All Collections
Customer Case Studies
From Sketch to Production-Ready Asset: The End-to-End Journey with Layer
From Sketch to Production-Ready Asset: The End-to-End Journey with Layer

How to create production-level game assets without limits.

Updated over a week ago

What is Layer?

Layer is an AI-powered artist tool that empowers game teams to execute faster.

From a quick sketch, you can product a final production-ready asset in minutes.

There are a few tools out there, but Layer is special...

Layer AI

Other generative AI tools

Focus on tools that enable artists & non-artists to work faster rather than replace entire steps.

Focus creating “net new” aesthetic assets that have less practical usage.

Emphasis on game art needs rather than wider consumer appeal.

Emphasis on the black box ability for AI models to experiment. This is excellent for consumer appeal but less useful to game production workflows.

Fine-grained control over generated assets.

Less artist control tools and simple single-layer canvas.

Unlimited generation model to maximize creativity.

Limited generation model to prevent consumer abuse.

So how exactly do we go from a sketch to production-ready assets quickly?

The end-to-end journey with Layer

Step 1: Training Your Art Style

Creating your art style from your game assets

Art style training with Layer involves teaching the tool to generate artwork that aligns with the specific artistic vision of your game. By uploading a curated set of art pieces in a desired style, Layer learns to understand the nuances, patterns, and characteristics that define that particular aesthetic. This enables Layer to generate original art that resonates with the chosen style, maintaining consistency and coherence with your game.

A. Uploading Images

By uploading anywhere from 10 - 25 images, you can train a specialized model. To ensure consistency and coherence in the generated artwork, the images you upload should adhere to the same art style, and the theme or content of the images should be similar:

Adhering to the Same Art Style:

  1. Consistent Visual Elements: All images in the dataset should share common visual elements, such as brush strokes, color palettes, shading techniques, and overall composition. For example, if you're aiming for a pixelated art style, ensure that all images display characteristic pixelated content.

Subject Consistency:

  1. Similar Subject Matter: It is essential to maintain theme consistency and avoid mixing different art types within the same style. The goal is to focus on a specific theme, such as character(s) or environment artwork, and ensure that all images within the style adhere to the same subject/theme. Mixing unrelated themes or including UI elements like frames or buttons when you have previously uploaded characters will compromise the effectiveness of the training process.

  2. Avoid Mixing Characters and Items:

    1. Character Design: If your primary focus is creating an art style for character design, the images you upload should exclusively consist of character illustrations. Avoid including unrelated content like UI elements or backgrounds, as they can introduce inconsistency and detract from the character-focused style you aim to achieve.

    2. Environment Artwork: When training Layer for background environment artwork, upload images that revolve around landscapes, cityscapes, architectural designs, and natural elements. Keeping the theme consistent will enable Layer to better understand and reproduce the specific style for environments without being influenced by unrelated character or UI designs.

    3. UI and Graphic Elements: Similarly, if your intention is to create an art style for UI and graphic elements, assemble a dataset containing buttons, icons, user interface components, and relevant graphic designs. By focusing solely on UI elements within this subject, Layer can learn and generate artwork that harmonizes with the desired UI style.


In this example, Thena is a character in the comic. Training 2 styles (one for general backgrounds and one for characters) will give the best results:

B. Input Captioning

After uploading images, you need to caption them. Captioning is just as important, if not more important as the images themselves, because the captions serve as essential guidance for Layer, allowing it to understand the visual content and context of each image accurately.

This simple process involves describing the images as though you were explaining them to a blind person, taking into account not only the subject matter but also the camera angle, the viewer's perspective, and other intricate details.

Tips to caption effectively:

  1. Describe the Visuals: Be descriptive and vivid in your image captions. Imagine explaining the image to someone who cannot see, and provide details about the characters, objects, settings, and any relevant actions or interactions.

  2. Camera Perspective: Include information about the camera angle and perspective of the viewer. Specify if it's a top-down view, side view, first-person perspective, etc. This helps the AI understand the intended composition better.

  3. Emotions and Mood: If applicable, convey the emotions and mood portrayed in the image. This can influence the AI's rendering and ensure that the generated artwork reflects the intended atmosphere.

  4. Color Palette: If color is a significant aspect of the image, mention the dominant colors or specific color themes. This allows the AI to consider color harmonies during the creative process.

  5. Additional Attributes: Include any additional attributes that might aid Layer's understanding, such as lighting conditions, weather, or specific design elements.


In this example, the Thena character style is being captioned. Notice how detailed the captions are...

A caption of one of the images:

Thena, a young woman with short messy green hair, face titled, side profile looking right, white collared shirt, sharp nose, angry, yelling, swearing, graphic background, portrait.

C. Variation Selection

After captioning your images, Layer will ask you for example prompts, which serve as reference points for Layer to understand and generate example outputs. These example prompts play a crucial role in refining the art style, as Layer will create several variations of your style from which to choose from. By comparing the outputs of each style variation, you can select the one that most closely matches your art style and desired output.

By selecting the best variation, you ensure that the generated art style truly reflects your creative direction and delivers the impact you desire in your game development project.

Iterative Refinement and Creativity

Remember that the process of selecting the best variant is not final. Layer allows for iterative refinement, enabling you to experiment further with the chosen style. You can provide additional example prompts, fine-tune details, or explore different combinations to continue evolving and enhancing the art style until it perfectly matches your creative vision.

D. Naming Your Art Style

Once you have selected the variant that best matches your desired output, it's time to give your art style a unique and memorable name. The name you choose should reflect the essence and personality of the style, allowing others to grasp its key characteristics quickly.

E. Choosing Between Public and Private Styles

After naming your art style, you have the option to decide whether it should be public or private. This choice determines who can access and use the art style:

  1. Public Style: By selecting the public option, you make your art style available to other users of Layer. This means that fellow creators can explore, use, and build upon your style in their own projects. Sharing a public style can lead to collaboration opportunities, inspire other artists, and foster a vibrant community around your unique art style.

  2. Private Style: Opting for the private option restricts access to the art style, making it exclusively available to you or specific team members. Choosing this option provides greater control over your artistic creations and allows you to keep your style exclusive and reserved for specific projects or purposes. This is best for game studios.

F. Begin Training!

After naming your art style and deciding its accessibility, Layer will initiate the training process. During this training phase, Layer will learn from the uploaded images, provided captions, and example prompts to continuously improve its understanding of the art style's intricacies.

Once the training is complete, you'll have a fully developed and distinctive art style that reflects your artistic vision and game theme. At this point, you can leverage your trained style and begin the forging process.

Step 2: Forging

Creating style-consistent assets quickly

Layer focuses on helping artists and designers accelerate the process. While many AI tools are good at generating aesthetic assets, they are not good at fine-tuning, meaning they are bad if you need a very specific output. With Layer's powerful capabilities, you can quickly bring your creative ideas to life in your specific art style.

Here's how...

A. Sketch

Many assets start with a sketch. Let’s imagine that we’re making a cover image for a comic about Thena. Here’s a quick sketch.

The sketch shows Thena standing on a rooftop before a cityscape with a long-barrelled gun like a sniper rifle. We want it to be dramatic and portrait-esque but still in the distinct comic style.

We can upload this sketch to Layer and use it as guidance in our generations.

B. Writing a Detailed Prompt

Utilizing the sketch as a guidance image, we can start to generate outputs.

But alongside the sketch, we also need a prompt.

Prompts are written instructions to help guide Layer on what asset to generate. However, just like learning a new language, there’s a specific way to talk Layer to get the results you’re looking for.

To help, we've written this guide on how to write amazing prompts.

For our case, let's use this prompt:

(Thena standing holding a sniper rifle gun on top of a building), cityscape in background, full body, green hair, black top, orange background, sun in sky, [deformed, ugly], full body, dramatic, single gun barrel, side profile.

These are our results...

Many results will have parts that are usable, and many parts that are not. This is normal.

If you do not like the images, you can continue to refine the prompt or sketch as needed.

In this case, I see one image that has the general mood of what I’m looking for, we can use that as a base.

C. Editing The Image

There is so much wrong with the selected image!

Yes, it has the general vibe we are going for, and yes, matches the art style quite well, but Thena is super deformed and our “sun” is just a weird black circle in the sky

On the other hand, the cityscape is really well done and the gun is also in style.

So how can we fix this?

Editing in Canvas

Let’s use Layer’s inpainting tools to fix this up. Thena’s legs first.

You can see I have drawn Thena's legs in black and painted in a platform for Thena to stand on in brown.

The painting is quite crude, yes, but these are simply guidance tools, so they don’t need to be too detailed.

We paint, then re-run the same prompt from the canvas with 50% image similarity.

Generating With More Guidance

After painting in the areas we want, it looks like we fixed up a lot of the lower body issues!

However, we notice that the whole image gets re-made every time we generate it.

In these examples, we’ve lost some of the intricacies of the cityscape.

Let’s see how we can guide Layer only to modify certain parts.

Generating With Masks

Using the Lasso tool in Canvas, we highlight the portion of the image that we want Layer to fix up.

This is represented by the blue-dotted area.

After highlighting this area, the prompt and generations will only affect that specific area:

You can see that with controlled guidance, we can selectively “fix” the parts of each generation that are not aesthetically pleasing.

The legs are now much better looking in each variation, but let’s pick one that has the fewest issues.

We then utilize masks + painting in each of the other problem areas and fix them one by one.

Here's us doing just that:

D. The Final Result

And here it is! The finished asset!

With controlled guidance, we selectively “fixed” the parts of each generation that were not aesthetically pleasing.

We fixed:

  • Extra limbs

  • Weird bends

  • Split up gun.

  • Thena’s hair, which was mistakenly too long.

  • Thena’s face, which is now more accurate to her character.

  • Black ring circle in the background

It would typically take an experienced artist around 4-8 hours for a piece like this, but by utilizing Layer, an artist was able to create it in around 1 hour.

It’s by no means perfect, and could be touched up manually by an artist, but the end result is very similar to the desired style.

The Overall Process

It's looking good!

What next?

It is pretty awesome huh? Well, now it is your turn! It is time to jump into Layer and train your own style!

Artist credits: Training images used are based on the famous graphic novelist Matteo Scalera

Did this answer your question?