Screenshots are work-in-progress development captures from the live AI-vs-AI spectator mode — the art pass is ongoing.
Kane’s Resurrection (a codename, not the final title) is a single-player, offline 3D real-time-strategy game: a clean-room spiritual successor to Command & Conquer 3: Tiberium Wars. It’s original IP — no C&C names, units, music, or assets — with two asymmetric factions: Foundry, a conventional military of tanks, artillery, and air power, and the Scavengers, Mad-Max-style scrappers running technicals, rocket jeeps, and salvage drones.
Why it exists
Nobody makes this game anymore. I grew up on base-crunching, harvester-guarding, superweapon-timing RTS campaigns, and the genre moved on without them. The design pillars, in build order: the feel (crunchy base-building and skirmish against a capable AI), asymmetric factions, then a campaign. Multiplayer is permanently out of scope — this is a single-player game, built like one.
Highlights
- A complete, closed match loop: economy → production → assault → victory. Matches against the AI can be won and lost, in play or spectator mode.
- 28 unit and 24 building specs across the two factions, with 63 hand-authored, silhouette-gated PBR meshes flowing through a Blender → GLB → engine pipeline.
- A four-mission campaign arc, an early map editor with live terrain preview and undo/redo, deterministic input-log replays, and a rematch/watch-replay results flow.
- Full product shell — title, skirmish setup, pause, settings, controls, accessibility — that cold-boots into a freshly generated match.
- Combat modeled as a damage-type × armor-type matrix with raycast line-of-sight and cover, over flow-field pathing with eight locomotion classes.
- Renders at 144 fps headroom on Apple Silicon Metal in dense spectator scenes, with automated packaged-launch, soak, and perf proofs gating the beta.
Technical approach
Built on a vendored copy of my RTS Engine (~105k lines of deterministic Rust), extended with a real mesh-import crate: glTF/GLB geometry, node transforms, tangents, and PBR materials bound through a content catalog, with hard budget gates on triangles and texture bytes. The whole repo runs ~140k lines with over a thousand tests. Beta packaging, soak tests, and performance baselines are automated; what remains before the tag is deliberately human — witnessed matches and an ears-on audio pass.
