---
title: "Wild Fields 1648"
url: "https://aigamehub.app/zh-tw/games/wild-fields-1648-052369"
source: "https://github.com/0xInsiderhq/wild-fields-1648"
source_type: "github"
play_url: "https://0xinsiderhq.github.io/wild-fields-1648/"
platform: "desktop"
login_required: false
pricing: "free"
origin: "original"
ai_tools: ["Claude"]
game_languages: ["en"]
author: "0xInsiderhq"
author_handle: "0xInsiderhq"
category: "動作"
categories: ["action", "simulation"]
tags: ["cossack", "open-source", "procedural-generation", "rts", "threejs", "webgl"]
added: "2026-10-24T02:45:36.873Z"
updated: "2026-10-24T02:46:11.282Z"
locale: "zh-tw"
content_locale: "zh-tw"
---

# Wild Fields 1648

一款設定在 1648 年烏克蘭草原的瀏覽器即時戰略遊戲：建立哥薩克城鎮、採集資源、訓練步兵、騎兵與野戰砲，並摧毀敵對的城鎮大廳。

## Description

Wild Fields 1648 是一款單一瀏覽器分頁的即時戰略遊戲，背景設定在 1648 年的烏克蘭草原。玩家建立哥薩克城鎮、務農、砍伐森林、採石與挖金，然後訓練長矛兵、火槍兵、哥薩克騎兵與野戰砲，在自家城鎮大廳被摧毀前燒毀敵對的城鎮大廳。模擬內容包括逐步增強的 AI 對手，會建造、採集、研究與攻擊；生長與成熟的小麥、磨損的地面路徑、倒下後留下樹樁的樹木、燒毀後留下廢墟的建築，以及使用關節骨架、兩骨 IK，具備行走、奔跑與奔馳步態的角色。美術使用 25 個 CC0 照片掃描模型與來自 Poly Haven 的 PBR 材質，從 112 MB 壓縮至約 9 MB，音效則以程式碼合成。它在單一瀏覽器分頁中以 three.js 執行（桌上型，掃描資源需要短暫載入）；儲存庫中包含離線單一檔案版本與建置腳本，作者表示程式碼由 Claude Opus 5.5 撰寫。

## Original post

RTS on the 1648 Ukrainian steppe, written by Claude Opus 5.5. three.js, 25 Poly Haven scans, every frame is code.

Topics: browser-game, claude, poly-haven, procedural-generation, rts, strategy-game, threejs, webgl

# Wild Fields 1648 A real-time strategy game set on the Ukrainian steppe in 1648, written by Claude Opus 5.5 and played in one browser tab. Raise a Cossack town, farm, fell forests, quarry stone, mine gold, train pikemen, musketeers, Cossack horse and field guns, and burn the rival town halls before they burn yours. **[Play it in your browser](https://0xinsiderhq.github.io/wild-fields-1648/)** · desktop, give it a few seconds to load the scans · follow the build on X [@0xInsiderf5](https://x.com/0xInsiderf5) ![Wild Fields 1648, a regiment marching through a Cossack village](media/preview.webp) ## What is in it - People on a jointed rig with two-bone IK, foot planting, gaits for walking, running and galloping horses, corpses that fall the right way - Wheat that grows and ripens, paths that wear into the ground where peasants keep walking, trees that fall and leave stumps - 25 photo-scanned models and PBR textures from Poly Haven (CC0), cut 40 to 60 times, 112 MB of downloads down to about 9 MB - AI opponents that build, gather, research and attack; a 20 minute AI vs AI match runs without errors - Burnt buildings leave ruins (log cribs, stone walls, a brick stove), new buildings clear the bushes on their plot - Sound synthesised in code, no samples ## From the first build to v3 The first build had procedural models only. v3 added 25 Poly Haven scans that Claude found and downloaded itself by driving a browser, after its sandbox could not reach the asset sites. ![v1 vs v3](media/post_before_after.jpg) ## Run it locally The page loads `assets/` with plain script and image tags, so any static server works: ```bash python3 -m http.server 8000 # open http://localhost:8000/ ``` `dist/Wild_Fields_1648_offline.html` is the same game as one 12 MB file with every scan and texture inline. It opens straight from disk. ## Build ```bash python3 build.py ``` `src/` holds the game as plain JavaScript modules (`core`, `map`, `entities`, `ai`, `game` for the simulation, `r_*` for rendering, `ui`, `audio`, `main`). `build.py` concatenates them in a fixed order into `src/shell.html` and writes `index.html` and the offline file. three.js r128 comes from cdnjs. ## Tools - `tools/assets/` the Poly Haven pipeline: weld, simplify with meshoptimizer, drop foliage cards, pack every mesh into `assets/scan_pack.js` (gltf-transform). Paths in these scripts point at the original workspace and need adjusting. - `tools/video/` how the comparison videos are made: `rec.js` records a seeded match frame by frame in headless Chromium, `interp.py` doubles the frame rate with optical flow, `compose.py` stacks two builds into a 1080x1920 video with labels. - `tests/` Playwright scripts used while building: boot, a long AI vs AI match, performance, a full match to the victory screen, screenshot sets. ## Credits - Code written by Claude Opus 5.5 (Anthropic), directed by [@0xInsiderf5](https://x.com/0xInsiderf5) - 3D models and textures: [Poly Haven](https://polyhaven.com), CC0. See [CREDITS