly atmospheric.

Next, we want to try our prompt with a new subject to see what we get. It won’t be the same, but it should be close. This is the step where I can spend several hours sometimes without meaning to. Your goal is to experiment and evolve this prompt into something that you like. Small wording changes can make big differences here too. If the color palette is off, just find some other picture or reference on the internet that you like, have the model analyze it and output the color scheme for you, and then update this in-progress image-generation prompt with your new colors.

The goal here is to have a workable prompt which is more versatile than one more heavily rooted in a specific artist or specific styles, because your resulting images in those cases will be more heavily influenced and grounded in those styles than you want. For example, when I was coming up with a style for these blog posts, I had this Asian style at one point, which I really liked, but it wasn’t versatile enough. Every subject ended up having an orange sun somewhere and mountain pine trees.

A rope bridge crossing misty mountains in a scroll-inspired painterly style

You’ll discover that problem, if you have it, in this next step where you try your prompt with a new subject.

This was my next prompt:

A tall narrow rocket on a launchpad in the far distance, taking up about a third of the frame. On either side are solid rocket boosters. Soft "god rays" illuminate the rocket from the sky.

Style:
The image reads as a painterly digital artwork with a strong oil-on-rough-canvas texture applied across the whole scene. The central figure is rendered as intricate wooden carvings, with visible grain, segmented panels, parquetry-like inlays, and maquette construction.

Composition:
The composition uses a 16:9 frame with a slight wide-angle feel and a spacious, solitary atmosphere. There are no distracting props.

Palette:
The palette is dominated by warm natural wood tones: honey, tan, amber, ochre, walnut brown, and darker carved accents. These are balanced by the muted blue-gray, green, and soft cream highlights. The lower tones are cooler grays, slate blues, and smoky blacks, adding urban contrast without overpowering the figure. Overall, the color mood is warm, earthy, and softly atmospheric.

This image wasn’t the first I got from the prompt. I did have to tweak the description of the rocket, then move it further in the distance, then add the “god rays.” But the heavy lifting, if you will, was done by our new image-generation prompt.

A distant wooden-toned rocket on a launchpad under soft rays of light

So the process is just this:

  1. Explore and find some artists and styles that you like.
  2. Experiment to see if they work together in the way that you want.
  3. Have the model describe the images.
  4. Use the description to create a new image-generation prompt.
  5. Experiment and refine the prompt using new subject matter. Update the color palette as needed.
  6. Save prompt!

I don’t know that I actually like this style I came up with for this post, but it’s fine as an example. If you’re creating a bunch of images for a slide deck, you want them to all be consistent, and this is a good way to do that without accepting a default look.

It can be a lot of fun coming up with new styles. And like everything with AI, it can be powerful if we use it as a lever for our own creativity.

More to Come!

I’ll be writing AI-related blog posts every day for a while. Even though I use AI for everything nowadays, none of these posts will actually be written by AI.

So, if you find this post useful, follow along! I’ll be posting on LinkedIn, but you can also follow along here at my site.

Work with Me

By the way, I’m looking for my next opportunity and I’m available for consulting. Shoot me an email and let’s get on a call and talk!