Sun
mend
A cozy cooperative restoration game for a world
that just needs people to put it back together.
Why This Game Exists
The Ethos
This idea came from the despair of the current state of the world and feeling powerless to change its fate. I think I'm not alone in this feeling — and there is much to be said for hacking and slashing games releasing some of this anger — but I wanted to create a game that built resilience and strength instead.
The goals of this game are to mend, to build, to repair. There are no enemies, no retribution for actions, no stressful time challenges or predatory game mechanics. This is a game built for those seeking agency over the future — and who yearn to make it a brighter one.
Game Description
Welcome to Sunmend.
Sunmend is a cooperative cozy crafting game for Roblox set in a post-oil world that never turned dystopian — just broken. Players travel between procedurally seeded towns, each with collapsed solar grids, dead food systems, and rusting infrastructure. Rather than fighting enemies, they repair, restore, and reconnect.
The world is not post-apocalyptic. It is post-negligent. The oil ran out, systems collapsed — but the sun still rises. Solar panels exist on every rooftop but nobody wired them correctly. Seed banks exist but nobody tended the germination beds. Water cisterns exist but pipes cracked and nobody fixed them. The setting draws directly from solarpunk's core philosophy: the tools for a sustainable future already exist — they just need people to put them to work.
Visually, the world takes inspiration from Studio Ghibli's naturalistic architecture, Art Nouveau overgrowth, and Afrofuturist community design. Vegetation reclaims old roads. Solar arrays gleam on thatched rooftops. Biogas digesters glow amber at night. Every restored town looks more alive than the last.
Each player chooses a specialty and earns rank through completed restoration work. Village Elder NPCs — powered by AI — remember players across visits and respond dynamically to the town's restoration state. The result is a live social world where every fix is permanent and visible, and the community "end game" is becoming a mentor in a place you helped build.
The Process
How I built this.
I began with a rigorous analysis of platforms and genre opportunities — comparing Roblox against alternatives, mapping the genre landscape, and identifying the blue ocean opportunity that Sunmend could occupy.
Roblox
Lua + Studio, massive F2P ecosystem
| Audience | 300M+ active users, discovery built-in |
| Dev time | Fast — templates, AI-assisted Lua (GPT/Claude) |
| Monetization | Robux marketplace, ready day one |
| AI leverage | High — code gen, asset gen, NPC scripting |
| Ship risk | Low — 1 month is very realistic |
Best shot at a shippable, playable F2P game in a month with a small team.
Fortnite (UEFN)
Unreal Engine + Verse language
| Audience | Large, but content island discovery is harder |
| Dev time | Medium — Verse is newer, less AI training data |
| Monetization | Creator economy share, no direct pricing |
| AI leverage | Medium — less Verse code in AI training sets |
| Ship risk | Medium — 1 month is tight, platform is complex |
Great quality ceiling and polished UX, but Verse code gen with AI is weaker than Lua.
Horizon Worlds
Meta's VR social platform
| Audience | Small, declining DAU, needs Meta Quest |
| Dev time | Slow — scripting is limited, tooling immature |
| Monetization | Very limited, not F2P ready |
| AI leverage | Low — proprietary scripting, little training data |
| Ship risk | High — platform instability, hardware gating |
Hardware-gated audience, weak tooling. Skip.
Meta Ray-Ban AR
Wearable + Meta AI SDK
| Audience | Tiny — requires specific $300 hardware |
| Dev time | Slow — SDK limited, no game engine support |
| Monetization | Essentially none — no marketplace |
| AI leverage | High for AI features, low for game logic |
| Ship risk | Very high — not a game platform at all |
Great demo material but not viable as an F2P game.
Roleplay / Social
Brookhaven, Royale High, Adopt Me
Anime RPG / Grind
Blox Fruits, Shindo Life, Anime Last Stand
Pet Simulator
Pet Sim 99, Adopt Me economy
Tower Defense
Anime Vanguards, All-Star TD, SpongeBob TD
Obby / Platformer
Tower of Hell, Flood Escape
Tycoon Builder
Restaurant Tycoon, Prison Tycoon
Survival / Horror
DOORS, Piggy, Rainbow Friends
Shooter / PvP
Arsenal, BedWars, Rivals
Farming / Idle
Grow a Garden (broke records in 2025)
AI-powered NPC worlds
Low compRoblox is adding real-time AI NPC infrastructure. Almost no games use it meaningfully yet.
Concept: a mystery town where every NPC has secrets — players interrogate them with actual typed questions to solve a case.
Cozy crafting / life sim
RisingGrow a Garden's record-breaking 2025 run proved there's massive demand for non-combat, cozy gameplay.
Concept: a shared village where players each own a cottage, grow things, trade, and decorate.
Puzzle / escape room
Low compDOORS showed that procedurally generated puzzle-horror works on Roblox. Straight puzzle/escape-room games with no horror are surprisingly thin.
Concept: a co-op "heist puzzle" game — players have different info and must communicate to crack the same vault.
Competitive cooking / crafting
ModerateThere's no quality Overcooked-style game on Roblox. Chaotic co-op cooking/crafting has massive viral clip potential.
Concept: rotating kitchen chaos — 4 players, time pressure, escalating orders. Seasonal menus keep it fresh.
Hyperlocal / cultural RP
Low compRoblox's own 2025 data showed strong searches for localized roleplay. Culture-specific RP games barely exist outside English-language defaults.
Concept: a city RP game built around a specific culture — a "Tokyo Street Life" or "Night Market" sim.
Sports prediction / fan games
RisingReal sports IP on Roblox is rare and expensive. Fan-made sports fantasy/prediction games tap real-world calendar events and pull in an older, higher-spend demo.
Concept: a fantasy sports draft + mini-game format — draft your team, then play skill mini-games to boost your players' weekly stats.
Cozy crafting village
Identity + decoration monetization
Revenue levers
Seasonal furniture, rare flower seeds, cottage themes. No gameplay impact — pure expression.
One-time purchase (299–499R). Unlocks a unique plot size, starter pet, or cosmetic title.
6–8 week seasons with a free + premium track. Premium track (299R) gives exclusive cosmetics.
Avoid: Any pay-to-progress mechanic. Monetize the look, never the loop.
AI NPC competitive crafting
Progression + advantage monetization
Revenue levers
Extra crafting slots, faster NPC respawn, auto-collect pass (499–799R).
Ranked seasons (4–5 weeks). Premium track unlocks prestige cosmetics + end-of-season title.
Avoid: Selling stronger NPC behavior for Robux. If AI gives paying players an unfair edge, trust collapses fast.
Cozy crafting — live ops calendar
Seasonal, low-pressure, community-driven
Cadence
Launch with a "Founding Festival"
Limited founder items (never returns). Drives urgency without FOMO toxicity.
Seasonal theme drop
New craftable item set tied to real-world season or holiday.
Community garden event
Server-wide collaborative goal. All players contribute, all get a shared reward.
New recipe unlock
One new free crafting recipe per week. Low dev cost, high habit-forming value.
New biome / area expansion
New zone with new crops, crafts, and an NPC story arc.
AI competitive crafting — live ops
Season-based, urgency-driven, content-dense
Cadence
Pre-season soft launch
Launch with unranked mode only. Let players learn mechanics before ranked opens.
Ranked season with prestige rewards
Leaderboard resets each season. End-of-season cosmetics go to top X%.
NPC personality / challenge rotation
Rotate which AI NPC types appear in lobbies. Keeps meta fresh.
Tournament event
48-hour bracket. Winner-take-all exclusive cosmetic.
New crafting mechanic / recipe tier
Major meta shift that forces players to re-engage.
| Dimension | Cozy crafting | AI competitive crafting |
|---|---|---|
| Revenue ceiling | High — large audience, proven by Grow a Garden's $12M/mo | Medium-high — smaller but higher-spend competitive audience |
| Session length | Long (45–90 min) | Medium (15–35 min) — round-based |
| DAU/MAU ratio | Lower (25–35%) | Higher (40–55%) |
| Live ops dev load | Light — seasonal content, asset swaps | Heavy — balance patches, NPC updates, season resets |
| Viral growth vector | Screenshots & "show your plot" social sharing | Tournament clips, NPC chaos moments on TikTok/YouTube |
| 1-month buildability | High — core loop is simple, AI not required at launch | Medium — AI NPC integration adds meaningful complexity |
I proposed the solarpunk genre, established the requirements for a pro-social cozy experience, and confirmed Roblox as the platform. This phase produced the world structure, matchmaking system, and core loop.
Next came the full Game Design Document — iterating on mechanics, NPC systems, monetization philosophy, and world structure until the vision was production-ready.
With the GDD locked, I worked with Claude on a detailed action plan and began building the MVP in Roblox Studio — including setting up MCP connections between Roblox Studio and Claude.
Action Plan — Phase 1
Action Plan — Phase 2
I made the MCP connections between Roblox Studio and Claude — but ran into issues even with Claude giving me exact instructions for Roblox Studio's AI assistant.
Roblox Studio MCP setup
⚠ The error
AI assistant error in Roblox Studio printing "then" over and over
I paused on development and turned to Midjourney to work on art direction. Claude generated the prompt order and descriptions, and I used Omni Reference adjustments to achieve visual consistency across generations.
Claude-generated Midjourney prompts
Omni Reference for visual consistency
The art board that emerged — including animated village and character concepts:
Village concept
Village Elder — concept 1
Village Elder — concept 2
Village — animated
Grower specialty
Village — dusk
Fabricator specialty
Village — animated 2
Village — aerial
Village — detail
Village — panorama
Where This Goes