What approach to work on art is practiced in the R&D department of Belka Games, – said Anastasia Morzina, the company’s lead artist.

Seditious question: is graphics so important?

The success of the game depends on many factors: the plot, gameplay, external circumstances, the presence of a loyal community and other things.

My area of responsibility is art. And I know that one of the success factors of the game is including graphics.

And no matter how painful it would be for artists to hear, graphics are important, but this is not the main thing.

To understand this, it is enough to go from the opposite: it is obvious that the art of a large number of top games will not take the ArtStation award. Vanya Smirnov will not give art from these games his like. In one game, the anatomy of the characters will be broken, in another – icons will be poorly read.

It’s not a problem. The available level of graphics is enough for these games to find a way to their audience, which means that you will have enough.

Often it is enough to provide a game with MVP—level graphics, a minimally viable product – and it will allow it to be successful.

This is our task as an R&D department — in 3-4 months to collect projects that can be used to assess the market potential of the underlying concepts.

Somewhere out there stands Horizon Zero Dawn and Cyberpunk 2077, examples of modern art, whose teams have been polishing gameplay and polishing textures for almost decades.

However, mobile projects do not have these decades. We don’t even know what will happen in a quarter or two — the bill goes not for years, but for months. While you are honing your project, a game made on your knee can pop up on the market at any moment and occupy your niche.

Here’s an example for you to understand: do you come to a good artist and ask how to draw a cool house for a mobile casual game?

Most likely, he will share with you the description of the next pipeline:

  1. Setting analysis
  2. Preparing a sketch
  3. Sketch Approval
  4. Revision of the sketch in accordance with the edits
  5. Three-dimensional visualization (modeling and then rendering)
  6. Refinement of details

Depending on the experience and skills of a particular specialist, this work takes from three full working days.

Let’s say there are five middle-level artists in the team. You need to draw 20 such buildings with different extensions and decorations. How long will it take?

I’ll count for you: 12 days per person. This is two and a half working weeks. We’ll put backlogs, edits, and polishing here — and now it’s been three weeks.

In an imperfect world, such tasks take much longer.

It makes no sense for us in R&D to spend three weeks on a perfect result.


That’s right, because success does not depend directly on graphics.

Therefore, you need to constantly accelerate. To accelerate is not to do it carelessly, they say, “and so it will do.” To accelerate is to optimize processes.

How exactly?

We propose to introduce boosters into pipelines (from the English a booster — amplifier), which will allow you to give a normal result within the framework of fast production sprints.

So, which boosters are we talking about?

Booster One — base and reuse

The database is an internal library of graphical assets. It’s not just about the graphics that went into the build, but also about the one that remained in the backlog.

As soon as an artist appears on the project and begins to draw something, the base immediately appears.

All available assets are collected into a single library, which is subsequently used as a Lego box. A kind of constructor with ready-made universal parts, from which you can design any content required at the moment.

It is the constant work with the database and the rest of its content that provides a good boost.

For example, I have a cool tree with green foliage. This means that due to color correction we will prepare the same steep trees with red foliage, and even with blue.

Moreover, half an hour of work with scissors and … voila — we have a set of good bushes in our database.

It is based on a single sprite, which is easy to visualize in completely different ways.

You can also always duplicate objects, add a bit of detail. There was a tree — there was a forest. There was one box — it suddenly became a warehouse.

A separate feature is the use in pipeline 3D. The model can be twisted as you like, getting a unique sprite each time.

Case: we needed a hammer icon. The same hammer was lying on the tables in the locations, the same hammer was given to the character, etc.

There is an opinion that it is impossible to repeat sprites — it “cheapens” the visual, and it catches the eye of the player.

I don’t agree with that.

A large number of visual lights of different shapes, many types of tables — rather confusing.

Plus, we don’t need a lot of different elements to immerse the player in the setting. A couple of elaborate details are enough. Therefore, you can duplicate elements. The main thing is to do it competently.

Besides, if the project takes off, you can always come back and draw whatever you want. But at the first stages it is worth paying attention to the functionality.

In the old days, at hackathons, we generally collected games on plugs and colored squares. First of all, it was important to check the gameplay — it would tighten or not.

After checking, we “stretched” the necessary textures on the plugs.

Booster Two — AI

Let’s talk about artificial intelligence, the burning Midjourney and Stable Diffusion, that’s all.

Many people think that neural networks can replace employees. They say that ChatGPT will write a script for narrators, Midjourney will draw cool art instead of an artist, and some plugin for Blender will immediately allow you to assemble a three-dimensional model without involving a 3D specialist.

That’s about what people who haven’t worked with neural networks think. I remember the riot on Artstation, when artists closed their profiles in protest against neural networks. That’s it, now the artists will be out of work.

At one time, when Photoshop first appeared, artists who painted posters and billboards by hand also rebelled against tricky machines. But time has passed, and there is almost more work — just more tools. The moral is this: neural networks are a tool. He will not replace the artist.

Now let’s talk about the real possibilities of neural networks.

We chose Stable Diffusion for a number of reasons. The most important thing is that the neural network has tools for correction, and this allows you to flexibly edit the results and make them more predictable. With it , you can also:

  • do research and concepts;
  • to fix the shortcomings of art (Inpaint) — to mask and redo the parts in which we are not sure (by the way, now the neural network copes even with hands);
  • upscale;
  • ControlNet allows you to use a concept, sketch or pose (and all this with a depth map) as a stencil.

There are many sites and channels in telegram where prompta are collected. By copying them or the LED with which they were created, you can get a similar result. One of the best assistants is Civitai.

Many have already successfully used neural networks for conceptualization. The predicted results can be obtained not only with your own embeddings, but also with existing models — it is no longer dangerous to borrow working prompta and rely on the base of stylistics or artists. I use a Lexica search engine.

Of course, we can do everything ourselves from scratch, but it will take much longer.

Booster three — collage (aka matte painting) and photo-bashing

Collage is a technique of drawing a picture from several photos with superimposing effects and finishing details if necessary.

Photobashing is one of the techniques of collaging, in which photographs are used to texture objects, rather than composing them from a variety of others.

Both approaches help speed up processes, especially if the style of your project is close to realism.

As for Belka Games, after creating a collage or a photo pencil, we make an overlay — we redraw the whole picture or part of it right on top of the illustration.

For example, you need to draw a seal. We collect an item from three or more refs. Plus, do not forget that you can always combine similar images. That is, it is not necessary to look for a print pen. You can take the handle of a fork, a spatula — anything.

The pipeline usually looks like this:

  • we break the object into pieces (the seal is a pen plus an impression);
  • we assemble a pen from two pieces and an impression from the shapes;
  • we add sealing wax with an impression if desired.

Assembling such a sketch takes about an hour.

Artists may be outraged: copy and paste is not art! Brueghel will curse and Djama Jurabaev will not appreciate it.

In fact, he will appreciate it. After all, once he himself uttered golden words: “I can draw a horse. But why should I prove it to someone if I can cut it out of a photo?“.


The tools, including photo-bashing, say nothing about the skill level of the author. Photobashing is the same art, and many of the coolest chromakeys are built on the photobash.

Booster Four — Team

The main booster is the team.

It is best, as a rule, when the team consists of generalists, universal soldiers who pull a bunch of versatile tasks at the same time and adapt to any wind of change.

But artists are not always fully developed. Someone has a good expertise, someone may have a limp, some guys do not take talent and perseverance, but they lack an academic base.

All this is normal. And all problems can be solved if the team has a lead, who often just needs to explain the basic principles to a junior colleague. Interested people always learn quickly. An example of this is our recent case study on the project.

And it is possible (and better) to go further and organize mini-teams with cross-leads, each of which will have a specialist with its own area of responsibility. It is important to first make sure that the areas of responsibility correspond to the strengths of the guys.

In this case, a magic circulation will occur:

  • everyone in the team will communicate with each other;
  • they will begin to understand the meaning of feedback, learn how to give it and use it;
  • they will get out of the habit of the word “edits” and stop being very stressed;
  • they will better understand how the project depends on their personal efforts.

It is important not to neglect the team spirit. Cool — while also teaching. The training itself can be carried out in different ways.

One option is to gather at a call of two or three people in the Slack channel or Google Meet, in which a more experienced participant streams the workflow with detailed comments, answering questions that arise.

In fact, this is an online backstage, to which there is usually no access. The format is suitable for artists of any level of training.

The second option: a detailed feedback in the task. Its advantage is that a person will be able to re-read the text several times and delve deeper into the meaning. It is important to attach references here so that a person immediately understands what is meant.

Specifically, I can advise the leaders to evaluate the weak areas of their artists and focus on them. That is, to compose tasks in such a way as to work out vulnerabilities on specific examples.


Let’s repeat and fix it: it allows you to speed up the production of content:

  • the base and its relaunch;
  • neural networks;
  • collage and photo-bashing;
  • and the main booster is a well—coordinated team, in which the exchange of experience is established.