素材来源 / Credits

这个试玩版里的素材大多是现成的,AI 负责挑、拼、调光。
Most assets here are off-the-shelf; the AI picked, assembled and lit them.

天空、材质、树 / Sky, materials, trees

Poly Haven(CC0):kloofendal_48d_partly_cloudy_puresky;asphalt_02、concrete_pavement、concrete_wall_008、painted_plaster_wall、long_white_tiles、worn_mossy_plasterwall、large_sandstone_blocks、rock_wall_08、leafy_grass、concrete_floor_worn_001、grey_roof_tiles、dark_wood、corrugated_iron、painted_metal_shutter;jacaranda_tree、island_tree_01/02/03、tree_small_02、searsia_lucida(远处的树是这几种模型从 36 个方向拍下来的图 / distant trees are these models photographed from 36 directions)。

路人 / Pedestrians

Blender Studio「Human Base Meshes」(CC0)。

楼的立面和列车的贴图 / Facade and train textures

李子坝楼、周围居民楼的立面和单轨列车的贴图是用 AI 出图模型(GPT Image 2.5)生成的,生成时参考了维基共享资源上的照片。这些贴图按 CC BY-SA 4.0 发布。参考的照片:
Facade and train textures were generated with an image model (GPT Image 2.5) using photographs from Wikimedia Commons as references; the textures are released under CC BY-SA 4.0. Reference photographs:

洪崖洞的木立面和石砌铺面同样是出图模型生成的,参考了 洪崖洞全貌(CC0)和 Hongya Cave 20180520(CC BY 2.0)。石墙、混凝土、路面、远处的树、行人和天边的远山也是出图模型生成的,没有用参考图;画面里的行人不是真人的照片。人行道的石板、行人走路的每一帧、每张立面的高度图(哪儿凸、哪儿凹)也是出图模型画的。
The Hongyadong timber and stone facades were generated the same way from the two photographs linked above. Stone walls, concrete, asphalt, distant trees, pedestrians and the far skyline were generated from text only; the pedestrians are not photographs of real people. The paving, the pedestrians’ walk-cycle frames and the height map of each facade were also generated.

洪崖洞的白粉墙木架立面和山墙(第四段新增)是同一个办法生成的,参考的也是上面两张洪崖洞照片。
The white-plaster half-timbered facades and gable walls of Hongyadong were generated the same way from the same two photographs.

黄昏和夜里的天空 / Dusk and night skies

Qwantani Dusk 1 (Pure Sky)、Qwantani Night (Pure Sky),Poly Haven(CC0)。

车 / Cars

路上的轿车和出租车是自己在 Blender 里用脚本建的虚构车型,比例参考了一张 AI 画的四视图线稿,不对应任何真实品牌。
The cars are fictional models built by script in Blender, proportioned from an AI-drawn four-view blueprint. They do not represent any real make.

第八段:出租车和路上的私家车换成了图生 3D 的模型。参考图是出图模型只凭文字画的「不对应任何真实车型的普通轿车」,形状由腾讯开源的 Hunyuan3D-2 在本地生成,贴图是照着模型的五个方向重新画了再烘上去的。车上没有任何车标。
Since phase 8 the taxi and the private cars are image-to-3D models: reference pictures drawn from text only (a generic car that matches no real model), shapes generated locally with Tencent's open-source Hunyuan3D-2, textures painted per view and baked. No badges.

街上的东西 / Street furniture and shop fronts

垃圾桶、配电箱、消防栓、公交站这些道具和行道树的树干是同一个做法的图生 3D 模型(见下面「AI 生成的模型」)。李子坝楼底的一排门面是出图模型画的,店名都是编的。压顶、路沿石、盲道、雨棚的贴图是代码现画的。
Litter bins, electrical cabinets, hydrants, bus shelters and tree trunks are made the same way (see “AI-generated models” below). The shop fronts under the Liziba building are AI-drawn; every shop name is invented. Coping stones, kerbs, tactile paving and awnings use textures drawn in code.

AI 生成的模型 / AI-generated models

下面这些三维模型是 AI 生成的。形状由腾讯混元的开源模型 Hunyuan3D-2(Tencent Hunyuan 3D 2.0 Community License)在本地电脑上生成;参考图和贴图由 GPT Image 2.5 生成,参考图只凭文字描述画,没有用任何照片或真实产品图。
出租车、两厢车、SUV、面包车、电动车;垃圾桶、配电箱、消防栓、公交站、石球、石凳、水马、锥桶、折叠桌和塑料凳、成箱的水和啤酒箱;行道树的树干(黄葛树、香樟);路灯的灯头和杆脚。
车和道具上没有任何品牌标志,车的造型不对应任何真实车型。

The 3D models listed here are AI-generated. Shapes were generated locally with Tencent Hunyuan's open-source model Hunyuan3D-2 (Tencent Hunyuan 3D 2.0 Community License); reference pictures and textures were generated with GPT Image 2.5 from text prompts only, with no photographs or product images.
Taxi, hatchback, SUV, microvan, electric scooter; litter bin, electrical cabinet, fire hydrant, bus shelter, stone bollard, stone bench, water-filled barrier, traffic cone, folding table with stools, crates of drinks; street-tree trunks (banyan, camphor); street-lamp heads and pole bases.
None carries a brand mark, and no car reproduces a real model.

第二代车(AI 生成):重庆的士、新能源轿车、小电车、中型 SUV、大 SUV、溜背轿车和它们的轮子也是这个做法(Hunyuan3D-2 生成形状,GPT Image 2.5 画参考图和贴图)。和上面不同的一点:画参考图时给出图模型看了维基共享资源上的真车照片,只用来定「这一类车」的大小和比例,要求画成没有车标、不对应具体车型的车;照片本身不在游戏里。照片的作者和许可见 ASSETS.md。车牌、车门上的公司名和电话都是编的。
Second-generation cars (AI-generated): the Chongqing taxi, the electric sedan, the micro car, the mid-size SUV, the large SUV, the fastback sedan and their wheels are made the same way (shapes by Hunyuan3D-2, reference pictures and textures by GPT Image 2.5). One difference: when drawing the reference pictures the image model was shown photographs of real cars from Wikimedia Commons, only to set the size class and proportions, and was asked for unbranded cars that match no specific model; the photographs themselves are not in the game (authors and licences are listed in ASSETS.md). Plates, the company name and the phone number on the doors are invented.

公交车(AI 生成):路上的公交车也是 AI 生成的模型(形状 Hunyuan3D-2,参考图和贴图 GPT Image 2.5;画参考图时看了真公交的照片当样式参考,去掉了一切标识,照片不在游戏里),线路和站名是编的。 The city bus is an AI-generated model too (shape by Hunyuan3D-2, pictures by GPT Image 2.5, with photographs of real buses shown as a style reference only); route numbers and stop names are invented.
骑车的人(AI 生成):路上骑电动车和骑摩托的两个人也是 AI 生成的模型(形状 Hunyuan3D-2,参考图和贴图 GPT Image 2.5),参考图只凭文字画,不是真人,车上没有车标。
Riders (AI-generated): the scooter rider and the motorcycle rider on the road are AI-generated models too (shapes by Hunyuan3D-2, reference pictures and textures by GPT Image 2.5, from text prompts only); they are not real people and the bikes carry no brand marks.

走路的行人:人行道上走路的人,人物模型和走路的动作(真人动作捕捉)来自微软开源的 Rocketbox 人物库,MIT 许可(Copyright (c) 2020 Microsoft),贴图缩小过;脸是 AI 照原位置重画成中国人面孔的(GPT Image 2.5,只凭文字,不是真人),头发和眼珠调了颜色。github.com/microsoft/Microsoft-Rocketbox
Walking pedestrians: the people walking on the sidewalks use avatars and motion-captured walk cycles from the Microsoft Rocketbox avatar library, MIT License (Copyright (c) 2020 Microsoft), with textures downscaled; faces were repainted by AI (GPT Image 2.5, from text only, not real people) and hair and eye colours adjusted.

站着的人(AI 生成):公交站和铺面前站着的六个人是 AI 生成的立体模型(形状 Hunyuan3D-2,参考图和贴图 GPT Image 2.5,只凭文字画),不是真人。
Standing people (AI-generated): the six people standing at bus stops and in front of the shops are AI-generated 3D models (shapes by Hunyuan3D-2, pictures and textures by GPT Image 2.5 from text prompts only); they are not real people.

位置说明:游戏里的地图是压缩过的。李子坝站的位置是照真实的来的;洪崖洞、解放碑、重庆大剧院真实都在李子坝下游几公里外,游戏为了一趟能看完,把它们搬到了李子坝附近:真实的洪崖洞和李子坝并不隔江相望,真实的解放碑在洪崖洞背后山上的步行街里、不在江边,重庆大剧院在江北嘴、和洪崖洞隔嘉陵江相对(这一点游戏里是对的)。走到这些地方时,左上角的名字卡上也会写。
About the map: the map is compressed. Liziba Station is where it really is; Hongyadong, Jiefangbei and the Chongqing Grand Theatre really stand a few kilometres downstream and were moved next to Liziba so one trip can take them all in. The real Hongyadong does not face Liziba across the river; the real Jiefangbei monument stands on a pedestrian street up the hill behind Hongyadong, not on the riverbank; the Grand Theatre is at Jiangbeizui, facing Hongyadong across the Jialing River (that part is right in the game). The name cards in the game say the same.

洪崖洞的楼和行道树的叶子(AI 生成):对岸吊脚楼群的五种楼(临街的、方形多层的、高塔楼、大屋顶的长廊楼、带城墙的楼)是 AI 生成的立体模型:形状 Hunyuan3D-2,参考图、白天和夜里的贴图 GPT Image 2.5。画参考图和白天贴图时给出图模型看了游客拍的洪崖洞照片当样式参考,照片本身不在游戏里。楼上的招牌字是代码画的,「洪崖洞」是地名,其余店名是编的。行道树的叶簇贴图是 GPT Image 2.5 只凭文字画的。
Hongyadong buildings and street-tree leaves (AI-generated): the five building types of the stilt-house complex across the river are AI-generated 3D models (shapes by Hunyuan3D-2; reference pictures and day/night textures by GPT Image 2.5). Visitor photographs of Hongyadong were shown to the image model as a style reference only and are not in the game. Sign lettering is drawn in code; “洪崖洞” is the place name, other shop names are invented. The street-tree leaf texture was drawn by GPT Image 2.5 from a text prompt.

同样是 AI 生成的图片(GPT Image 2.5):楼的立面、李子坝楼底的门面、公交站的广告、行人、列车和公交车的涂装、李子坝观景坝子后墙的马赛克壁画(图案是编的,不是现场那一幅)、江边那组带空中连廊的高楼的玻璃幕墙、洪崖洞的另外两套立面、螺旋匝道里的街心公园地面、对岸远处成片的老城屋顶、江上游船的船身。店名、广告和站牌上的线路都是编的。
Also AI-generated images (GPT Image 2.5): building facades, the shop fronts under the Liziba building, bus-shelter posters, pedestrians, train and bus liveries, the mosaic mural behind the Liziba viewing plaza (an invented design, not the real one), the curtain walls of the riverside towers with the sky bridge, two more Hongyadong facade styles, the park ground inside the spiral ramp, and the old-town rooftops across the river. Shop names, posters and bus routes are invented.

地图数据 / Map data

这一版没有用到 OpenStreetMap 的数据,地形和路是自己摆的。
No OpenStreetMap data is used in this build.

字体 / Typeface

标题和景点名用的是思源宋体(Noto Serif SC,Adobe、Google)的一个子集,按 SIL Open Font License 1.1 使用。
Titles and place names are set in a subset of Noto Serif SC (Adobe, Google), used under the SIL Open Font License 1.1.

景点介绍 / Place notes

名字卡上的介绍出自维基百科的「李子坝站」和「洪崖洞民俗风貌区」两个条目。
The notes on the place cards come from the Wikipedia articles on Liziba station and Hongya Cave.

代码 / Code

three.js(MIT)。

← 回到游戏 / Back