Arcane Battles

The Challenge

Designing a deep, competitive card game with fair free-to-play mechanics, real-time PvP, and a scalable content pipeline that could produce new cards weekly posed challenges across game design, technical architecture, and AI integration. The card battle engine needed to support complex interactions between 500+ cards with unique abilities, chain reactions, and edge cases — a single bug in card interaction logic could ruin competitive play. Real-time PvP on mobile requires careful netcode design to handle disconnections, slow connections, and potential cheating without ruining the experience for either player. The AI art generation pipeline needed to produce card illustrations that were stylistically consistent with the game fantasy aesthetic, while being unique enough that each card felt special. The matchmaking system needed to be fair (matching players of similar skill) while also being fast (nobody wants to wait 60 seconds for a match). The seasonal content system needed to allow the 3-person studio to create and release new cards, events, and balance patches on a weekly cadence without requiring our ongoing involvement. Finally, the game needed to be lightweight enough to run on devices with 2GB RAM while still looking beautiful — a significant constraint given the particle effects and card animations.

Our Solution

We built the game on Unity with a custom card battle engine that uses a data-driven approach — every card ability, interaction rule, and combat mechanic is defined in JSON configuration files rather than hard-coded, allowing the Mythic Studios team to create and balance new cards without touching code. The engine includes a simulation mode that can play thousands of automated matches to test balance before a card goes live. For real-time PvP, we implemented a lockstep networking model with rollback prediction, ensuring gameplay feels responsive even on 4G connections with up to 200ms latency. The anti-cheat system validates all game actions server-side, making client-side cheating impossible. The AI art pipeline uses Stable Diffusion fine-tuned on 10,000 hand-curated fantasy art images in the game specific style, with a ControlNet conditioning system that ensures consistent card framing, character poses, and elemental color palettes across all generated cards. Each card goes through a 3-step generation process: base image generation, style consistency check (automated), and final review by the art director. The matchmaking system uses a modified Glicko-2 rating system with confidence intervals, achieving a median queue time of 8 seconds and a skill gap of less than 150 rating points between matched players. The entire content management system was built as a web dashboard (Next.js) where the studio can create cards, schedule events, push balance patches, and monitor game health metrics. Development took 10 months for the initial release, with LIAWEB providing 3 months of post-launch support for optimization and the first two seasonal updates.

Results

Tech Stack