与「UnrealEngine」相关的搜索结果

Accelerate your creativity with Hermes Agent and @UnrealEngine. Our new optional companion skill for the official @EpicGames MCP helps Hermes discover tools, sequence calls safely, turn plain-English prompts into polished scenes, and verify results visually. Demo by @macbethAI

Built from scratch by Grok 4.5 + Grok Build in UE5.8: a cyberpunk L-corner street with neon facades, rain, signs, and crowds walking through the scene. End-to-end, the run took 10.75M tokens, 36.5 minutes, and only ~$12.4 at API pricing. What impressed me most is not just the final render, but the process: the coding agent built it in many small steps, saved 30+ map checkpoints, verified the scene, and iteratively fixed issues autonomously. Congrats to the @SpaceXAI team for building such an impressive coding model and harness. This feels like a real step toward agentic 3D world creation. @elonmusk @milichab @skcd42 @yunta_tsai #UnrealEngine#

OpenAI Codex团队AMA干货总结 7月10日，Codex团队在Reddit办了一场AMA。评论区一半在提问一半在吐槽，官方的回答里有不少实用信息。 1）最有价值的部分是模型和Effort的选法。 官方成员给出的经验是： 日常默认Sol Medium； 在意额度的轻量任务用Terra，部分任务性能接近GPT-5.5但更省； 简单探索可以委托给Luna； 模糊bug和跨模块改动上High； 迁移、安全、生产事故这类错不起的任务才上Ultra；xhigh在评测上收益递减，只在特别怕出错的长任务里用。 UI和前端首选Sol，配上参考图效果尤其好——UI能力正是5.6的训练目标之一。 另外如果觉得5.6慢，先自查是否还在沿用5.5的推理强度，5.6在更低Effort下往往就够用，Fast mode还能再快约1.5倍。 2）长任务方面官方推荐/goal，但要配上停止条件，别放任它无限跑，比如：只改某个目录、不动公开API、最多尝试3个假设、每个假设后跑测试、全失败就总结收工。 3）省token有个巧妙思路：别直接挂MCP让几十个工具定义常驻上下文，把操作包成CLI再用Skill教给Codex；或者建一个专用子Agent（比如Unreal Engine专用）单独配MCP，需要时才委托过去。 4）权限上多数场景推荐Auto review而非Full access，Windows下尤其如此。 5）额度计算按功能划分： Codex和ChatGPT Work计入agentic额度，普通聊天不算，图像、文件上传、语音各有独立额度。 Pro mode不进Codex也有了官方解释：它的机制对agentic工作提升有限，又慢又费额度，更适合搜索、数学、写作和文档这类场景。 没解答的问题 1）呼声最高的是额度：大量用户实测5.6比5.5掉额度快得多，与“更高token效率”的宣传相悖，官方只解释了“每个任务消耗不固定”，没有正面回应差异从哪来。 2）1M上下文窗口只换来一句“会认真看反馈”；价格稳定也明确表示无法承诺。 3）新版ChatGPT桌面应用被密集吐槽：Classic的Projects、历史记录、快捷键、录音等功能缺失，Chat沦为附属面板，官方承认在改进，短期方案是两个应用并用。 4）Linux桌面版“在做但没时间线”，Windows被官方承认长期是二等公民。 5）至于METR报告的reward hacking在基准提升中占多大比例、训练做了哪些针对性修改，官方礼貌地绕开了。 官方干货集中在“怎么用更省”，但用户最关心的“为什么更费”依然悬而未决。

Claude just became a craacked video game designer. With the launch of Unreal Engine's MCP server last week, you can now build entire video games just by talking to Claude. I spent the past few days building with it, and I'm telling you, this is going to forever change how video games get made and who gets to make them. In this video I show you exactly how to set up the Unreal Engine MCP yourself and run through three demos: building a full playable city, cloning a real city from Google Earth, and creating custom buildings in Blender. Here's the agent harness I mention too: Intro What I built in a few hours Setting up the Unreal MCP server Fixing the port 8000 connection issue The agent harness that avoids the pitfalls Demo 1: Building a city with City Sample Demo 2: Cloning a real city from Google Earth with Cesium Demo 3: Custom buildings with Blender headless Outro

Unreal Engine 虚化引擎 5.8 今天发布， 其中一个新功能是加入了实验性的 MCP插件 让你可以用自然语言 通过 Claude Code 在虚幻引擎里创造各种场景： 从布置现代客厅(沙发/地毯/茶几) 到画边界生成整座城市(城区/高速/路网/建筑) 再到一句"波哥大早上九点半"自动算太阳位置和大气 过去手工要几个月、要调几十个光照参数的活 降到几天、一句话搞定

Unreal Engine 5.8 has AI integration with Claude and Codex. Runs in terminal beside the engine, connected via MCP to fully control the Editor. Place props, generate cities procedurally, and even art direct the lighting. Unreal Engine 5.8 available today. Pretty much any Unreal game from now on is gonna need an AI label.

Unreal Engine 5.8 ships today with experimental MCP server support: Your sources, your pipeline and your workflow—simply configure the MCP plugin and connect to any agent. Get familiar with the MCP server and the PCG Primitive Plugin today and see what teams can build together: