您需要 登录 才可以下载或查看，没有账号？注册
x
本帖最后由 chenpan85 于 2022-9-4 23:44 编辑
Etienne Carrier： My name is Etienne Carrier and I've been working as a technical artist at Ubisoft Montreal for 3 years. When I joined Ubisoft, the Far Cry 5 project was just getting started, which was the first 3A project I worked on. Since I had never worked on a triple-A game before, my secret weapon in joining Ubisoft was my experience in procedural development, which I had been doing for six years. Before I joined Ubisoft, I worked on indie games, and in the process I added procedural generation to my pipeline, so today I'm happy to share with you the procedural generation technology that Ubisoft used on Far Cry 5. 我的名字是Etienne Carrier，在育碧蒙特利尔担任技术美术师已经3年了，我加入育碧的时候，《孤岛惊魂5》项目刚刚开始，这是我参与的第一个3A项目。由于我之前没做过3A游戏，不过我加入育碧担任这个岗位的秘密武器是在程序化研发方面的经验，这方面我已经做了六年。 加入育碧之前我是做独立游戏的，这个过程中我把程序化生成加入到了我的研发管线当中，所以今天很高兴能够与大家分享我们育碧在《孤岛惊魂5》项目上用到的程序化生成技术。
Before we get started, here's a list of things we're going to share today, starting with a quick introduction and what challenges we want to address, followed by the goals of the pipeline, followed by a look at procedural generation tools, which are all from the user's perspective. Then we'll learn more about the pipeline, detail the Cliff and BiOME tools, talk about some of the changes made during development, and conclude. 1. The challenge 开始之前，这是今天我们要分享的内容列表，首先是做一个快速的介绍并说明我们希望解决的是哪些挑战，随后是这个管线的目标，接着我们了解程序化生成工具，这些工具都是从用户角度看待的。 随后我们会对这个管线进行更多的了解，然后详细介绍Cliff和biome工具，接着会说一些研发过程中做出的改变，最后是总结部分。 1.挑战
One of the challenges we had in developing Far Cry 5 was the constantly changing terrain. Gifs showed the terrain changing many times in two and a half years, so the challenge was how to manage so much content with so many iterations. Let's take a look at a small part of the terrain. This is a forest terrain that we started with for our level art. This vegetation distribution is consistent with the initial terrain, but as the terrain iterates, we see that the vegetation placement becomes problematic. 在《孤岛惊魂5》的研发过程中，我们遇到的一个挑战是持续变化中的地形，通过动图我们可以看到地形在两年半之内的多次变化，所以我们的挑战是迭代如此频繁的情况下，如何管理这么多的内容。 我们来看地形当中的一小部分，这是我们的关卡美术最初做的一个森林地形，这个植被分布与最开始的地形吻合，但随着地形的迭代，我们会看到植被的位置就会有问题。 Therefore, we wanted to make sure that the vegetation was properly distributed according to the terrain at each iteration, and we also wanted to determine how the game world would be consistent and coherent when multiple players were in the game world. Of course, in the face of these problems, we could have locked in these terrain locations from the beginning of the project, but this was obviously not practical, as we would have to iterate continuously to achieve the quality required for the final release. 2. The target 因此，我们希望能够在每次迭代的时候，这些植被都能够根据地形合理分布，另外我们还希望确定的是，当游戏世界里有多个玩家的时候，如何让他们看到的游戏世界是持续和连贯性的。 当然，面对这些问题，我们也可以在项目一开始就锁定这些地形的位置，但这很明显是不现实的，因为我们必须不断地迭代，才能达到最终发布的质量要求。 2.目标 The first goal is to build micromanagement tools that populate the game world with more natural game content. These then need to be consistent with terrain topology, such as the same case where the vegetation is distributed exactly as the terrain changes.
第一个目标就是打造微观管理工具，用更加自然的游戏内容填充到游戏世界当中。然后，这些内容需要与地形拓扑（terrain topology）一致，比如这是同一个案例，但植被完全按照地形的变化进行分布。 Then, the pipeline needed to be automated, so we used Houdini and the Houdini engine, and a Nightly build on the Build Machine, which was used to refresh the world every night. We have multiple tools for dealing with different parts of the game world, namely maps. 随后，这些管线需要是自动化的，所以我们使用了Houdini和Houdini引擎，以及在build machine上打造的一个Nightly build，后者主要用于每天晚上刷新游戏世界。我们有多个工具处理游戏世界里的不同部分，也就是地图。 All the data generated must be deterministic, meaning that the same input needs to lead to the same result. No matter how we bake the terrain, we can see that the terrain in the middle is constant, because we have Nightly build, so the connection between the maps needs to be seamless. 产生的所有数据必须是确定的，意味着同样的输入需要导致同样的结果，不管我们对地形如何烘焙，都可以看到中间的地形是不变的，因为我们有Nightly build，所以地图之间的衔接需要是无缝的。 As a final goal, we wanted these things to be player friendly. Nightly Build alone is not a guarantee that the user will get the latest data, and we need to change the data as we work, so we added a lot of tools. 3. The tool Now let's take a quick look at the tools available. Initially, we only worked on tools to solve the BIOME distribution, but as the project expanded, we created a number of tools. 最后一个目标，我们希望这些东西是对玩家友好的，只有Nightly build是无法保证用户获得最新数据的，我们还需要在工作过程中随时改变这些数据，所以我们加入了很多的工具。 3.工具 现在我们快速了解一下能够使用的工具，最初我们只做了解决biome分布的工具，但随着项目的扩大，我们创造了很多的工具。 The first is freshwater tools, which are used to create bodies of water such as lakes, rivers and waterfalls. 首先是淡水工具，它用来生成湖泊、河流、瀑布等水体。 在陡峭表面生成悬崖的工具 We also have a cloud generator that can generate different concentrations of fog on a 2D map depending on the terrain. 我们还有一个云雾生成工具，可以在2D地图上根据地形生成不同浓度的雾气。 Finally, we have a world map that generates worlds based on terrain, trees, etc. 4. User perspective 最后，我们还有一个根据地形、树木等因素生成世界的世界地图。 4.用户视角 First of all, users do Terraforming Pass, and we do use Wall Machine Pass to start projects, but most of the Terraforming is using existing tools. 首先，用户会做terraforming pass，我们的确用了wall machine pass开启项目，但大部分的terraforming都是用已经存在的工具。 The artist can then distribute fresh water on top of it, mainly by using a curve, like here we use the fresh water tool directly, which will generate a body of water on the ground and some objects in the water. After that, the user can generate cliffs, and the tool basically does this through terrain ramps, which can be micro-controlled, but most of the time the user doesn't have to do anything, just run the tool and create cliffs in these terrain areas. 然后美术师可以在上面分布淡水，主要是通过使用曲线实现，比如这里我们直接使用淡水工具，这时候就会在地面上生成水体以及水中分布的一些物体。 随后，用户可以生成悬崖，工具基本上是通过地形斜坡实现的，虽然可以微观控制，但大部分时候用户都不需要做任何事，只需要运行工具就可以在这些地形区域生成悬崖。 Next, the level artist can add vegetation, mostly using the Biome Painter, and the Biome tool automatically generates properly distributed forest vegetation across the terrain, which can also interact with water resources. These tools are great, but users want to customize it, so an artist can clean up an area and add a path, and because we're doing it naturally, we can add it directly to the map. 接下来，关卡美术师可以增加植被，主要是使用biome painter实现，然后biome工具就会在整个地形上自动生成合理分布的森林草木，而且它们还可以与水体资源互动。 这些工具都很不错，但用户还希望对它进行定制化，美术师可以对一片区域进行清理，然后增加一条道路，因为我们做的内容比较自然，所以可以直接加到地图上。 Vegetation can be seen casting reflections on the road, which has been cleared of trees. However, there are some resources that need to be replaced manually, such as vehicles, which are all done by the level artists, and we may need to tweak the Biome painting a little bit, such as adding driving roads, adding trees around, etc. After refreshing the BiOME tool, we can see the effect: 可以看到植被会在路上投下倒影，路上的树木也已经被清理掉。 不过，还有些资源需要手动替换，比如车辆，这些东西都是关卡美术师做的，在此基础上，我们可能要稍微调整一下biome painting，比如增加驾驶道路，在周围增加一些树木等。 刷新biome工具之后，我们可以看看效果：
The user can also add wires in the terrain. Here, we can add two kinds of wires connected to each other, so we add a single wire and connected wires. If the user does not design wires for the distant terrain, the system will automatically generate wires. 用户还可以在地形当中增加电线，这里我们可以增加两种相互连接的电线，所以我们增加了单条线和相连接的电线，如果用户没有对远处的地形设计电线，系统会自动生成电线。 At this point, trees are blocking the path of power lines, so we need to refresh the BiOME tools, because these tools can interact with each other and exchange information. After one change, other changes can be made. 这时候，树木挡住了电线的道路，所以我们需要刷新biome工具，因为这些工具是可以相互影响并交换信息的，一个变化之后，其他可以对应做出改变。 So that's the user perspective tool. But to show some uninterruptible workflows, the terrain can be changed at any time, such as lowering the terrain, or in this case adding an island, and once we're happy with the changes, we just need to refresh the tools, in this case I just need to refresh the Cliff and Biome tools. 以上就是用户视角的工具。不过为了展示有些无法被打断的工作流程，地形是可以随时改变的，比如降低地形，或者在这里增加一个小岛，一旦我们对这些改变满意，只需要刷新这些工具，这个案例中，我只需要刷新cliff和biome工具就行了。 5. The line Next we look at how all the tools work in the Houdini engine. We use the internal engine DUNIA as the input and then Houdini as the output, basically exchanging data between the two engines. You probably know what Houdini is, but this is a quick illustration of what it can do. Some of the inputs are from Dunia to Houdini in Python, such as world information and world names. Instead of making one world, we make multiple worlds for testing and run these pipelines on top of them. We also send file paths to ensure that the files are saved in the same folder on the PC, and these file paths are also sent to the Houdini engine. We also sent terrain, curves and shapes and so on, but also terrain, like fences, not just the distribution of fences, but the shape and style of fences. 5.管线 接下来我们看所有工具在Houdini引擎是如何工作的，我们使用了内部引擎DUNIA作为输入，然后通过Houdini输出，基本上是在两个引擎之间交换数据。 你们可能知道Houdini是什么，不过这张图可以简单展示它能够做什么。 有些输入是用Python从dunia到Houdini，比如世界信息、世界名字，我们不止做了一个世界，而是为了测试做了多个世界，并且在其之上运行这些管线。 我们还发送了文件路径，以确保文件在PC上被保存在同一个文件夹内，这些文件路径也发送到了Houdini引擎。我们还发送了地形，以及曲线和形状等等，还包括地形，比如栅栏，不止有栅栏分布情况，还有栅栏的形状与风格。 However, the most important input of the programmatic tool is the terrain itself. Since the generation is done in a specific area of the terrain, the terrain is divided into many small sectors, namely the pink part in the figure (64mx64m), which is the smallest area that users can adjust when they want to use the tool. 不过，程序化工具最主要的输入还是地形本身，因为生成是在地形上的特定区域完成的，所以地形是被分为很多小扇区（sector）的，也就是图中的粉色部分（64mx64m），当用户想要使用工具的时候，这是他们可以调整的最小区域。 When generating areas, the user can make choices based on different situations, such as the entire map or an area, and all adjustments can be seen directly from the camera's perspective. Then we look at the data that Houdini can send back, such as 2D terrain data, terrain texture layout, programmatically generated geometry, and terrain logical areas, which are mainly terrain ID numbers, all of which are temporarily cached on disk because they avoid loading time for data transfers. So it's easier to transfer data between Dunia and Houdini. 生成区域的时候，用户可以根据不同情况做出选择，比如整个地图或者某个区域，所有的调整都是可以直接通过摄像头视角看到的。 然后我们看看Houdini可以发回的数据，比如2D地形数据、地形纹理布局，几何体则是程序化生成的，还有地形逻辑区域，后者主要是地形的id编号，所有这些数据都是暂时以缓存形式存在于磁盘，因为避免了数据传输的加载时间，所以这样在Dunia和Houdini之间传输数据比较简单。 Some specific information is stored in Entity Point Cloud, such as specific location information of certain objects, vegetation resources, rocks, collectible items, decals and VFX data, prefabricated buildings, etc., and even animals in the environment. All of these objects have corresponding ids for accurate adjustment. 有些具体信息存储在Entity Point Cloud，比如某些物体的具体位置信息，植被资源、岩石、可收集物品、decals以及VFX特效数据、预制建筑等，甚至可以是环境中的动物，这些物体都有对应的ID以便精准做出调整。 The purpose of this tool is to create cliffs on large, steep surfaces and create detail on top of the terrain. In this session, we'll start by talking about the techniques used in the Far Cry series, then look at tool inputs, layering, geometry, shading methods, terrain data, corrosion, vegetation surfaces, and exporting data out of the tool. 这个工具的目的是在比较大的陡峭表面制作悬崖，并且在地形之上创造细节。 在这个环节，我们首先会谈谈之前孤岛惊魂系列使用过的技术，随后会了解工具输入、分层、几何形状、着色方法、地形数据、腐蚀、植被表面以及向工具外输出数据等内容。 In fact, far cry series have not used this tool before, before the far cry 4 and cliff just terrain in the game, although I hadn't worked in these projects, but through this chart we can see that project no details on the cliff before filling, because the manual of these resources will make the rise in the cost of r&d. But on Far Cry 5, we wanted to do a bigger project, to make the cliff a deeper representation, which is why we developed tools for it. 实际上，孤岛惊魂系列之前没有用过这样的工具，在《孤岛惊魂4》以及之前，游戏里的悬崖就只是地形，虽然我之前没有在这些项目工作过，但通过这张图我们可以看出来，以前的项目没有对悬崖进行细节填充，因为手动做这些资源会让研发成本提高。 但在《孤岛惊魂5》项目上，我们要做更大的项目，让悬崖有更深层次的表现，这也是我们为之研发工具的原因。 这是Cliff工具程序化生成之后的悬崖 We actually deleted all surfaces by using Terrain Slope as a starting point and within a certain slope threshold, such as the right side reduction, which is the cliff input geometry, which gives the player a visual reminder that there is a cliff and you can't walk on it. 我们实际上是通过使用Terrain slope作为起始点，并且在特定的坡度阈值之内删掉了所有的表面，如右侧缩减，这就是悬崖输入几何体，这种几何体还给玩家视觉提醒，这里有悬崖，是不能在上面行走的。 Then there's another interesting phenomenon that we hope to reproduce with these tools, which is geographic stratification. The yellow line in the picture is the formation of rocks over time. For this reason, we built the strata tool. Each layer is random, and we gave each layer a unique ID to make it easier to debug these strata. 随后，还有一种有趣的现象是我们希望通过这些工具重现的，那就是地理分层。图中的这条黄线是岩石随着时间变化而形成的，为此我们打造了地层工具，每一层都是随机的，我们给每一层都有独特的ID，以便让这些地层做debug的时候更方便。 Then we need to control the angles of these formations to get the desired effect, which we usually do on RGB terrain, and Houdini gives us the Angle parameters. Instead of selecting an infinite amount of data, we used only four to allow the user to choose the terrain shape that best fits their world. But just being that way, it didn't look natural, so we took it a step further. 然后我们还需要控制这些地层的角度，达到我们想要的效果，通常我们是在RGB地形上实现，随后Houdini就会给出角度参数。 选择参数的时候，我们并没有选择无穷多的数据，而是只用了4个，以便让用户选择最适合他们游戏世界的地形形状。 不过，只是这样的话，看起来是不自然的，所以我们做了进一步处理。 Then, using the attributes we just transformed, we generate color layouts that can control macro changes in the game world. Another thing to note is that this color could have been made a cliff color, but we didn't do it for two main reasons. First, we wanted to keep the distant content even if it wasn't loaded yet because of the distance. In addition, we wanted the memory footprint for the programmed generation of texture color processing to be as low as possible, and because we used Terrain Shader, Cliff Mesh didn't require any UV. 然后，用刚刚转换的attribute，我们生成颜色布局，可以控制游戏世界的宏观变化。另外需要说的是，这个颜色本来可以做成悬崖颜色，但我们没有这么做，主要有两个原因，其一，我们希望保留远方的内容，即使是因为距离的原因它们还没有被加载；另外，我们希望用于纹理颜色处理程序化生成的内存占用尽可能低，而且因为我们使用的Terrain shader，悬崖网格（cliff mesh）不需要任何uv。 From the Cliff Mask that we moved to the terrain, we expanded the cliff by running a slow simulation. We could see the points scattered across the cliff face sliding down to accommodate the corrosion effect, and it was worth noting that the original color remained in the corroded area. 从我们转移到地形上的cliff mask，我们通过运行缓慢模拟的方式对悬崖进行拓展，我们可以看到分散于悬崖表面的点向下滑落以适应腐蚀效果，值得注意的是，原来的颜色依然保持在腐蚀区域。 We can also distribute vegetation on cliff faces, and the way we detect it is by separating the up-facing cliffs individually, and then we can distribute trees and other plants on those cliff faces, as shown in the picture. 我们还可以在悬崖表面分布植被，我们检测的方式是将面朝上的悬崖单独分离，然后可以在这些悬崖表面分布树木和其他植物，如图所示。 Finally, this data is transferred to the editor via the Cliff tool. It includes geometry and collisions, rocks and entities point clouds on slopes, terrain texture ids, Cliff colors on the terrain, and Cliff masks. 最后，这些是通过Cliff工具传输到编辑器的数据，包括几何体与碰撞，还有岩石与斜坡上生长的树木entities point cloud，有地形纹理ID，还包括地形上的悬崖颜色以及cliff mask。 Here's a screenshot of some of the cliffs in the game. 7. Biome tools The purpose of this tool is to spread biome content across the game world, and here we cover a lot of features, mainly because there are so many forms of content that we want to make, so this tool can do a lot of things. The first step is to generate the terrain from Heightmap, which is simple. 这是游戏里一些悬崖的截图。 7.Biome工具 这个工具的目的是将生物群系内容遍布游戏世界，这里我们会覆盖大量的功能，主要是因为我们想要制作的内容形式有很多，所以这个工具可以做很多事。 第一步是从Heightmap生成地形，很简单。 Then, from the topological topology we generated Abiotic Data, which is the physical characteristics of the land. It's important to note that all of these topographic properties are going to be key to how we do most biome formations. 然后，从地形拓扑我们生成了Abiotic Data，也就是这片土地的物理特征。 值得注意的是，所有这些地形属性都会成为我们做大多数生物群系的关键。 Once this is done, the terrain data is stored on Terrain Attributes, and we also introduce other 2D data so that the Biome Painter data generated by the user through the tool is consistent with the programmatically generated data, such as fresh water, paths, fences, wires, cliffs, etc. What's interesting about the Biome tool is that it doesn't generate all the creatures in the game world at once, but rather breaks them down into major creatures and minor creatures. 完成这些之后，地形数据会存储在terrain attributes上，我们还引入其他2D数据，以便让用户通过工具生成的Biome Painter数据与程序化生成的数据一致，比如淡水、路径、栅栏、电线、悬崖等。 Biome工具比较有趣的是，它并不会一次性生成游戏世界里的所有生物，而是分为主要的生物和次要生物进行制作。 This tool gives us a macro world that looks very natural. The Main Biome also makes other things, such as creating a place for the user to place power lines in the generated forest. We can see the Power Lines Mask being added to the Biome tool to replace data and grass in the mountains. 这个工具给了我们外观看起来非常自然的宏观世界。 Main Biome还制作其他东西，比如生成的森林当中留出给用户放置电线的地方，我们可以看到power lines mask被加到了biome工具之上，以取代山区当中的数据和草地。 Then we did Sub-Biomes Recipes, which are made up of many parts, each representing a different creature, like forest, grass, etc., and at the center of these Biomes is Generate Terrain Entities. It can distribute units over terrain, and it can modify and create terrain attributes, but a core part of it is determining viability for each species. Each species strives for their growth and prosperity, and survivability is about determining for each species its favorite terrain attribute. The species best adapted to the terrain will outperform the others. 随后我们做了Sub-Biomes Recipes，它由很多部分组成，每一个都代表了不同的生物，比如森林、草地等，在这些biome的中心是generate terrain entities。 它可以将单位分布在地形上，还可以修改并创造地形属性，但其中核心的部分是对每个物种确定生存能力。 每个物种都为它们的生长和繁荣而努力，生存能力就是为每个物种确定其最喜欢的地形属性。最适应地形生存的物种会（在这个地形上）胜过其他物种。 For example, species A prefers to live on this closed terrain, and species B prefers to live on the bottom FlowMap terrain. If the FlowMap value is high enough, then species B will outperform species A. A survival radius is added to the selection of the winning species. For example, in this case, if the blue species grows within the radius of the green species, it is removed, which is the red species in the picture. Also, we want grass under the trees, so these won't be removed. 比如A物种喜欢生存在这种封闭地形之上，B物种喜欢生长在底部FlowMap地形，如果Flowmap值足够高，那么B物种就会胜过A物种。 在选取获胜物种的时候，还会加入一个生存半径， 比如在这个案例中，如果蓝色物种生长到了绿色物种的半径之中，它就会被移除，也就是图中红色的物种。另外，我们也希望树底下有草丛，所以这些不会被移除。 To do that, we have a priority radius, so we evaluate the priority first, and if two species have the same priority, then we do a survival of the fittest. In this diagram, for example, the blue tree has a longer priority radius, while the yellow species has a priority of 0, so it will grow closer to the green trees. However, we don't want it too close to the blue trees, so whenever the yellow grass gets too close to the blue trees, they will be removed. Viability plays a big role in the distribution of vegetation. Now let's look at some natural phenomena. 为此，我们设计了优先半径，所以首先会评估优先级，如果两个物种的优先级相同，我们这时候会采取优胜劣汰。比如在这张图中，蓝色树木的优先半径更长，而黄色物种的优先级是0，所以它会生长在离绿色树木较近的地方。不过，我们不希望它离蓝色树木太近，所以只要黄色草丛离蓝色树木距离太近，它们就会被移除。 生存能力在植被分布方面起到了很大的作用，接下来我们看一些自然现象。 In the image on the left, there is no vegetation on the flowing line, and in the image on the right, there is almost no vegetation on the saline surface of the mountain. To mimic something like this, we need to combine different topographic data. By fusing the topographic abiotic data, we can create a very specific rhythm for the vegetation distribution, and then we can combine the two topologies, and we can also fuse the noise into the topographic data. We also want to add the exclusion masks generated by the previous tools, such as fresh water, roads, cliffs, etc. 左侧图片中，在流动的线上没有任何植被，右侧的图中，山脉的盐碱表面几乎没有任何植被。为了模仿类似的事物，我们需要将不同的地形数据结合起来。 通过融合地形非生物数据，我们可以为植被的分布创造比较特别的节奏，然后我们将两种地形结合起来，还可以将噪点融合到地形数据中。 随后我们还希望加入之前工具生成的exclusion mask，比如淡水、路、悬崖等。 So we've got a map of the distribution of a species in this terrain. In addition, our biome tool includes a species resources, different size in the far cry before the series production, the size of these species is randomly selected, and the living conditions of the terrain and species preference is not consistent, now our tools can choose different size for these species according to different situations. 这样我们就得到了一个物种在这个地形的生存分布图。 另外，我们的biome工具还包含了一个物种不同大小的资源，在之前的《孤岛惊魂》系列制作中，这些物种的大小只是随机选择的，与地形以及物种偏好的生存环境并不一致，如今我们的工具可以根据不同情况为这些物种挑选不同的大小。 For example, in this scene, many factors can affect the size of trees. For example, small saplings will be distributed at the edge of the forest, and tall trees will be in the center of a large forest. In addition, altitude also affects tree size. Although we limited the size of the footage, we designed a lot of options for the tools in order to achieve diversity of vegetation. For example, we start with 50 meters of trees, then we can add 40 meters, 30 meters, etc., and make sure that trees of different sizes are distributed on the right terrain for them. 比如在这个场景中，很多因素可以影响树木的大小，比如小树苗会分布在森林边缘，高大的树木会在比较大片的森林中心。 另外，海拔也会影响树木大小。 虽然我们限制了素材大小，但我们给工具设计了很多的选择，以便实现植被的多元化。比如我们开始是50米的树，随后可以加入40米、30米等等，并确保不同大小的树木分布在适合它们的地形上。 However, there is a problem that trees of different sizes form a ladder distribution. To solve this problem, each size of tree can be scaled down or increased to fill in the gaps between different sizes. If you don't want to mess up the diversity, you can also use random distribution. In addition, different sizes of the same tree may not be exactly the same, so our tool also provides a variety of options, such as in this case each tree has its own diversity. 不过，这也会遇到一个问题，那就是不同大小的树木会形成阶梯状分布，为了解决这个问题，我们让每个大小的树木都可以按比例缩小或者增加，以填补不同大小之间的空隙。 如果不希望把多样性搞砸，还可以使用随机分布。 另外，不同大小的同一种树不一定是完全一样的，所以我们的工具还给出了多样化选择，比如这个案例当中每棵树都有自己的多样性。 Let's look at another phenomenon, which is the ecological succession of the tree canopy. As mentioned earlier, small trees tend to be located at the edge of the forest, but sometimes in the middle of the forest, and when you go into the forest, you will find that some small trees have already begun to grow and have reached a certain height. Viability influences tree selection, but the resulting diversity is not always smooth because of the terrain we use, which can sometimes result in the tallest trees being distributed at the forest edges. For this reason, we added the age parameter, and we can see different effects through age. 我们再来看另一个现象，也就是树冠的生态演替，前面提到，小树往往位于森林边缘，但有时候也有些会分布在森林中间，当你走进森林的时候，你会发现一些小树也已经开始生长，并且已经达到一定高度。 生存能力会影响到树木的选择，然而由于我们使用的地形不同，最终带来的多样性并不总是那么流畅，这就会在有些时候让最高的树木被分布在森林边缘。 出于这个原因，我们增加了年龄参数，我们可以看到通过年龄的变化会出现不同的效果。 We need to control another important factor is the density of the trees, if you take the same density, so the distance between the young trees may seem more natural, but for the tall trees, it would seem to be very busy, and we are done for different trees resources can also lead to different performance overhead, such as small trees occupy GPU is lower than the tree, It is therefore necessary to control the density of different types of trees. 我们需要控制的另一个重要因素是这些树木的密度，如果采取同样的密度，那么小树之间的距离可能看起来比较自然，但对于高大的树木，就会显得很拥挤，而且我们为不同树木做的资源也会导致不一样的性能开销，比如小树占用的GPU比大树低，因此控制不同类型树木的密度也很必要。 Basically, our density is controlled rampwise by size, but there are also factors like age. In addition, we can also adjust the vegetation density by slope. 基本来说，我们的密度是通过大小进行斜坡式控制的，但也会考虑到年龄等多个方面的影响。 另外，我们还可以通过坡面来调整植被密度。 A good in-game example of this is that leeward slopes have fewer trees than windward slopes. 这就是一个不错的游戏内案例，比如背风坡的树木就比迎风坡稀少。 The other thing to say IS THAT WE WANTED TO HAVE a variety OF colors IN the GAME, like the grass here has a lot of colors. To save on resource creation costs, we added a separate color distribution feature that could be controlled by survivability or time. 另外一个要说的是，我们希望游戏里的颜色也有多样化，比如这里的草丛颜色就非常多。为了节约资源创作成本，我们增加了单独的颜色分布功能，比如可以通过生存能力或者时间来控制。 These vegetation distributions also need to be rotated individually, and by default, they are rotated according to the slope of the terrain. For example, in the game, the grass along the river always falls towards the water because there is always a slope on the water bank. 这些植被分布还需要有各自的旋转，默认设置下，他们是按照地形坡度进行旋转的。 比如在游戏里，河边的草始终倒向水面方向，那是因为水岸上总会有斜坡存在。 Another example is the bending of a tree trunk, which you may have seen in nature, but if you haven't noticed it, you can look at it. Unfortunately, we don't do that in the game, and I wish our system would allow it. We did some other cool things, like the grass here swaying in the wind. 另外一个案例，就是树干的弯曲，你们可能在大自然中看到过这种现象，如果没有注意到，你们可以专门去观察。不幸的是，我们并没有在游戏里这么做，我很希望我们的系统能够允许这么做。 我们做了其他一些比较酷的事情，比如这里的草丛会随风摇曳。 Here we rotate the grass according to the wind vector map, but also adjust the wind direction for the effect of the terrain, so whenever the wind blows, the grass will rotate in the direction of the wind. Our vegetation rotation can be rotated horizontally based on the slope Angle, which is controlled by percentage. 我们这里根据wind vector map对草丛做了旋转，但还按照地形影响对风向进行了调整，所以只要有风吹来，这些草丛都会按照风的方向旋转。 我们的植被旋转可以基于斜坡角度进行水平旋转，这是通过百分比控制的。 There's also rotation jitter, and we use these rotations randomly on the vegetation, except for shore and fixed locations where rotation is inconvenient, but most places where the vegetation rotates in a random way. [backcolor=rgba(255, 228, 181, 0.7)]Having said most of the parameters that govern vegetation generation, however, the biOME depends on resource placement, because some vegetation resources can also affect the terrain resources, and even the surrounding resources, such as two different trees affecting the distribution of objects around them. 还有旋转抖动，我们对植被使用这些旋转的时候是随机的，除了岸边和固定地点的旋转是不便的，但大多数地方的植被旋转方式都是随机的。 说完了大部分控制植被生成的参数之后，然而biome取决于资源放置，因为有些植被的资源也会影响到地形资源，甚至会影响周围资源，比如两种不同的大树就影响了周边的物体分布。 For example, this tree trunk with cracked bark has some dead branches around it, and the shade created by these trees can also affect the growth of other species. To achieve this effect, we pass some resource data back into the terrain in the system for four different things, including terrain distortion, terrain texture, terrain data output, and terrain color. We can use terrain distortion examples to generate a mask at these distribution locations, which can also be combined with terrain data, such as using road mask to get information about the adjacent road. 比如这个树皮皴裂的树干周围有一些枯枝分布，另外，这些树木产生的阴影也会影响其他物种的生长。 为了实现这样的效果，我们在系统中将一些资源数据传回到地形当中，主要用于4种不同的事情，包括地形畸变、地形纹理、地形数据输出和地形颜色。 我们可以通过地形畸变举例，我们可以在这些分布位置生成一个mask，还能够与地形数据相结合，比如使用road mask获得与路相邻的信息。 In this case, we actually raised the terrain by 1 meter, which would result in terrain distortion in the game. I think this is a very interesting feature because most of the time in the game the trees are directly on the ground and we don't fill in any more information about where they are, but in nature the roots tend to be higher than the ground and they also help prevent soil erosion. 这里实际上我们把地形提高了1米，对应在游戏里就会出现地形畸变。 我觉得这是一个非常有趣的功能，因为大多数时候，在游戏里树木都是直接长在地上，我们不会填补它所处位置的更多信息，但在大自然中，树根往往会会比地面高一些，他们还可以防止水土流失。 Taking a closer look at these roots, we really want them to blend seamlessly into the terrain, which we can do by generating matching terrain textures. To generate the mask for the distribution unit again, we set the number of textures we want to generate for this mask, then select Air Texture from the menu, and our tool will then select which texture to apply where. The terrain ID is then sent to the editor, where it is mixed with the terrain as needed. 接下来近距离看看这些树根，我们非常希望它们与地形无缝融合，可以通过生成匹配的地形纹理来实现。 再次对分布单位生成mask，我们设置想要对这个mask生成的纹理数量，然后从菜单选择air texture，我们的工具随后会选择哪个纹理应用到哪些地方。 随后，地形id会被发送到编辑器，它就会按照需要与地形混合在一起。 Next, we might want to add something extra to the tree, which we can do by generating attributes for the tree, but the mask is stored on the terrain and we can set the parameters and ranges we want. Then we go to the next node, Forest Rock, and reuse the generated mask. 接下来，我们或许还想要给树木添加额外的东西，可以通过树木生成attributes的方式实现，不过，这时候mask会存储于地形上，我们可以设置自己想要的参数和范围。 然后我们来到下一个节点，森林岩石，重新使用生成的mask。 The corresponding behavior between these resources is then triggered. Going back to the ponderosa pine case, we can also add age data, which allows us to add new elements based on the age of the tree, and if you live in such an environment, you can see that some natural phenomena are present in the game, not just shaded. We can also add small trees among the trees to increase the ecological succession effect. 然后就会触发这些资源之间的对应行为。 回到黄松树案例，我们还可以加入年龄数据，这就可以让我们根据树木年龄增加新的元素，如果你是生活在这样的环境中，就可以发现有些自然现象也出现在了游戏里，而不只是被阴影挡住。 我们还可以在树木中间增加小树，以增加生态演替效果。 When we do vegetation, we iterate and add more and more elements until we achieve this complete combination. A Far Cry 5 map is one square kilometer, and such a map has more than 600,000 units distributed, all generated programmatically using the Biome tool. In FACT, WE added about 70 OF these maps TO THE GAME to make a FULL, lifelike world, so THE amount of resources generated was huge. 做植被的时候，我们会不断迭代，增加更多的因素，直到实现这样完整的组合。 《孤岛惊魂5》的一个地图是一平方公里，这样的一个地图当中有60多万个单位分布，都是通过biome工具程序化生成的。实际上，我们在游戏里增加了大概70张这样的地图，以组成一个栩栩如生的完整世界，所以生成的资源量很大。 This is a Google Earth image, and what's interesting is the difference in topography, you can see that the green color is different in the dry and wet areas. In order to make the world more realistic, we also made terrain texture tones and mixed them with natural texture colors. This allowed for a variety of terrain colors. The neutral sod was a grey-green color, and we mixed it with the terrain colors to achieve light and dark color changes. 这是一张谷歌地球图片，比较有趣的现象是地形差异，可以看到干燥区域和湿润区域的绿色是不一样的。 为了让游戏世界更真实，我们还做了地形纹理色调，并且将它与自然的纹理颜色混合，这实现了地形颜色的多样化，中性的草皮是灰绿色的，我们通过与地形颜色的混合来实现光暗效果下的颜色变化。 Zooming in on a part of the world, we can see the effect of the topographic colors blending with the vegetation. It is also worth noting that the terrain and vegetation colors mentioned above can also be applied to the sod, as the grass shader can also select the terrain color. 放大世界一部分，我们可以看到地形颜色与植被混合的效果。 另外值得注意的是，以上提到的地形与植被颜色还可以用到草皮上，因为草丛着色器也可以选择地形颜色。 Finally, we passed data from Houdini to the editor, including Entities Point Cloud, terrain height maps, terrain texture ids, and more. 最后，我们从Houdini将数据传送到编辑器里，包括entities point cloud、地形高度图、地形纹理id等等。
I'll jump right to the tips section, but first of all, more power requires more responsibility, and procedural tools can generate a lot of data, which gives us great control over performance, but it also affects overall gameplay and art. For example, they wanted to make a very dense forest in Far Cry 5, but in reality it wasn't that interesting because trees and animals couldn't live in such a forest, so we had to make a special set of tools to control the biodistribution, to increase the density without affecting the gameplay. Second, elegantly designed tools open up more opportunities. Our Biome, for example, gives us a lot of opportunities. Also, keep things simple. While this already seems complicated, most of the time the tools are partial engineering, and once the overall view is established, you can invest some time in simplifying it. Listen to your users, sometimes they might prefer to control the UI over the process, like our Fresh Water tool, where you can automatically generate the river, or you can have the artist design it manually, but most of the time you don't need the manual feature. Stay flexible, as initial plans are not always the best. Also, make sure you have a good balance between control and automation. Too much emphasis on manual control can increase costs, but too much automation can make things get out of hand. 我直接跳到心得部分，首先要说的是，更大的权利需要更多的责任，程序化工具可以生成大量数据，这让我们可以很好的控制性能，但这也会影响整体玩法以及美术。比如他们想在《孤岛惊魂5》做非常浓密的森林，但在实际上它并不是那么有趣，因为树木、动物不可能在这样的森林里生活，因此做生物分布的时候我们必须专门做一套工具进行控制，在提升密度的同时确保不影响玩法。 其次，设计优雅的工具带来更多机会。比如我们的biome就给我们带来了大量机会。 另外，将事情做的简单一些。虽然这已经看起来很复杂，但大多数时候工具都是偏工程化的，一旦确定了整体观，就可以投入一些时间对其进行简化。 聆听用户的意见，有些时候，他们可能更喜欢控制UI，而不是处理过程，比如我们的淡水工具，你可以自动生成河流，也可以让美术师手动设计，但大多数时候可能都用不到手动功能。 保持灵活，因为最初的计划并不总是最好的。 还有，确保在控制与自动化之间有比较好的平衡，过于强调手动控制会增加成本，但太多的自动化会让事情失去控制。 最后，感谢这些人对管线做出的贡献。
| 
评分
盆子
LingTreeBi...
li19880217
强
