GenAI + 3D || Spaitial CTO || Ex-Google Labs + Google Beam || PhD University of Washington || Alaska enthusiast || ricardomartinbrualla.com
Seattle, WA
- Tweets 240
- Following 774
- Followers 3,319
- Likes 902
Pinned Tweet
Creating immersive 3D worlds has never been easier! Super excited to share Echo, Spaitial's first 3D foundational model!
Converting interiors to rococo style is my favorite application so far 😃
🚀 Announcing Echo — our new frontier model for 3D world generation.
Echo turns a simple text prompt or image into a fully explorable, 3D-consistent world. Instead of disconnected views, the result is a single, coherent spatial representation you can move through freely.
This is part of a bigger shift in AI: from generating pixels and tokens to generating spaces. Echo predicts a geometry-grounded 3D scene at metric scale, meaning every novel view, depth map, and interaction comes from the same underlying world — not independent hallucinations.
Once generated, the world is interactive in real time. You control the camera, explore from any angle, and render instantly — even on low-end hardware, directly in the browser. High-quality 3D world exploration is no longer gated by expensive equipment.
Under the hood, Echo infers a physically grounded 3D representation and converts it into a renderable format. For our web demo, we use 3D Gaussian Splatting (3DGS) for fast, GPU-friendly rendering — but the representation itself is flexible and can be easily adapted.
Why this matters: consistent 3D worlds unlock real workflows — digital twins, 3D design, game environments, robotics simulation, and more. From a single photo or a line of text, Echo builds worlds that are reliable, editable, and spatially faithful.
Echo also enables scene editing and restyling. Change materials, remove or add objects, explore design variations — all while preserving global 3D consistency. Editing no longer breaks the world.
This is only the beginning. Echo is the foundation for future world models with dynamics, physical reasoning, and richer interaction — environments that don’t just look right, but behave right.
Explore the generated worlds on our website and sign up for the closed beta. The era of spatial intelligence starts here. 🌍
#Echo #WorldModels #SpatialAI #3DFoundationModels
Check it out: spaitial.ai/
6
11
133
14,944
NeRFs on Google Search 🤯!
When you look for *shoes* on google.com, you can now find interactive 360 spins for some items generated using NeRFs! These are reconstructed from 5-10 views that were already available in Google Shopping!
only available in US & on mobile
17
141
867
406,654
We present nerfies! We use selfie videos to create 3d free-viewpoint portrait visualizations of yourself using Deformable NeRFs! More details at nerfies.github.io and below (1/8)
9
117
718
My team at Google Seattle is hiring researchers & engineers to work on improving and scaling neural rendering for object reconstruction. We're a small, focused team trying to bring the latest advances in 3D reconstruction to products at Google. Please reach out if interested!
19
60
464
How can we use test-time compute for spatial understanding? 🤔
In InterPose, we propose to repeatedly sample generative video models to help two-view pose estimation and reconstruction, by leveraging the video models' keyframe interpolation abilities.
A 🧵... (1/8)
1
31
244
21,692
After 8 wonderful years at Google, I decided to leave Google and try something new!
Working at Google was a dream come true for me, and 4 internships and 8 years later, I couldn't feel more grateful about my experience.
A 🧵
9
4
221
26,622
Project Starline is publishing a technical paper about the system at Siggraph Asia 2021 in December. @jlawrence176 and @danbgoldman also made a terrific presentation video, check it out!
paper: research.google/pubs/pub5090…
presentation video: piped.video/watch?v=kDgToq5a…
1
58
208
I'm taking 6 months off work to cross Alaska on foot, paddling and biking with Salomé. We are starting on Monday!🚶♂️🚶♀️🗻🌊
If you are curious about what it will be like, follow us at @north2arctic and our blog north2arctic.com! 1/3
9
4
196
So happy that Project Starline was just announced! Using real-time 3D imaging, compression, spatialized audio and a lightfield display, Project Starline makes you feel like you are across someone else!
blog.google/technology/resea…
13
19
156
While updating NeRF-in-the-Wild, we added 3 more scenes to the paper: Prague's Old Town Square, Hagia Sophia & Taj Mahal.
They came out pretty well! I could not resist myself and created a demo reel / virtual world tour video showcasing the results at 720p60hz! Enjoy! 😃
We've updated our NeRF-in-the-Wild paper and video with more scenes. I particularly like the Taj Mahal results!
piped.video/watch?v=mRAKVQj5… nerf-w.github.io/
2
17
143
Excited about 3D GenAI? There’s something super exciting brewing…
If you know of researchers, or 3D / ML / infra engineers, there are positions open in Munich and London. Reach out!
4
14
109
12,259
My larger team at Google is hiring 2024 PhD interns working on generative media (video, image, 3D, etc)! Come work with great collaborators across Google.
See intern call below for details, links are available in the attached doc.
docs.google.com/document/d/e…
5
18
95
29,406
My group at Google, Project Starline, is looking for interns for summer 2022! Please consider applying!
Application link: bit.ly/3HG60bc
Project Starline announcement: blog.google/technology/resea…
3
13
91
Nerfies code is out! 🚀 I love the self-contained video2nerfie colab demos, no tools or installs needed!
Next: make my bio pic a nerfie 🤓
Today we're releasing the code for Nerfies!
We've also included a Colab demo so that you can try out a basic version of our method right in your browser.
Project Page: nerfies.github.io
Github: github.com/google/nerfies
[1/n] 🧵👇
4
3
75
Our latest work takes a DSLR to the 19th century to reproduce historical figures as they looked like! 📸
What did Abraham Lincoln really look like? Our new project simulates traveling back in time with a modern camera to rephotograph famous subjects.
Web: time-travel-rephotography.gi…
Video: piped.video/eNOGqNCbcV8
w/ @ceciliazhang77, @RealPaulYoo, @rmbrualla, @jlawrence176, Steve Seitz
8
4
66
NeRF without per-scene training! Image-based (volumetric) rendering using ray transformers.
Training NeRFs per-scene is so 2020. Inspired by image based rendering, IBRNet does amortized inference for view synthesis by learning how to look at input images at render time. 15% drop in error, 80% fewer FLOPs than NeRF. Great work @QianqianWang5! ibrnet.github.io
1
4
64
Check out our latest work! NeRF + appearance modeling + transient reconstruction + uncertainty handles the variation found in internet photos 😃
Our paper, “NeRF in the Wild”, is out! NeRF-W is a method for reconstructing 3D scenes from internet photography. We apply it to the kinds of photos you might take on vacation: tourists, poor lighting, filters, and all. nerf-w.github.io (1/n)
1
6
53
We are working on making these more widely available, and are currently hiring PhD Research Scientists! Ideal candidates should have experience in generative modeling (2D or 3D) and/or 3D reconstruction. If interested, please send your CV to reconstruction-at-scale@google.com!
1
4
43
11,738
Nice smooth meshes in our CVPR paper! It was a fun collaboration with the awesome folks at TUM.
Excited to share @DejanAzinovic's Neural RGB-D Surface Reconstruction #CVPR2022!
We reformulate NeRF to operate on a signed distance field to obtain high-quality 3D surface reconstructions.
piped.video/iWuSowPsC3g
dazinovic.github.io/neural-r…
@rmbrualla @DanBGoldman @JustusThies
6
48
No shadow left behind. Or how to virtually declutter your house! We use a deep version of differential rendering to remove objects and their shadows from scenes.
Fun to work with Edward Zhang (UW), @jannekontkanen and Brian Curless!
grail.cs.washington.edu/proj…
1
6
52
HyperNeRF featured in @twominutepapers! Congrats Keunhong!
3
45
Do you use @GoogleColab or interactive notebooks? Do you make videos?
Checkout mediapy -- a package that makes reading, writing and showing images and videos in notebooks a bliss!
import mediapy
mediapy.show_video([f for f in frames])
colab.sandbox.google.com/git…
2
8
45
Deformable Neural Radiance Fields //Nerfies is on arxiv! 😃
abs: arxiv.org/abs/2011.12948
pdf: arxiv.org/pdf/2011.12948
Also, check the 7m video on youtube with more examples and applications!
piped.video/watch?v=MrKrnHhk…
1
7
40
Salome and I are going back to Alaska to try to finish our trip we started 2 years ago — I’m so looking forward to this adventure and to share it with you once we’re back.
I’m sad that I’ll be missing CVPR due to this trip — see you at ICCV!
Two years ago we tried to cross Alaska from Ketchikan to Kotzebue, but only managed to do two thirds of it.
This summer we are going to try to finish the trip, from Healy all the way to the Arctic to Kotzebue biking, hiking and paddling 🚵♀️🥾🚣♀️
1
33
5,456
Salomé and I finished our trip! What an amazing life experience, one that will be hard to forget. Thanks everyone who helped us make this possible! 🙏
Now we will slowly go back to normal life — back at work in October. Hoping to seeing many of you at virtual ICCV!
We finished our trip! A few days ago we made it Icy Bay Lodge, where we stopped in the spring, connecting the dots between Ketchikan, Alaska’s southernmost city, and Healy, that’s just north of Denali National Park.
What a trip!! We enjoyed so much Alaska’s wilderness!
(1/5)
4
31
My favorite part is how stable the reconstructions are overall, thanks to reasoning only in 3D without 2D nets. You can render with the same quality at day time...
1
1
24
So cool that folks can try Starline at Siggraph! Happy to hear you had a great experience!
Wow. Just tried the Google Project Starline demo, amazing work by the team!
It’s hard to describe how cool and futuristic the experience felt, and I can’t wait for the public to try it out.
My impressions as someone working in 3D and ML:
1
1
23
2,362
📢 Check out GeLaTO: Generative Latent Textured Objects 📢
We combine coarse proxy geometry, neural textures, a neural rerenderer, and a generative model to model *real* objects with complex appearances like eyeglasses frames, with real-time rendering capabilities!
Generative Latent Textured Objects - GeLaTOs! - are a new approach for rendering objects with complex geometry and appearance using simple proxies, neural textures, and U-nets! gelato-paper.github.io (1/6)
3
6
20
Nerfies also allow for some fun effects, like taking virtual vacations using high quality matting. (7/8)
1
19
We found a neat way to improve NeRF, check it out!
When cameras view scene content at a variety of scales/resolutions, NeRF produces renderings that are aliased or blurry. We tried to fix this with mip-NeRF!
@jon_barron @BenMildenhall @PeterHedman3 @rmbrualla
jonbarron.info/mipnerf/
2
20
I feel very lucky and thankful to work with such a talented team, including @KeunhongP, @philipphenzler, and @kemelmi and other many great collaborators across Google! @jon_barron @_pratul_
1
16
8,752
PS: this was a cool article abour trip that appeared in the Mountaineers magazine!
ricardomartinbrualla.com/dat…
1
16
2,290
Tonight at 7pm PST, Salomé and I will be giving an online talk at the Mountaineers Walking the Wild series about our trip crossing Iceland on foot in 2016!
If you are avaible, consider joining the zoom call! If you miss it, happy to share slides!
mountaineers.org/locations-l…
15
1,725
If you are excited about stepping out of the comfort zone into the growth zone with me, please reach out! I'll be at NeurIPS in Vancouver starting tomorrow!
Looking forward to sharing more about my new project soon!
4
13
2,756
I then had the opportunity of doing fun research in NeRFs, taking them to all around-the-world in Nerf-in-the-Wild and to my personal photos in Nerfies.
This eventually led to me to create and lead a team that worked on landing NeRFs in Google Search, especially shoes :-)
1
13
2,042
I’ve been working on it for 4.5 years, so I’m over the moon that it’s been announced! It’s such a fun project, touching computer vision, graphics and machine learning. Looking forward to continue working on telepresence! 😃
1
14
We finally made it to Juneau. The 10 day section turned into 13 days!
We spent 6 days paddling and walking the beach on our way to Juneau. We had beautiful weather but challenging wind conditions! Check out our blog post:
north2arctic.com/of-sun-and-…
13
This was announced in SearchON back in September :-)
See #4 here:
blog.google/products/shoppin…
1
11
7,000
To do that, Deformable NeRFs recover a per-frame deformation field that brings the scene into correspondence with a canonical model. We use a nerf-like MLP that takes 3D points to the canonical frame of reference and is conditioned by a latent observation code. (3/8)
3
13
I still look at these results every day to see if I can find artifacts! It's so realistic!
I wonder if the technique could reproduce this waterfall of clouds though... piped.video/mvgBJpdcgrg
Excited to show off our new project on single-image cinemagraphs. Our method automatically turns a _single image_ into a seamlessly looping video!
Website: eulerian.cs.washington.edu
Video: piped.video/watch?v=4zKliOMi…
w/ Brian Curless, Steve Seitz, Rick Szeliski
More in thread! [1/5]
1
13
Alaska 1 - Ricardo and Salomé 0. Next round in a few days :)
Our first section turned into a training run after unexpected winter conditions forced us to turn around. See more in the blog: north2arctic.com/lingering-w…
1
12
Really cool work from my neighbor Mira Slavcheva!! 😃
The cooking dataset makes me *so hungry*! Seems like it is a pandemic-inspired dataset, like the ones we got for nerfies.
Congrats y'all! @LvZhaoyang @MZollhoefer #futureofcooking
We are proud to introduce "Neural 3D Video Synthesis", our recent team efforts in FRL Research that can enable high-fidelity high resolution wide-angle 3D video synthesis.
Arxiv: arxiv.org/abs/2103.02597
Page (with full video): neural-3d-video.github.io/
1
2
12
I'll miss reading about the great work happening across Computer Vision and Graphics labs -- can't wait to catch up when I get back! 3/3
1
10
Some fun stats:
- 3030 miles in 174 days = 17.5 miles per day
- 620 pounds of food packed, containing 1.38 M calories
- 23 pounds of peanut butter
- 12 kg base weight (no food) including >5kg of boating gear
- 13 days: longest stretch between resupplies (hopefully!!) 🤯
2/3
1
10
In 2016, I joined Google to work on Project Starline, and I loved building the first full-fledged demos of the 3D teleconference system with an incredibly talented team, and later on exploring real-time neural rendering with amazing collaborators.
blog.google/technology/resea…
1
1
9
2,141
Nonetheless, it’s a bit funny to follow the announcement from Yakutat, Alaska, where the internet is not fast enough for me to watch the live feed 🙃
11
I'm pretty adventurous in my personal life, e.g. my COVID project with my wife was to cross Alaska human-powered, on foot, packraft and bicycle.
And now, it's time for me to take on a new adventure in my professional life.
1
9
1,454
I'm immensely thankful to all my mentors at Google, my colleagues over the years, my research collaborators, and especially, my reports, for pushing me to be the best version of myself. Thank you!
1
9
2,931
We went back on the trail!
I never thought a drone would be so helpful for hiking, not only we used it to navigate the countless crevasses in a massive glacier, but also to scout the slopes with less vegetation in a steep valley to minimize the bushwhack :)
We crossed the Alaska Range between the Parks and Richardson highways. It was stunning, but challenging with many rivers and glacier crossings!
Next, we will continue heading East towards Slana.
Read more here: north2arctic.com/alaska-rang…
9
Hard to wrap your mind around NeRF and all its followups? Check this comprehensive post by @fdellaert!
2020 was the year in which *neural volume rendering* exploded onto the scene, triggered by the impressive NeRF paper by Mildenhall et al. I wrote a post as a way of getting up to speed in a fascinating and very young field and share my journey with you: dellaert.github.io/NeRF/
9
To create the shots in the demo reel, I load the SfM point clouds into blender, and create smooth camera paths, it feels like being a movie director (which is pretty hard!). We then render the frames offline, as it takes 5 minutes per frame on 8 GPUs 😱
2
8
A feature film made by rendering NeRFs! More immediately, real-time reconstruction and editing tools
9
If you are interested in high quality 3D human capture at Google, please DM me! We are considering candidates for internships for next summer and full-time positions. (9/8+2)
1
1
9
My favorite National Park in the US: Wrangell - St. Elias National Park
(also the largest!)
We finally made it beautiful McCarthy, Alaska, following the trail that many miners used to get across the Wrangells during the late gold rush of Chisana! What a scenery!!
More in the blog: north2arctic.com/mccarthy/
8
En #CVPR, cuánta gente!
Hoy a la 1pm estaré en el panel de mentores del LatinX workshop representando a Google! Ven y pregúntame lo que quieras! O hazme un DM en Twitter si no estás en la conferencia! #visionporordenador
1
1
7
Two simple steps:
1) Apply here if you're graduating by 2025: google.com/about/careers/app…
otherwise here:
google.com/about/careers/app…
2) Send us a message with your CV to genmedia-interns-2024@google.com
Applications before January 15 will be given priority.
1
5
1,623
Not the easiest and quickest way to travel, but the packrafts take us to truly incredible places! Today we saw whales, orcas and the aurora from the lighthouse 🤩
New post on our blog! We didn’t make it to Juneau yet, but we are happy to be at the incredible Five Finger Lighthouse.
north2arctic.com/little-beri…
1
7
In particular, we generate videos starting in one input view, and ending in the other, and compute pose estimates from an augmented input set containing the input views and some generated in-between frames, that help bridge the gap between the input views.
(2/8)
1
1
7
718
It is hard to stay still during selfie captures, and small motions break the photometric consistency, which leads to severe artifacts in NeRF. Deformable NeRFs model the small frame-to-frame motion to generate a consistent reconstruction. (2/8)
1
6
Invisibility Shield Kickstarter 🤯
kickstarter.com/projects/inv…
Thanks @danbgoldman for sharing!
2
6
We are continuing our Alaska trip, this time heading East and South from Denali National Park towards where we left it in Icy Bay.
Super excited to be back in Alaska! 🏔🐻🏕
We had to interrupt our trip due to an injured foot followed by a viral infection. We made the best out if it by spending a week in beautiful Cordova and then going to see our families in Europe. We are now back in Alaska and will start hiking tomorrow.
north2arctic.com/return-to-a…
5
Please see more details at inter-pose.github.io/
Congrats to Ruojin Cai (@ruojin8) who led the project! 🥳She is the last intern I hosted at Google.
Joint work with my amazing (ex-)colleagues at Google @jasonyzhang2, @philipphenzler, @zhengqi_li, and @Jimantha.
(8/8)
5
1,093
Although the practicality of calling large video models during pose estimation is limited due to their excessive compute requirements, this work points towards new synergies between pose estimation and generative video models…
(6/8)
1
3
457
The pace of research in this field is mindblowing. I can't wait to see the new products and experiences that will come out of this.
5
Agreed!
Coordinate-based models is a good name for models that store "stuff" in MLPs. Or neural fields too. @jon_barron
5
We started blogging again about our second trip to Alaska, which we did a year ago. Expect one post a week for a few weeks :-)
Funnily enough, I somehow miss the mosquitoes.
1/3 A year ago, we went back to Alaska to finish our trip. We finally started blogging about it, and already published two posts! We plan to publish a post a week, until we make it to the Arctic Ocean (we did!).
Why did we go back?
north2arctic.com/alaska-keep…
6
1,323
This project is the fruit of Keunhong Park's (@KeunhongP) amazing internship at our group in Google Seattle. The project was fully remote, and working on selfie reconstruction seemed pretty appropriate during socially-distancing COVID times :-) (8/8+2)
1
4
Because video models sometimes generate implausible (but cool!) interpolations, like ones with morphing artifacts or shot cuts, we propose a simple method to select high quality video interpolation samples: ...
(3/8)
1
1
4
609
... where pose estimates on static datasets like RE10K or DLVD3D generate training data for pose-conditioned video models (like in CAT3D), and in turn keyframe interpolation from video models provides a valuable prior for pose estimation.
(7/8)
1
4
1,206
🔥 ¿LO HAS VISTO YA? 🔥
Con la Inteligencia Artificial es posible reconstruir con todo lujo de detalles MONUMENTOS EN 3D a partir sólo de darle imágenes a analizar 🏛🤖
Más info aquí 👇
✅ LINK: piped.video/d35Yjz_tpBA
¡Compártelo! 🔄💘
2
4
It was created by my colleague Hugues Hoppe, and I have enjoyed using it internally at Google. So happy he open-sourced it.
4
Today I’ll be at the LatinX workshop at #CVPR in the mentoring session at 1pm representing Google. Swing by and ask any question, or DM me on Twitter if you are not at the conference.
3
Oops, here is a fixed doc link!
docs.google.com/document/d/e…
4
1,175
Remembering our great time in Chisana with Arthur.
The footage from the 360 camera on the tip of the wing was amazing (yes, we were in that little bumpy plane), and Arthur's flying skills are next level 🛩️🧑✈️
At the end of our trip in Alaska, our friend Arthur picked us up in his bushplane and hosted us for 2 weeks in beautiful gold rush "town" of Chisana. We had a blast there, only topped by the spectacular flights over the Wrangells.
More info in blogpost: north2arctic.com/chisana/
2
4
New post on our blog about the short section from Wrangell to Petersburg, and long weather delay in Petersburg, where we also got vaccinated. Tomorrow we set off to Juneau!
north2arctic.com/fishing-tow…
4
Very cool work from our intern Xiaoming! Check it out!
Wondering how to easily relight an object?
Inverse rendering, maybe the first thing that comes to mind, is brittle and expensive due to differentiable Monte Carlo rendering.
Check out IllumiNeRF for simple, effective 3D relighting without it!
illuminerf.github.io
(1/n)
3
1,163
The video was created using Internet photos from the PhotoTourism dataset. Images are about 1 megapixel and have a ton of issues, including noise & occluders. I can't only imagine what NeRF-W could do with 10+ megapixel DSLR pics!
cs.ubc.ca/research/image-mat…
1
3
... for a given video interpolation sample, we compute multiple pose estimates by randomly sampling different sets of in-between frames from the generated video, and choose the video interpolation sample with the most consistent pose estimates.
(4/8)
1
1
3
551
This is a neural version of billboard clouds commonly used in computer graphics! In the paper, we also show it can generalize to more complex objects using AtlasNet patches too!
3
I got your message! I’m assuming it was that huge grizzly bear that came running towards us super excited! :) Woohoo, congrats!
3
Thanks to all the great people who worked on this project: @KeunhongP, @utkarshsinha, @jon_barron, @_sofien_, @danbgoldman, and Steve Seitz! (10 / 8 + 2)
3
My spellchecker just corrected arxiv to arctic. Surprisingly appropriate! @north2arctic
3