Testing The Most Insane Things Genie 3 Can Create

章节 1:初探 Genie 3 与史前恐龙世界

📝 本节摘要

演讲者 Bilawal Sidhu 介绍了 Google DeepMind 提供的 Genie 3 早期访问权限,将其定义为一个互动式的“世界模型” (World Model)。他测试了第一个场景——恐龙世界,虽然指出恐龙的生理结构并不完美,但他对能够像进入电影场景一样“走进”视频画面的体验感到极为震撼,并观察到了镜头光晕等细节。

[原文] [Bilawal Sidhu]: google DeepMind gave me early access to Genie 3 This is their world model You type in words and it builds an interactive universe that you can explore for a minute

[译文] [Bilawal Sidhu]: Google DeepMind 给了我 Genie 3 的早期访问权限。这是他们的“世界模型”。你输入文字,它就会构建一个你可以探索一分钟的交互式宇宙。

[原文] [Bilawal Sidhu]: Oh my god Wo wo wo wo wo wo Holy crap Now is this tech going to replace Unreal Engine or is this something new entirely let's go hands-on with the product and find out

[译文] [Bilawal Sidhu]: 噢,我的天。哇,哇,哇,哇,哇,哇。该死(惊叹)。这技术会取代虚幻引擎(Unreal Engine),还是某种全新的东西?让我们亲自上手这个产品来一探究竟。

[原文] [Bilawal Sidhu]: What if you could step inside an iconic sequence from a movie imagine if every YouTube video or every Netflix movie you could pause at any time and hit the freaking genie button and step inside that world

[译文] [Bilawal Sidhu]: 如果你能走进电影中的经典片段会怎样?想象一下,如果每一个 YouTube 视频或每一部 Netflix 电影,你都能随时暂停,按下那个该死的“精灵”按钮,然后走进那个世界。

[原文] [Bilawal Sidhu]: So let's try and do that Oh my god it did it It did it And I got I got this thing is buzzing That's so cool

[译文] [Bilawal Sidhu]: 所以我们来试一下。噢,我的天,它做到了。它真的做到了。而且我……我这个东西在震动。太酷了。

[原文] [Bilawal Sidhu]: Okay let's check out these dinosaurs So as you can see like the physiology of these dinosaurs isn't perfect but my god is it magical to be able to step inside any of these environments

[译文] [Bilawal Sidhu]: 好的,来看看这些恐龙。正如你所见,这些恐龙的生理结构并不完美,但天啊,能够走进这些环境里真是太神奇了。

[原文] [Bilawal Sidhu]: And by the way what I found really funny is here I'm going to finish recording this Maybe I can Oh my god it's got a lot of legs Oh check out the lens flare right there Y

[译文] [Bilawal Sidhu]: 对了,我发现很有趣的是……我要结束这段录制了。也许我可以……噢,我的天,它有好多条腿。噢,快看那边的镜头光晕。


章节 2:太空漫游与光影物理测试

(Space Exploration & Lighting Physics)

📝 本节摘要

演讲者切换至太空场景,测试模型对光影和反射的理解。在一个接近恒星的飞船驾驶舱场景中,他惊叹于太阳能板上的反射效果和太阳耀斑的细节,认为这对概念艺术家具有极大的游戏设计辅助价值,并开玩笑说或许需要一场巨大的太阳风暴来应对日益先进的AI。

[原文] [Bilawal Sidhu]: ou see that guys as we look at the sun that's an implicit effect that it's learned Reflections Can I go into the water that's an interesting test Come on Come on Come on I guess not

[译文] [Bilawal Sidhu]: 你们看到了吗,各位,当我们看向太阳时,这是它学到的一种隐式效果:反射。我能进入水中吗?那是个有趣的测试。来吧,来吧,来吧。我想是不行。

[原文] [Bilawal Sidhu]: So I really want to show you this Okay So what is the environment we are in the cockpit of a spaceship approaching a massive star with fiery tendrils infusion explosions massive solar flares

[译文] [Bilawal Sidhu]: 所以我真的想给你们看这个。好的。那么环境是什么?我们在飞船的驾驶舱里,正接近一颗巨大的恒星,伴随着火焰般的卷须、聚变爆炸和巨大的太阳耀斑。

[原文] [Bilawal Sidhu]: So object spaceship Let's see if we can do this I mean this really showcases some of the game design capabilities right let's say you're a concept artist You're like cooking up a bunch of cool stuff and now you want to go play with this thing

[译文] [Bilawal Sidhu]: 所以对象设为飞船。看看我们能不能做到。我的意思是,这真的展示了一些游戏设计的能力,对吧?假设你是一位概念艺术家,你构思了一堆很酷的东西,现在你想进去玩一玩。

[原文] [Bilawal Sidhu]: Oh my god Yep It's working Look at the reflections on the on the solar panels here I can go down now Look at the tendrils and the solar flare

[译文] [Bilawal Sidhu]: 噢,我的天。是的,它在运行。看这里太阳能电池板上的反射。我现在可以下去了。看那些卷须和太阳耀斑。

[原文] [Bilawal Sidhu]: You know these are things that people talk about AI is going to get so advanced y'all We're going to need a big ass solar flare to just take out all the electronics on the planet A massive EMP pulse Quite interesting

[译文] [Bilawal Sidhu]: 你知道人们常说 AI 会变得非常先进,各位,我们需要一个巨大的太阳耀斑来摧毁地球上所有的电子设备。一个巨大的电磁脉冲(EMP)。很有趣。


章节 3:致敬《火线》与都市夜景

(Urban Night Scene & The Wire Tribute)

📝 本节摘要

受到美剧《火线》(The Wire) 的启发,演讲者尝试创建一个“夜晚巡警”的场景。Genie 3 成功生成了一个充满细节的城市街道,包括路面纹理、冒出的蒸汽和路灯效果。令演讲者惊喜的是,场景中出现了行人和看似剧中“复印店”的建筑,展示了模型对特定文化符号和复杂环境光影的还原能力,以及类似 SLAM(即时定位与地图构建)的实时世界构建体验。

[原文] [Bilawal Sidhu]: Let's try another one I've only recently been getting into the TV show The Wire And uh let's see if we can recreate like a beat cop at night right like there we go See that that did it Like let's go create a world

[译文] [Bilawal Sidhu]: 让我们试另一个。我最近才开始看电视剧《火线》(The Wire)。看看我们能不能重现一个像夜晚巡警那样的场景,对吧?这就对了。看,它做到了。就像去创造一个世界。

[原文] [Bilawal Sidhu]: Yo check this out This is exactly what I wanted We got the steam popping up Oh gosh look at the look at the the texture detail on the road right there My god And there's life in the scene for a change

[译文] [Bilawal Sidhu]: 哟,快看这个。这正是我想要的。我们看到了冒出来的蒸汽。噢天啊,看那边路面上的纹理细节。我的天。而且场景里终于有生机了。

[原文] [Bilawal Sidhu]: Gosh there are people walking by Let's see Can I can I bump into this person no I guess I just missed them Let's walk towards that car Yeah the lights Yeah look at the look at the light the street lamp too

[译文] [Bilawal Sidhu]: 天啊,有人走过。看看,我能……我能撞上这个人吗?不,我想我刚好错过了。让我们走向那辆车。是的,灯光。看那灯光,还有路灯。

[原文] [Bilawal Sidhu]: Let's see if we can go inside here Yeah this is where I mean by like the slam mapping thing As you like go inside it's building the world world for you

[译文] [Bilawal Sidhu]: 看看能不能进去这里。是的,这就是我所说的类似即时定位与地图构建(SLAM mapping)的东西。当你走进去的时候,它正在为你构建这个世界。

[原文] [Bilawal Sidhu]: This does not look like a diner This looks like the photocopying shop honestly from uh the photocopying shop in the wire uh where Edris Elbow works or his front business I should say,

[译文] [Bilawal Sidhu]: 这看起来不像是餐馆。老实说,这看起来像是《火线》里的那家复印店,也就是伊德里斯·艾尔巴(Idris Elba)工作的地方,或者该说是他的门面生意。


章节 4:交互测试——自动驾驶与工业废土

(Interaction Tests: Autonomous Car & Hazmat Suit)

📝 本节摘要

演讲者测试了与动态物体的交互能力。在街道场景中,他尝试站在路中间,发现车辆会自动在他面前停下,且车内无人,仿佛是“自动驾驶警车”。随后,他切换至一个“超现实工业地狱景象”,生成了一个穿着防护服(Hazmat suit)的角色。他特别测试了模型对“涉油而行”这种高阻力物理环境的模拟,并成功体验了在彩虹色油污淤泥中行走的视觉效果。

[原文] [Bilawal Sidhu]: Oh I got a car coming at me Let's see if I can walk in the middle of the street and kind of see what the car does Is it going to slow down before our time ends over here again the lighting and the detail Look at all of this So good

[译文] [Bilawal Sidhu]: 噢,有一辆车朝我开来了。看看如果我走到马路中间,车会有什么反应。它会在我们的时间结束前慢下来吗?再说一次,这光影和细节,看这所有的一切。太棒了。

[原文] [Bilawal Sidhu]: Or is it going to clip right through me i think we might just find out Oh it stops And there's nobody driving Bro what this is like autonomous police vehicle That's amazing

[译文] [Bilawal Sidhu]: 还是说它会直接穿模穿过我?我想我们马上就知道了。噢,它停下了。而且没人驾驶。兄弟,什么?这就像是自动驾驶的警车。太神奇了。

[原文] [Bilawal Sidhu]: Is it going to just drive off on its own and it just starts moving Okay that was that was quite fascinating

[译文] [Bilawal Sidhu]: 它会自己开走吗?它刚开始动了。好的,那真是相当迷人。

[原文] [Bilawal Sidhu]: So here's another one that we can try which is like waiting through oil Here we say okay yeah surreal industrial hellscape We're going to go with Aszmat suited figure waiting through the oil thick resistance right and we can see how well this does

[译文] [Bilawal Sidhu]: 所以我们可以试另一个,比如涉油而行。这里我们设为,好的,超现实的工业地狱景象。我们要选一个穿着防护服(Hazmat suit)的人在厚重的油污阻力中涉行,对吧?看看效果如何。

[原文] [Bilawal Sidhu]: So if things go well we'll ideally have a character walking in front of us and then you know maybe we'll be walking behind him There we go This guy is very static

[译文] [Bilawal Sidhu]: 所以如果一切顺利,理想情况下我们会有一个角色走在我们前面,然后你知道,也许我们会走在他后面。好了。这家伙非常静止。

[原文] [Bilawal Sidhu]: Let's see if we look down Do we see our hazmat suit there we go Ladies and gentlemen we are stepping through that iridescent dirt the sludge

[译文] [Bilawal Sidhu]: 看看如果低头看,能不能看到我们的防护服。有了。女士们先生们,我们正踏过那彩虹色的污垢、那淤泥。


章节 5:月球引力与物理跳跃

(Moon Gravity & Physics Mechanics)

📝 本节摘要

在月球场景中,演讲者测试了物理引擎的表现。他发现模型生成了类似喷气背包的“二段跳”机制,尽管没有完全配套的特效,但体验十分有趣。他尝试降落在登月舱上,引用阿姆斯特朗的名言,并对比了不同生成尝试中“尘土飞扬”粒子效果的有无,感叹角色惊人的跳跃能力。

[原文] [Bilawal Sidhu]: What about going to the moon then we also looked at uh the primary energy source of this planet There we go Oh I guess it gave me third person Can I jump aha Oh I can even double jump

[译文] [Bilawal Sidhu]: 那去月球怎么样?我们刚才看了这个星球的主要能源。好了。噢,我想它给了我第三人称视角。我能跳吗?啊哈,噢,我甚至可以二段跳。

[原文] [Bilawal Sidhu]: That's so funny I guess there's like a little bit of a jetack that you can kind of see It doesn't have like the effects to go with it but I can totally Oh my god I can totally

[译文] [Bilawal Sidhu]: 太搞笑了。我猜好像有点喷气背包的意思,你能隐约看到。它虽然没有配套的特效,但我完全可以……噢,我的天,我完全可以。

[原文] [Bilawal Sidhu]: Oh no No Jump again Jump again There we go Let's land on the module There we go Okay that's kind of cool Now let's take a Let's soak it in Let's soak in the sights ladies and gentlemen That is beautiful

[译文] [Bilawal Sidhu]: 噢不。不。再跳。再跳一次。好了。让我们降落在登月舱上。好了。好的,这有点酷。现在让我们……让我们沉浸其中。女士们先生们,让我们沉浸在这景色中。真美。

[原文] [Bilawal Sidhu]: And now let's lift off One small step for man one giant leap for mankind Yeah I definitely am not going to reach the end there But let's see if we get a dust dust puff when we land on the ground Nope

[译文] [Bilawal Sidhu]: 现在让我们起飞。“这是一个人的一小步,却是人类的一大步”。是的,我绝对到不了尽头那里。但看看当我们落地时会不会有尘土飞扬。不,没有。

[原文] [Bilawal Sidhu]: Oh this guy can this guy this guy got hoops Omie got hoops man In this one I got dust So as the guy jumps around you can see the dust puffs and particle effects on the ground

[译文] [Bilawal Sidhu]: 噢,这家伙能……这家伙,这家伙跳跃能力真强(got hoops)。兄弟跳得真高啊,伙计。在这一幕里我有尘土效果。所以当这家伙跳来跳去时,你可以看到地上的尘土团和粒子效果。


章节 6:高级餐厅与 NPC 行为局限

(Fine Dining & NPC Limitations)

📝 本节摘要

演讲者试图模拟巴黎米其林餐厅的服务员体验。他调侃了使用 ChatGPT 生成提示词的过程,并描述了私密的灯光与水晶餐具带来的氛围,直言这让他感觉仿佛置身于电影《教父》之中。然而,他也指出了模型的技术局限:当角色移动过快时,NPC的面部细节会丢失(变得模糊),且存在“穿模”现象——他无法推挤人群,而是直接穿过了他们,缺乏真实的物理互动反馈。

[原文] [Bilawal Sidhu]: What if we do something funny like a fine dining waiter like what if you're in a Michelin star restaurant in Paris right like you got the intimate lighting you got tablecloths and crystal wear

[译文] [Bilawal Sidhu]: 如果我们做点有趣的,比如高级餐厅的服务员?就像如果你在巴黎的一家米其林星级餐厅,对吧?你有那种私密的灯光,你有桌布和水晶餐具。

[原文] [Bilawal Sidhu]: and you're like baffled over which exact fork to use And you know of course you ask chat GPT all that right like as any other person would Just kidding

[译文] [Bilawal Sidhu]: 然后你对到底该用哪把叉子感到困惑。而且你知道,当然你会问 Chat GPT 所有的这些细节,对吧?就像其他人一样。开个玩笑。

[原文] [Bilawal Sidhu]: Uh let's do first person Carrying silver tray precisely plated arms arms in hand visible informal waiter apire moving between tables Okay Oh god we got to roll with this

[译文] [Bilawal Sidhu]: 呃,让我们做第一人称。端着银托盘,摆盘精致,手臂……手里拿着东西可见,非正式的服务员服装,在桌子之间移动。好的。噢天啊,我们得继续这个。

[原文] [Bilawal Sidhu]: Let's just see what happens Okay let's stop Stop Stop Stop Stop Stop Stop Stop Stop Homie's just doing his thing Oh there we go Now I can take over

[译文] [Bilawal Sidhu]: 让我们看看会发生什么。好的,停下。停,停,停,停,停,停,停,停。这哥们只是在做他自己的事。噢,好了。现在我可以接管了。

[原文] [Bilawal Sidhu]: This looks like I don't know why I feel like I'm in the Godfather You'll notice this limitation too is like notice how like the faces like refine when I slow down a bit but otherwise they just keep on keeping on

[译文] [Bilawal Sidhu]: 这看起来像……我不知道为什么我觉得我在《教父》里。你也会注意到这个局限性:注意看当我慢下来一点时,面部是如何变得精细的,但其他时候它们就只是一直在那儿动。

[原文] [Bilawal Sidhu]: Let's see if I can like push into people if they'll move aside Nope I just clip right through them In some situations I've noticed that I can like they'll respond to you here This is just not compelling enough

[译文] [Bilawal Sidhu]: 看看我能不能挤向人群,看他们会不会让开。不,我直接穿模穿过去了。在某些情况下,我注意到我可以……他们会回应你。但在这里这还不够有说服力。


章节 7:建筑漫游与高斯泼溅风格

(Architectural Exploration & Gaussian Splatting Style)

📝 本节摘要

演讲者测试了大教堂场景,并结合自己对“高斯泼溅”(Gaussian Splatting) 技术的喜爱,形容 Genie 3 生成的画面具有类似的点云视觉效果。他惊叹于彩色玻璃和“泼溅点”组成的吊灯所呈现出的印象派风格。随后,他通过对比自己在 Google 沉浸式视图 (Immersive View) 的工作经历,赞扬了 Genie 3 能够绕过复杂的渲染流程直接生成逼真环境的能力。最后,他控制角色“起飞”,在类似旧金山的天际线中漫游,并幽默地辨认出一座类似 Salesforce 大厦的建筑。

[原文] [Bilawal Sidhu]: Now y'all know me I'm big into Gossian splatting Let's try a cathedral and see if we can fly around it Let's see what it looks like

[译文] [Bilawal Sidhu]: 大家都了解我,我很热衷于高斯泼溅(Gaussian splatting)。让我们试一个大教堂,看看能不能绕着它飞。看看它长什么样。

[原文] [Bilawal Sidhu]: I've also got a bunch of images I can upload and test out here Wow There we go I can go around Free viewpoint Let's look around the whole thing

[译文] [Bilawal Sidhu]: 我还有一堆图片可以上传到这里测试。哇。好了。我可以到处走。自由视角。让我们环顾整个场景。

[原文] [Bilawal Sidhu]: Let's look at the stained glass How cool is that y'all a chandelier splattered points It is very impressionist looking

[译文] [Bilawal Sidhu]: 让我们看看彩色玻璃。那多酷啊各位,一个由泼溅点(splattered points)组成的吊灯。这看起来非常具有印象派风格。

[原文] [Bilawal Sidhu]: Can we clip through this see what happens Wow Okay we just got into the back rooms in the cathedral Amazing Amazing Absolutely fantastic

[译文] [Bilawal Sidhu]: 我们能穿模(clip through)过去看看会发生什么吗?哇。好的,我们刚进入了大教堂的幕后空间(back rooms)。太惊人了。太惊人了。绝对精彩。

[原文] [Bilawal Sidhu]: Yeah the stained glass looks really cool Let's see if we can do this This would be freaking cool I mean like I mean I know I worked on immersive view which is cool

[译文] [Bilawal Sidhu]: 是的,彩色玻璃看起来真的很酷。看看我们能不能做这个。这会非常酷。我的意思是,我知道我曾参与过沉浸式视图(Immersive View)的开发,那很酷。

[原文] [Bilawal Sidhu]: It's like basically Earth inside of Breal game engine with a bunch of other simulation tech But this is a cool way to just bypass that al together Holy craperoni

[译文] [Bilawal Sidhu]: 它基本上就像是在 Breal 游戏引擎(注:此处可能指 Unreal 虚幻引擎或类似引擎)里的地球,加上一堆其他的模拟技术。但这是一种完全绕过那些步骤的酷方法。天啊(Holy craperoni)。

[原文] [Bilawal Sidhu]: Can I jump up can I fly oh yes I can Oh yes we can Ladies and gentlemen we have achieved liftoff

[译文] [Bilawal Sidhu]: 我能跳起来吗?我能飞吗?噢是的,我可以。噢是的,我们可以。女士们先生们,我们起飞了。

[原文] [Bilawal Sidhu]: Let's see if we can Can I go into one of these buildings what happens no Is that is that the Salesforce tower what is that h too funny But this looks very realistic

[译文] [Bilawal Sidhu]: 看看能不能……我能进入这些建筑物之一吗?会发生什么?不。那是……那是 Salesforce 大厦吗?那是什么?哈,太搞笑了。但这看起来非常逼真。


章节 8:虚拟制片与身份错乱故障

(Virtual Production & The Identity Glitch)

📝 本节摘要

演讲者试图创建一个虚拟制片片场,模拟“镜头后的镜头”。结果模型产生了一个极其怪诞的循环:他在直升机里看到了自己(或者像他朋友 Matt Wolf 的人),甚至感觉到自己变成了直升机的一部分(长出了螺旋桨)。这种“我和我自己被锁在一起”的视觉反馈被演讲者形容为“非常致幻” (trippy) 的恐怖谷体验,。

[原文] [Bilawal Sidhu]: Uh I don't know why that looks like Matt Wolf It kind of likes mixing both of us up I don't know Okay Yeah dude Vibes Vibes bro Dog we filming dog This is how guys Virtual freaking production

[译文] [Bilawal Sidhu]: 呃,我不知道为什么那看起来像 Matt Wolf。它好像喜欢把我们俩搞混。我不知道。好的。是的兄弟,氛围,氛围啊兄弟。哥们,我们在拍片呢,哥们。这就是大家所说的,该死的虚拟制片(Virtual Production)。

[原文] [Bilawal Sidhu]: What am I in a helicopter what am I even Oh yeah I guess I got propellers I'm locked in with myself in a chopper See can I space out of this no dude This is a little uncanny Not going to lie

[译文] [Bilawal Sidhu]: 我是在直升机里吗?我到底在……噢是的,我猜我有螺旋桨。我和我自己被锁在一架直升机里。看看能不能退出去?不,兄弟。这有点恐怖谷效应(uncanny),不说谎,。

[原文] [Bilawal Sidhu]: Oh my god it's me again dude What the hell guys what the heck is going on here this is trippy as hell bro Oh my god I just looked at myself This is crazy

[译文] [Bilawal Sidhu]: 噢,我的天,又是“我”,兄弟。搞什么鬼,各位,这到底是怎么回事?这简直太致幻了,兄弟。噢,我的天,我刚看着我自己。这太疯狂了。


章节 9:软件界面模拟与姿态估计

(Software UI Simulation & Pose Estimation)

📝 本节摘要

演讲者测试了非游戏环境的生成能力,尝试重现 Chrome 浏览器和 Microsoft Word 的界面。他惊讶地发现,模型不仅能模拟鼠标移动,甚至能产生“悬停提示”(tool tip)的交互反馈。随后,他尝试生成一个实时“姿态估计”(Pose Estimation)的场景,虽然渲染效果有些古怪(wonky),但他成功在画面中看到了代表骨架的火柴人和关键点,验证了模型对计算机视觉概念的理解。

[原文] [Bilawal Sidhu]: So I did another one where I was like trying to recreate just a browser like a like Chrome and Microsoft Word And I could do the same thing I could use left and right to move the mouse around and it would even do hover events and stuff like that

[译文] [Bilawal Sidhu]: 我还做了另一个测试,试图重现一个浏览器,比如 Chrome 和 Microsoft Word。我可以做同样的事,我可以用左右键移动鼠标,它甚至会有悬停事件之类的反应。

[原文] [Bilawal Sidhu]: Notice how I gave you the little tool tip as you kind of hovered over it It's really trippy and weird And then at some points basically if I push to the right enough I can like zoom in and out So it's kind of like those images where I was stuck in the zooming motion I moved to another screen on the right which is really trippy Throw this in here and see what happens

[译文] [Bilawal Sidhu]: 注意看当你悬停在上面时,我是如何给你那个小小的工具提示(tool tip)的。这真的很致幻、很奇怪。然后在某些时候,基本上如果我向右推得足够多,我可以放大和缩小。所以这有点像那些我被卡在缩放动作里的图像,我移动到了右边的另一个屏幕,这真的很致幻。把这个丢进去看看会发生什么。

[原文] [Bilawal Sidhu]: Okay So what does your environment look like camera view flying around Let's just see how that looks Does it do something interesting and give me Does it simulate what I'm hoping for yes it does Holy crap Ladies and gentlemen ladies and gentlemen we are doing real time pose estimation on a bunch of characters right here,

[译文] [Bilawal Sidhu]: 好的,那么你的环境看起来像什么?摄像头视角到处飞。让我们看看那是怎样的。它会做些有趣的事并给我……它能模拟我期望的东西吗?是的,它做到了。该死(惊叹)。女士们先生们,女士们先生们,我们正在对这里的一群角色进行实时姿态估计(Pose Estimation)。

[原文] [Bilawal Sidhu]: It's not perfect It really does a little bit of a wonky rendition but let's see if we rotate around if it'll show us all those attributes again like the stick figures and the the points Not exactly Oh there we go Here we kind of see it

[译文] [Bilawal Sidhu]: 它并不完美。它的渲染确实有点古怪,但看看如果我们旋转周围,它是否会再次向我们展示所有那些属性,比如火柴人和那些点。不完全是。噢,有了。这里我们大概能看到了。


章节 10:自然景观与激光雷达风格

(Nature Scenes & Lidar Style)

📝 本节摘要

演讲者尝试生成一个基于激光雷达(Lidar)点云风格的场景,随后进入了一片森林。在此处,他引用了罗伯特·弗罗斯特的著名诗句来形容环境的幽寂。在探索过程中,他意外触发了“自由落体”故障,跌出了地图边界。恢复后,他评价生成的树木像 Google Earth 中的“西兰花树”(网格质量一般),推测这是基于大量摄影测量数据训练的结果,而非纯粹的激光雷达数据。

[原文] [Bilawal Sidhu]: Can we fly around a lightar point cloud that's pretty good Not bad Not bad at all This looks very cool

[译文] [Bilawal Sidhu]: 我们能绕着激光雷达点云飞吗?那相当不错。不错,一点也不错。这看起来非常酷。

[原文] [Bilawal Sidhu]: Let's see if we go off the beaten track What happens will it synthesize something new past this that's always the fun part about this

[译文] [Bilawal Sidhu]: 让我们看看如果我不走寻常路会怎样。会发生什么?它会在之后合成新的东西吗?这总是这东西有趣的部分。

[原文] [Bilawal Sidhu]: I guess like I'm in the in the the woods are lovely lonely dark and deep And there's miles to go before we sleep But I see some light over there

[译文] [Bilawal Sidhu]: 我想我就像是在……“树林美丽、幽寂、深邃,但在睡觉前还有很长的路要走”(注:引用罗伯特·弗罗斯特诗句)。但我看到那边有一些光。

[原文] [Bilawal Sidhu]: Let's see if I can make it over See what's going on over there I can't quite see What the hell is that did I just fall through stuff oh my god

[译文] [Bilawal Sidhu]: 看看我能不能过去。看看那边发生了什么。我看不大清。那是什么鬼?我刚穿过东西掉下去了吗?噢,我的天。

[原文] [Bilawal Sidhu]: Yo I didn't even know you could freef fall See if I can jump back up We're stuck in a freef fall

[译文] [Bilawal Sidhu]: 哟,我都不知道居然可以自由落体。看看能不能跳回去。我们卡在自由落体里了。

[原文] [Bilawal Sidhu]: Let's try this one more time Oh very cool Very cool Not bad at all Not bad at all These look like broccoli trees That's what we call the the meshes of trees in Google Earth

[译文] [Bilawal Sidhu]: 让我们再试一次。噢,非常酷。非常酷。一点也不错。一点也不错。这些看起来像“西兰花树”。我们就是这么称呼 Google Earth 里的树木网格的。

[原文] [Bilawal Sidhu]: Liidar technically would give you much richer rendition than this but it's probably trained on enough photoggramometry that I can't really distinguish Or maybe my prompting thing just sucks I don't know

[译文] [Bilawal Sidhu]: 从技术上讲,激光雷达(Lidar)会提供比这更丰富的渲染,但这可能是在足够的摄影测量数据上训练出来的,所以我真的分不出来。或者是我的提示词写得太烂了,我不知道。


章节 11:微缩赛车与照片记忆重现

(Hot Wheels & Photo Memory Recreation)

📝 本节摘要

演讲者首先展示了一个类似“风火轮”(Hot Wheels)的微缩赛车场景,以第三人称视角在后院驾驶小车。他高度赞赏了车漆上的菲涅尔反射(Fresnel highlights)和光影曝光的动态变化。
随后,他尝试“走进”一张包含多位科技博主(如 Rowan Cheung, Matt Wolfe 等)的旧合影。虽然原本期望展示“记忆捕捉”的潜力,但结果却意外地滑稽:人脸变得模糊,甚至改变了种族特征——他开玩笑说模型把他的朋友 Matt 变成了印度人,且两人长得一模一样。

[原文] [Bilawal Sidhu]: So this is a cool one It's sort of like a Hot Wheels experience You get to like have this third person perspective uh driving around a car in like a backyard Super cool

[译文] [Bilawal Sidhu]: 这个很酷。这有点像风火轮(Hot Wheels)的体验。你可以拥有这种第三人称视角,呃,在后院之类的地方驾驶一辆车。超级酷。

[原文] [Bilawal Sidhu]: So there we go We're off to the races quite literally Let's see if I can manage to keep this on rails barely There we go Notice the exposure shifts as we're going into darker areas as well It's really quite cool

[译文] [Bilawal Sidhu]: 好了。我们真的开始比赛了。看看我能不能勉强保持在轨道上。好了。注意看当我们进入较暗区域时曝光的变化。真的很酷。

[原文] [Bilawal Sidhu]: And you can see my keyboard controls on the bottom left and right and to see what's actually happening The reflections are really really cool The Fresnel highlights on like the car paint very very cool itself

[译文] [Bilawal Sidhu]: 你可以看到我在左下角和右下角的键盘控制,看看实际发生了什么。反射效果真的非常酷。比如车漆上的菲涅尔(Fresnel)高光,本身就非常非常酷。

[原文] [Bilawal Sidhu]: And of course I can totally go off the I can go off unchartered territory right like and you could see that the reflections look all super accurate

[译文] [Bilawal Sidhu]: 当然我完全可以离开……我可以去那些未知的领域,对吧?你可以看到反射看起来都超级准确。

[原文] [Bilawal Sidhu]: Let's see if I can just go through this fucking fence or it'll it'll collide Ah unfortunately Yeah Notice uh you're getting the um barn door lighting effect with like the light peeking through

[译文] [Bilawal Sidhu]: 看看我能不能直接穿过这个该死的栅栏,还是会撞上?啊,不幸的是(撞上了)。是的。注意,呃,你会看到那种挡光板(barn door)照明效果,光线透过来。

[原文] [Bilawal Sidhu]: Let's see if we can actually like fly around this photo Like this is sort of like the memory capture use case right oh god Let's see what happens here I guess this is just letting me zoom in and out of the photo Can I Oh there we go

[译文] [Bilawal Sidhu]: 看看我们是否真的可以飞进这张照片。就像,这有点像记忆捕捉的应用场景,对吧?噢,天啊。看看这里会发生什么。我猜这只是让我放大缩小照片。我能……噢,有了。

[原文] [Bilawal Sidhu]: Holy crap Look guys it's Rowan Schwang Matt Wolf me Lionus Ekinstan and this is Oh you can barely make out the face unfortunately

[译文] [Bilawal Sidhu]: 该死(惊叹)。看各位,这是 Rowan Cheung、Matt Wolfe、我、Linus Ekenstam(注:均为科技圈博主),还有这是……噢,不幸的是你几乎看不清脸。

[原文] [Bilawal Sidhu]: Oh gosh I made him bald No it's a very famous tech YouTuber who will remain unnamed at the moment But yeah this isn't that crazy like stepping inside a photo

[译文] [Bilawal Sidhu]: 噢天啊,我让他秃顶了。不,那是一位非常著名的科技 YouTuber,暂时不提名字。但是是的,这就是像走进照片里一样,是不是很疯狂?

[原文] [Bilawal Sidhu]: Oh gosh it turned both of us Matt and I look the same Oh my god I made Matt Indian

[译文] [Bilawal Sidhu]: 噢,天啊,它把我们俩变得一样了,Matt 和我看起来一样。噢,我的天,我把 Matt 变成印度人了。


本章聚焦于最后的水下场景测试,以及演讲者对 Genie 3 技术的总结评价与致谢。

章节 12:深海探险与结语

(Underwater Exploration & Conclusion)

📝 本节摘要

最后一个测试场景是水下沉船。演讲者对画面中的浮游生物、气泡以及“焦散”(caustics)光影效果赞不绝口,尤其是从水面射下的“上帝之光”。他尝试游向水面但未能成功,随后展示了沉船细节。
在结语部分,他向 Google DeepMind 致谢,评价该技术目前“秒杀”(blows everything else out of the water)他所见过的所有公开同类产品,并鼓励观众点赞订阅,预告了更多关于世界模型和 Genie 技术深度的内容。

[原文] [Bilawal Sidhu]: Let's do an underwater sequence Okay so Wow Whoa whoa whoa whoa whoa whoa This is Oh look at the plankton and the air bubbles Whoa whoa whoa This is really cool

[译文] [Bilawal Sidhu]: 让我们做一个水下片段。好的,那么……哇。哇,哇,哇,哇,哇,哇。这是……噢,看那些浮游生物和气泡。哇,哇,哇。这真的太酷了。

[原文] [Bilawal Sidhu]: Here I can orbit all around You've got amazing costic effects or costics are looking really nice You can see some of the god rays as they're coming off the top

[译文] [Bilawal Sidhu]: 这里我可以环绕四周。你有惊人的焦散(caustic)效果,或者说焦散看起来真的很棒。你可以看到一些从顶部射下来的“上帝之光”(god rays)。

[原文] [Bilawal Sidhu]: See if we can approach the the ship here in time Amazing fishes just floating around having a good time The flipper mechanism is really cool

[译文] [Bilawal Sidhu]: 看看我们能不能及时靠近这艘船。令人惊叹的鱼群就在周围游荡,享受美好时光。那个脚蹼机制真的很酷。

[原文] [Bilawal Sidhu]: Yeah part of me wants to go to the surface I've tried this in other generations I could never make it to the top

[译文] [Bilawal Sidhu]: 是的,我内心有一部分想浮出水面。我在其他几次生成尝试中试过,但我从来没能到达顶部。

[原文] [Bilawal Sidhu]: But check out the shipwreck y'all How cool is this huh let's see Reach for the light Reach for the light Yeah ain't no way But let me orbit around and show you what's in the surrounding So cool

[译文] [Bilawal Sidhu]: 但是快看这个沉船,各位。这多酷啊,哈?让我们看看。伸向光芒,伸向光芒。是的,没门儿。但让我绕一圈,给你们看看周围有什么。太酷了。

[原文] [Bilawal Sidhu]: All right there we go Hey Google Deep Mind thank you so much for the early access I really enjoyed playing with this I am just super excited about a lot more people playing with a world model of this quality

[译文] [Bilawal Sidhu]: 好了,就这样。嘿 Google DeepMind,非常感谢给我这次早期访问机会。我真的很享受玩这个。我只是超级兴奋能有更多人玩到这种质量的世界模型。

[原文] [Bilawal Sidhu]: I mean this blows everything else out of the water at least so far that we've seen publicly that I can talk about publicly But there's so many undiscovered things So jump into it

[译文] [Bilawal Sidhu]: 我的意思是,这绝对秒杀(blows everything else out of the water)其他一切,至少目前为止在我们公开看到的东西里,或者我能公开谈论的东西里是这样。但还有很多未被发现的事物。所以快来体验吧。

[原文] [Bilawal Sidhu]: If you enjoy this video please drop a like drop a comment below you know subscribe for more content like this It really does help help and make a difference and cost you nothing of course

[译文] [Bilawal Sidhu]: 如果你喜欢这个视频,请点个赞,在下面留个言,你知道的,订阅以获取更多这类内容。这真的很有帮助,能带来改变,而且当然不需要你花钱。

[原文] [Bilawal Sidhu]: And if you want to dive deeper into world models check out this video over here And if you want to go deeper into Genie and kind of what it's capable of we've got a video on that right over here as well Belavo signing off and I'll see y'all on the next one Cheers

[译文] [Bilawal Sidhu]: 如果你想更深入地了解世界模型,看看这边的这个视频。如果你想更深入地了解 Genie 以及它的能力,我们也在这边有一个相关视频。Bilawal 签退,我们在下一期视频见。干杯。