Documentation

Technical documentation and insights on how steamcapsule.com analyzes and presents Steam artworks.

Overview

This project aims to help developers, artists particularly, create more effective capsule art that improves visibility and connects with their target audience.

A total of 861 games were analyzed, extracting patterns in color schemes, title placement, and visual composition.

Genre Classification

Adventure
Arcade
ARPG
JRPG
Platformer
Puzzle
Roguelike
Sandbox
Shooter

If a game has specific genre-related tags within its top 3 tags on Steam, it's marked with the corresponding genre. This approach ensures that our genre classification reflects how the Steam community and developers categorize games.

Games can belong to multiple genres.

All Games List

Here's the list of all ... unique games analyzed in this project.

Methodology

My methodology combines visual analysis, color theory, and compositional principles to extract insights from Steam capsule art:

  • Data Collection: Using Steam Web API and SteamSpy API to gather artworks, genres, tags, and review statistics.
  • Image Processing: Analyzing each capsule using k-means clustering to extract its color palette, dominant color percentages, and visual zone distribution.

Data Collection

The data collection process involves both API usage and manual curation:

  • Collecting capsule images via Steam Store API
  • Gathering game metadata including genres, tags, and review statistics
  • Retrieving review data including review counts and review sentiment
  • Focus on games from the global Steam catalog with US region metadata

Selection Criteria

Games are selected through a weighted scoring system:

Minimum Requirements
At least 100 reviews on Steam
Quality Focus
Higher ratings increase score
Logarithmic Scale
Prevents any one game from dominating

Weighting Formula

Score =
0.6
×
log(reviews)
+
0.4
×
percentage
Popularity Weight (60%)Quality Weight (40%)
Review Count
Rating

Models & Analysis

The analysis employs several computer vision techniques in a sequential process to extract meaningful data from capsule art:

1

Color Extraction

Using a multi-stage approach, I first convert RGB data to the LAB color space for more perceptually accurate clustering, then apply k-means clustering via OpenCV to extract the 5 dominant colors from each capsule. The LAB values are normalized before clustering and converted back to RGB after. When analyzing colors, I convert them to HSV color space for better color classification and naming. For each color, I calculate percentage coverage based on pixel counts, allowing for analysis of color harmony, contrast levels, and genre-specific color trends.

Color Analysis Pipeline

RGB Input
LAB Conversion
k-means Clustering
Blue
Purple
HSV Color Naming

Color Extraction Process

GAME TITLE
Sample Steam Capsule
35% coverage
25% coverage
20% coverage
12% coverage
8% coverage
2

Text Detection

EasyOCR is used to detect text regions in capsules, helping to identify title placement and analyze how text integrates with the visual elements.

Text Detection Process

ADVENTURE
Input Capsule
ADVENTURE
Confidence: 96%
Position: CenterSize: 45% of width
Detected Text:"ADVENTURE"
Zone Placement:Zone 5 (Center)
EasyOCR detects text position, content, and confidence level
3

Zone Distribution Analysis

By dividing each capsule into a 3×3 grid (9 zones), I analyze where visual and textual elements are placed. This reveals patterns in composition, focal points, and title placement across different game genres.

Zone Analysis Process

STRATEGY
Sample Capsule Art
STRATEGY
1
2
3
4
5
6
7
8
9
Element Distribution
Character: Zones 4, 7
Title: Zones 6, 9
Coverage Analysis
Visual Elements:60%
Text Elements:25%
Popular Zone Usage
5%
5%
5%
25%
5%
15%
40%
15%
25%
Heatmap of zone usage frequency

These three analytical approaches work together to create a comprehensive understanding of what makes capsule art effective on Steam, providing data-driven insights for game developers.

Improvements & Future Plans

Several enhancements are planned to improve the analysis and user experience:

Text Detection Improvements

The current text detection system sometimes struggles with highly stylized or "unique" fonts. Future updates will implement advanced recognition models to better detect uniquely styled text elements in capsule art.

Export Custom Reports

You will be able to export detailed analysis reports based on your specific filtering criteria. For example, you'll be able to export a custom report showing color trends for roguelike games with high review scores, or a report showing zone usage statistics for games with low review scores.

Example Export Options
Color palette report by genre
Zone usage statistics by review score
Custom comparison between two genres
Text placement data for specific game types

Frequently Asked Questions

Work in progress
Coming soon...

Last updated: March 2025.