Utilizing ChatGPT to make Civ5 art style icons using GIMP

[Putmalk]

Hi everyone! Wanted to document my workflow for creating Civ 5-like icons. I'll provide some for reference here:

Spoiler Sample icons following this process :

Other icons I've made can be found in this thread: https://forums.civfanatics.com/threads/putmalks-icon-dump-thread.699760/

Tools used in this guide:

1. GIMP 3.0.4
2. ChatGPT (I pay for the paid version, not sure how the free version is. You can certain still use this process with the free version, but your results may vary)
   1. Bonyduck Campersang has some thoughts on other models that may assist here: https://forums.civfanatics.com/thre...-style-icons-using-gimp.699744/#post-16864681
3. Civ 5 icon template

Knowledge needed:

1. Need to know how to use GIMP
2. Patience and willingness to learn
3. Patience with ChatGPT

Step 1: Generate a reference image

The number one issue with icon generation is to generate a reference image. In the past, we've had to crawl google for a relevant reference image. Today, we are empowered to use LLM tools to generate unique art in any style or time period possible. Results may be hit or miss.

To begin, give ChatGPT a rule:
LLMs are extremely powerful, and can do things like interpret the art style of an image you paste in. To save you some time, I usually add a ChatGPT rule that says the following:

When I ask you to generate an icon, you are to follow the below art style.

You must generate whole images, not circles.

1. Stylized Realism
Figures, objects, and architecture are realistically proportioned but simplified.
Musculature, armor details, and architecture are clean and readable at small scale.
No photorealistic texturing—details are crisp, graphic, and painterly.

2. Bold Lighting & Shading
Strong directional light, usually from above or behind, creating a dramatic glow.
Backgrounds often use radiant light beams or gradient glows to emphasize importance.
Shading is smooth and gradient-based, rather than textured or rough.

3. Geometric Composition
Figures and buildings are composed with clear shapes (triangles, circles, rectangles).
Silhouettes are bold and instantly recognizable.
Outlines are minimal or absent; form separation is achieved through lighting and color zones.

4. Vivid, Saturated Colors
Palettes are bright but controlled—golden yellows, strong reds, deep blues, and natural greens.
Metallics (gold, steel, bronze) are highlighted with sharp reflective gradients.
Natural elements (grass, wood, cloth) are painted in solid, clean tones with subtle shading.

5. Heroic & Dramatic Poses
Units are shown in dynamic but clear stances—soldiers at attention, warriors mid-action, or craftsmen dramatically working.
Buildings are angled in perspective but centered for symmetry and impact.
Wonder/technology icons often use symbolic or idealized imagery rather than literal realism.

Summary
This style blends digital realism with painterly clarity:

Realistic anatomy and architecture.
Bold highlights and radiant backgrounds.
Simplified forms for clarity at small scale.
Saturated, clean colors with smooth shading.
Circular framing that prioritizes immediate recognizability.

Once this is set up, you can prompt the AI to describe a scene. Prompts I've used in the past has been "describe an ancient Mesopotamia fisherman", and then it will give you a detailed prompt that you can feed into image generation.
For this tutorial, I've already set up a valid prompt:

generate an icon of ancient literature following this description of the art style. Depict a man in an ancient library, with a table in front of him. On the table is a scroll. He is currently writing on the scroll using ancient ink pen (With a feather in it). On the table is also a small ink bottle. Do not generate a background, instead leave everything in the background behind the man fully transparent.

Result is:

Spoiler :

Good start, but not good enough. This violates the art style by:

1. AI-generated images are almost always yellow
2. The colors are not good, very flat.
3. Outlines are no good. The only outlines are faint and tiny in civ 5's icons.

We will fix all of this in GIMP for this tutorial. If you are unhappy with the generation, feel free to command the AI to do your bidding, but I've had to do several rounds of image generation to get something decent in some cases. Be patient.

Step 2: Manual editing in GIMP (WARNING: Difficult)
I'm going to cut out the outline of the bookshelf to get the scene that I want:

Spoiler :

Imperfections are okay. We will clean them up later.
Now, copy the civ 5 icon template, then paste this image and shrink it to frame:

Spoiler :

Next, remove the outlines. First, Lock the Alpha Channel of the layer.

Spoiler :

Now, I gotta clean this photo up to resemble the Civ 5 icons. This is the hardest step, and it will require training your skills. I like to find a Firaxis Civ 5 image similar to what I’m doing as a reference image, and then look at things to find similarities. In this case, I’ll be using the Library icon, as it has similar themes of what I want:

Spoiler :

First I need to remove outlines. To do this I’ll use a combination of the painbrush tool and the Eraser tool. If you’re using the Eraser tool, remember to unlock the Alpha channel first and relock when painting.

I cannot teach you how to do this, you will have to learn how to do this with patience. Here's some in-progress screenshots:

Spoiler :

This is detailed process, and is the majority of your skill expression will lie. In some instances, I’ll use select to focus on areas, or even update their colors.

I'll change the feather color to add more contrast to the scene.
For the feather I used a combination of Free Select + Contrast + Colorize. This is mostly digital art manipulation so pick what feels right for your image. One cannot prescribe a perfect solution for all images.

Spoiler :

After doing work for about 20 minutes, I got something like this:

Spoiler :

Step 3: The background
Use the gradient tool and select colors that contrast the image. This is at your discretion:

Spoiler :

Then use a free lasso tool to select a sunburst (I do this manually). I recommend you Feather the selection by 1px.

Spoiler :

Use the Dodge and Burn tool to brighten the selected region:

Spoiler :

De-select the region, then bright around the focal point of the sunburst. Very intense by focal point, much less around it.

Spoiler :

At this point our icon is almost complete. If you want to add more props or a background, you can. Use a similar process and repeat until you have the desired icon.

For now, I will color correct the image, as a lot of AI-generated images have a yellow tint. You can try a similar curve, results may vary:

Spoiler :

Civ 5 images tend to be high contrast, so bump that up as well. Note, every image should have a different style, so do what’s best for the image.

Spoiler :

Adjust image lighting by using dodge / burn:

Spoiler :

Step 4: Complete the icon

Create a new layer, invert the selection, color it black, and change opacity to 50:

Spoiler :

Without de-selecting the selection, move the selection 3 pixels to the right, and then 3 pixels down:

Spoiler :

Delete the selection. You should now have the top-left corner shadowing found on all Civ V icons.

Spoiler :

The last step is to repeat the alpha to selection on the border image, shrink the selection by 1 pixel, hide that layer, merge all visible layers, and then delete the selection.

Spoiler :

Because we shrank the selection by 1px, there is a border on the outside. Just select a rectangle around the image, invert the selection, and delete it.

Spoiler :

And viola, a brand new Civilization 5-style icon, ready to use (once added to icon templates)!

 

[Putmalk]

for good measure, here is another icon I've made today:

 

[hokath]

The prompt gets such a good base, wow!
1px feather on the sunburst is a nice hint, ty.

 

[Bonyduck Campersang]

priceless tutorial

I would add two things

1) the yellow tint is usually found in ChatGPT's Dall-E model (probably due to overtraining on Ghibli frames) but not on most other image-generating models (though those other models generally aren't as good at artistic images and following detailed prompts); the more images you generate wtihin the same chat the yellower they get so you would want to create new chats for each generating session

2) ChatGPT restricts free users to a limited number of image generations per day, on the other hand if you have a Microsoft Copilot account you can do the same things as in ChatGPT's premium version with no restrictions; Copilot uses the same model as ChatGPT