Pinned Tweet
Find the best #3DGS tool for your workflow with a simple quiz: 3dgstoolfinder.xraispotlight.com
It filters a database I created of 35 different tools, from casual free apps to on-device reconstruction pipelines to cloud services specialised in 360 video.
You can use it to test new alternative to your current workflow or simply discover something new.
Let me know what you think in the comments below
It filters a database I created of 35 different tools, from casual free apps to on-device reconstruction pipelines to cloud services specialised in 360 video.
You can use it to test new alternative to your current workflow or simply discover something new.
Let me know what you think in the comments below
1
5
28
1,942
Can you replace a drone with a VEEEEEEEEEERY long stick?
This was one of the questions I was trying to answer when using the @insta360 on a 3.2m extendable invisible stick, combined with some ground-level close-up photos taken with my phone.
The idea was to combine the two r the 360 camera was good enough, and whether both types of images could be merged into the same processing pipeline using Spirula.
The answer is yes and no.
My first processing run, on my #DellProMax T2 Tower equipped with an RTX6000, was successful and produced a 1M-splat .ply model at just 200 MB in around 40 minutes.
It was good to see that the 3.2m stick itself didn't seem to cause any problems. And the additional viewpoints definitely give the splat more freedom than a standard ground-level capture.
360 images and the one taken with my phone aligned beautifully making it possible to get area of the scene at a much higher quality... but that came at a cost.
At just 1 M splats, I believe there weren't enough splats to get enough detail everywhere it was actually possible.
But I think I need more splats. 🤔
What would you try: 3M, 5M, or 10M?
Let me know below.
#DellProPrecision #NVIDIA
This was one of the questions I was trying to answer when using the @insta360 on a 3.2m extendable invisible stick, combined with some ground-level close-up photos taken with my phone.
The idea was to combine the two r the 360 camera was good enough, and whether both types of images could be merged into the same processing pipeline using Spirula.
The answer is yes and no.
My first processing run, on my #DellProMax T2 Tower equipped with an RTX6000, was successful and produced a 1M-splat .ply model at just 200 MB in around 40 minutes.
It was good to see that the 3.2m stick itself didn't seem to cause any problems. And the additional viewpoints definitely give the splat more freedom than a standard ground-level capture.
360 images and the one taken with my phone aligned beautifully making it possible to get area of the scene at a much higher quality... but that came at a cost.
At just 1 M splats, I believe there weren't enough splats to get enough detail everywhere it was actually possible.
But I think I need more splats. 🤔
What would you try: 3M, 5M, or 10M?
Let me know below.
#DellProPrecision #NVIDIA
5
2
32
1,792
These are the 12 principles to create believable characters.
Wendelin Reich, founder and CEO of VIRTUAL BEINGS, has spent more than a decade studying how to make digital characters feel alive and found out that to make a character feel alive it need to be:
👀 Observable: ing the player can perceive. Internal states such as a racing heart only become behavior when they produce something observable, such as nervous tapping or knocking over a glass.
🔄 Continuous: Characters are always behaving, even when they appear to be doing nothing. Sitting, sleeping, or standing still are still ongoing behavior rather than an absence of behavior.
🤝 Interactive: Behavior always happens in relation to the character's environment and other agents. A believable character needs to respond and adapt to what is happening around it rather than simply playing predefined animations.
🧱 Constrained: Behavior is shaped by the physical and contextual constraints of the world, including the environment, objects, physics, sound, and the character's own body. The character needs to adapt its behavior to these constraints.
🧩 Sequenced: Behavior consists of sequences that can be reused and adapted. Real humans rely on learned patterns of movement but adjust them to the specific situation rather than generating every movement from scratch.
⚡ Interruptible: Behavior can change unexpectedly. Characters can change their minds, react to new events, or abandon an action while still maintaining a continuous flow of behavior.
🎭 Patterned variation: The same behavior should never look exactly identical every time. Natural behavior contains variation, but that variation follows underlying patterns rather than being completely random.
🪜 Hierarchical: Complex behavior is composed of layers of coordinated actions. For example, a handshake involves the torso positioning the arms, the arms controlling the hands, and the head and eyes coordinating with the other person.
🔀 Parallel: Characters perform multiple behaviors simultaneously. Someone can walk, smoke, look around, and react to another person at the same time. These behaviors may involve different or overlapping parts of the body.
🧠 Cognitively caused: Behavior has an underlying mental cause. What a character does should communicate intentions, emotions, decisions, or other internal states rather than being movement without meaning.
👁️ Monitored: Characters continuously interpret what is happening and use that information to influence their behavior. Their actions are connected to their perception of the situation rather than operating independently of it.
💬 Readable: Other characters, and especially the player, automatically interpret behavior and infer its meaning. A gesture can communicate intention, failure, embarrassment, or emotion even when these states are never explicitly stated.
Check out the full interview with additional links and resources here: youtu.be/Oy47sSNXeYk
Wendelin Reich, founder and CEO of VIRTUAL BEINGS, has spent more than a decade studying how to make digital characters feel alive and found out that to make a character feel alive it need to be:
👀 Observable: ing the player can perceive. Internal states such as a racing heart only become behavior when they produce something observable, such as nervous tapping or knocking over a glass.
🔄 Continuous: Characters are always behaving, even when they appear to be doing nothing. Sitting, sleeping, or standing still are still ongoing behavior rather than an absence of behavior.
🤝 Interactive: Behavior always happens in relation to the character's environment and other agents. A believable character needs to respond and adapt to what is happening around it rather than simply playing predefined animations.
🧱 Constrained: Behavior is shaped by the physical and contextual constraints of the world, including the environment, objects, physics, sound, and the character's own body. The character needs to adapt its behavior to these constraints.
🧩 Sequenced: Behavior consists of sequences that can be reused and adapted. Real humans rely on learned patterns of movement but adjust them to the specific situation rather than generating every movement from scratch.
⚡ Interruptible: Behavior can change unexpectedly. Characters can change their minds, react to new events, or abandon an action while still maintaining a continuous flow of behavior.
🎭 Patterned variation: The same behavior should never look exactly identical every time. Natural behavior contains variation, but that variation follows underlying patterns rather than being completely random.
🪜 Hierarchical: Complex behavior is composed of layers of coordinated actions. For example, a handshake involves the torso positioning the arms, the arms controlling the hands, and the head and eyes coordinating with the other person.
🔀 Parallel: Characters perform multiple behaviors simultaneously. Someone can walk, smoke, look around, and react to another person at the same time. These behaviors may involve different or overlapping parts of the body.
🧠 Cognitively caused: Behavior has an underlying mental cause. What a character does should communicate intentions, emotions, decisions, or other internal states rather than being movement without meaning.
👁️ Monitored: Characters continuously interpret what is happening and use that information to influence their behavior. Their actions are connected to their perception of the situation rather than operating independently of it.
💬 Readable: Other characters, and especially the player, automatically interpret behavior and infer its meaning. A gesture can communicate intention, failure, embarrassment, or emotion even when these states are never explicitly stated.
Check out the full interview with additional links and resources here: youtu.be/Oy47sSNXeYk
0
2
5
597
56% of people think Animated Gaussian Splatting will have the biggest impact on 3DGS in 2027.
The new chapter of the 2027 edition of my Gaussian Splatting Handbook will focus exactly on that 😉
This is your chance to ask anything you would like to know and drive the series of interviews with experts, practitioners and founders I’m about to run.
Just drop the questions in the comments 👇
Let me tell you how I found this out.
When I asked on LinkedIn what could have the biggest impact on 3DGS, only 24% said 360 cameras, and just 13% mentioned SLAM scanners.
There is clearly a lot of hype around animated Gaussian Splatting. So I think it’s time to figure out where we actually stand and where this technology is headed.
One of the biggest challenges is making sure that what I capture in the next edition of the Gaussian Splatting Handbook doesn’t become outdated in a matter of weeks.
That’s why I’m trying to structure the interviews around more than just the latest models, software and technical workflows and trying to capture the knowledge and decision-making process that professionals have gathered through years, sometimes decades, of working with volumetric capture. The things they have learned through experience that help them make better decisions today.
The interviews will follow the workflow:
1- The creation of the initial footage. How do you capture a real performance? What changes when you use a professional camera rig, a high-end setup or even phones? And how can synthetic footage complement or completely replace real footage?
2- Then there is the training and reconstruction itself. How do you take that footage and turn it into a volumetric performance using some of the latest models and approaches? After that comes compression and packaging. How do you turn the result into something that creative professionals can actually use?
3- And perhaps the biggest question: how is 4DGS being deployed in the field today? On the web, on a headset, as part of a game or interactive experience, or simply as an element in a VFX production.
4-There is also the licensing side. Probably one of the less sexy parts of the workflow, but something that becomes very relevant when you start creating projects for clients.
These are the areas I’m planning to cover in the interviews.
What am I missing?
What would you want me to ask the people working on this every day?
Add your questions below. I’ll use them to shape the interviews and the new chapter of the Handbook.
The new chapter of the 2027 edition of my Gaussian Splatting Handbook will focus exactly on that 😉
This is your chance to ask anything you would like to know and drive the series of interviews with experts, practitioners and founders I’m about to run.
Just drop the questions in the comments 👇
Let me tell you how I found this out.
When I asked on LinkedIn what could have the biggest impact on 3DGS, only 24% said 360 cameras, and just 13% mentioned SLAM scanners.
There is clearly a lot of hype around animated Gaussian Splatting. So I think it’s time to figure out where we actually stand and where this technology is headed.
One of the biggest challenges is making sure that what I capture in the next edition of the Gaussian Splatting Handbook doesn’t become outdated in a matter of weeks.
That’s why I’m trying to structure the interviews around more than just the latest models, software and technical workflows and trying to capture the knowledge and decision-making process that professionals have gathered through years, sometimes decades, of working with volumetric capture. The things they have learned through experience that help them make better decisions today.
The interviews will follow the workflow:
1- The creation of the initial footage. How do you capture a real performance? What changes when you use a professional camera rig, a high-end setup or even phones? And how can synthetic footage complement or completely replace real footage?
2- Then there is the training and reconstruction itself. How do you take that footage and turn it into a volumetric performance using some of the latest models and approaches? After that comes compression and packaging. How do you turn the result into something that creative professionals can actually use?
3- And perhaps the biggest question: how is 4DGS being deployed in the field today? On the web, on a headset, as part of a game or interactive experience, or simply as an element in a VFX production.
4-There is also the licensing side. Probably one of the less sexy parts of the workflow, but something that becomes very relevant when you start creating projects for clients.
These are the areas I’m planning to cover in the interviews.
What am I missing?
What would you want me to ask the people working on this every day?
Add your questions below. I’ll use them to shape the interviews and the new chapter of the Handbook.
2
0
13
1,200
We live in amazing times.
We can now capture reality with consumer-grade devices like 360 cameras, reconstruct entire scenes in very high fidelity on a laptop and without any need for an internet connection.
I went out this weekend to capture a scene here in Leiden using this sta360 to get more coverage and viewpoint freedome.
I then processed it directly on my new #DellProPrecision 7 sent by #Dell, equipped with an #NVIDIA RTX Pro 3000 running Spirula as the reconstruction software.
Everything was flawless, and I will be sharing the process and final result this week.
It’s fascinating to see how quickly this technology has moved. And I think we’re only at the beginning. From here, things will become faster and more accessible. Mobile devices will increasingly be able to achieve similar levels of quality, making reality capture something that anyone can do.
The interesting question now is what we’re going to do with these models and with the ability to capture the world around us at this level of detail.
We can now capture reality with consumer-grade devices like 360 cameras, reconstruct entire scenes in very high fidelity on a laptop and without any need for an internet connection.
I went out this weekend to capture a scene here in Leiden using this sta360 to get more coverage and viewpoint freedome.
I then processed it directly on my new #DellProPrecision 7 sent by #Dell, equipped with an #NVIDIA RTX Pro 3000 running Spirula as the reconstruction software.
Everything was flawless, and I will be sharing the process and final result this week.
It’s fascinating to see how quickly this technology has moved. And I think we’re only at the beginning. From here, things will become faster and more accessible. Mobile devices will increasingly be able to achieve similar levels of quality, making reality capture something that anyone can do.
The interesting question now is what we’re going to do with these models and with the ability to capture the world around us at this level of detail.
8
4
90
5,290
Viewing city-scale data on #SmartGlasses has never been easier.
I got the chance to try Reality Import, a new platform that does a LOT of the heavy lifting:
- Find relationships between huge data sets using AI
- Creates geolocated maps
- Compile projects that can be directly or Unity
- Deploy on SPECS or other AI glasses
In the case you see, I just dropped all parking spots and chargiing station in the Netherlands, and built a simple app that finds the closest available spots.
If you want to give it a shot, leave a comment below or drop me a DM.
I got the chance to try Reality Import, a new platform that does a LOT of the heavy lifting:
- Find relationships between huge data sets using AI
- Creates geolocated maps
- Compile projects that can be directly or Unity
- Deploy on SPECS or other AI glasses
In the case you see, I just dropped all parking spots and chargiing station in the Netherlands, and built a simple app that finds the closest available spots.
If you want to give it a shot, leave a comment below or drop me a DM.
0
0
0
396
After testing the new Echo-3 from @SpAItial_AI I am convinced Gaussian Splatting is a solved problem. So what's next?
Well, maybe not exactly solved yet for all use cases, but I think it will be in a matter of months.
Generating Gaussian Splats has become easier than ever. You nsumer grade 360 camera, a phone, or with an AI world model that can generate high-quality environments that are also measurable and accurate in the real world, particularly for robotic simulation.
Of course, there are still applications where you need higher fidelity or better quality around specific objects. But those problems will probably be solved too.
The bigger problem now is the application.
The technology is moving much faster than companies are adopting it.
Now we need to figure out what comes next.
How do we turn these environments into meaningful, valuable experiences that people and companies can actually use?
That, to me, is where the interesting part of Gaussian Splatting starts and where not so many have an answer for.
Well, maybe not exactly solved yet for all use cases, but I think it will be in a matter of months.
Generating Gaussian Splats has become easier than ever. You nsumer grade 360 camera, a phone, or with an AI world model that can generate high-quality environments that are also measurable and accurate in the real world, particularly for robotic simulation.
Of course, there are still applications where you need higher fidelity or better quality around specific objects. But those problems will probably be solved too.
The bigger problem now is the application.
The technology is moving much faster than companies are adopting it.
Now we need to figure out what comes next.
How do we turn these environments into meaningful, valuable experiences that people and companies can actually use?
That, to me, is where the interesting part of Gaussian Splatting starts and where not so many have an answer for.
12
9
112
8,132
I nearly made a mess of my interview with Shadnam Khan by focusing on the wrong thing.
I went into my conversation with @multiset_ai COO, focused on the technical details: centimetre-level accuracy, LiDAR scanners versus 360 cameras, and the trade-offs between different ways of t I took away was that how you build the map matters less than I expected, provided there’s a reliable positioning platform behind it.
Different capture methods still have their strengths and limitations. But the goal is to turn those captures into a shared spatial reference that people, devices and robots can use to understand where they are and what’s around them.
That also shifted how I think about the bottleneck. For many of these applications, the technology seems ready. The harder part is understanding a company’s workflow and finding a use case that delivers a measurable return on investment.
Helping a technician find the right asset or reducing the time needed for an inspection gives that map a practical purpose. That’s where I’d put more of the attention.
I went into my conversation with @multiset_ai COO, focused on the technical details: centimetre-level accuracy, LiDAR scanners versus 360 cameras, and the trade-offs between different ways of t I took away was that how you build the map matters less than I expected, provided there’s a reliable positioning platform behind it.
Different capture methods still have their strengths and limitations. But the goal is to turn those captures into a shared spatial reference that people, devices and robots can use to understand where they are and what’s around them.
That also shifted how I think about the bottleneck. For many of these applications, the technology seems ready. The harder part is understanding a company’s workflow and finding a use case that delivers a measurable return on investment.
Helping a technician find the right asset or reducing the time needed for an inspection gives that map a practical purpose. That’s where I’d put more of the attention.
0
1
4
459
For my submission to the #WorldBackupProject challenge, I couldn’t think of a better answer than something that has already disappeared.
To join this competition from #XGRIDS, you don't even need a PortalCam.... but more on that below 😉
What you see is Fatherhood, a monumental d by Colombian artist Stinkfish in Amsterdam in 2015. This project is particularly meaningful to me because was the one that got me started in spatial computing and had a huge influence and impact on everything I've done since.
Back in 2019, I had the chance to work with the @StrArtMuseumAms and its founder Anna Stolyarova to preserve Fatherhood before the mural was demolished.
SAMA is not a traditional museum. Its collection lives and dies on the streets of Amsterdam. Around 70% of the artworks the museum had documented had already been removed, demolished or painted over. Fatherhood was one of those pieces.
So when I saw the World Backup Project asking: “What would you save before it’s gone for good?” I immediately thought of Fatherhood.
The competition is built around exactly this idea: capture something from the real world, reconstruct it as a 3D Gaussian Splat, and then use the XGRIDS ecosystem to turn that capture into something new.
And you don't need an @XGRIDS_OFFICIAL scanner to participate. You can capture with their new mobile app, use video and images you process in LCC studio or any of the SDKs for web Unity or Unreal.
Since now LCC Studio supports video as input, I simply went back to the drone footage captured in 2019 to reconstruct the splat, and used LCC Studio to clean it up, edit it, create transitions and turn the result into the video you're seeing here.
And I'm very grateful to SAMA and Anna for giving me the opportunity to work on this project all those years ago. It genuinely changed the direction of my work.
Maybe that’s what a spatial backup is really about: keeping a door open to places that no longer exist.
To join this competition from #XGRIDS, you don't even need a PortalCam.... but more on that below 😉
What you see is Fatherhood, a monumental d by Colombian artist Stinkfish in Amsterdam in 2015. This project is particularly meaningful to me because was the one that got me started in spatial computing and had a huge influence and impact on everything I've done since.
Back in 2019, I had the chance to work with the @StrArtMuseumAms and its founder Anna Stolyarova to preserve Fatherhood before the mural was demolished.
SAMA is not a traditional museum. Its collection lives and dies on the streets of Amsterdam. Around 70% of the artworks the museum had documented had already been removed, demolished or painted over. Fatherhood was one of those pieces.
So when I saw the World Backup Project asking: “What would you save before it’s gone for good?” I immediately thought of Fatherhood.
The competition is built around exactly this idea: capture something from the real world, reconstruct it as a 3D Gaussian Splat, and then use the XGRIDS ecosystem to turn that capture into something new.
And you don't need an @XGRIDS_OFFICIAL scanner to participate. You can capture with their new mobile app, use video and images you process in LCC studio or any of the SDKs for web Unity or Unreal.
Since now LCC Studio supports video as input, I simply went back to the drone footage captured in 2019 to reconstruct the splat, and used LCC Studio to clean it up, edit it, create transitions and turn the result into the video you're seeing here.
And I'm very grateful to SAMA and Anna for giving me the opportunity to work on this project all those years ago. It genuinely changed the direction of my work.
Maybe that’s what a spatial backup is really about: keeping a door open to places that no longer exist.
1
0
13
791
Last week I posted a full walkthrough from scanning with an Insta360 X6 to creating a robot simulation in Isaac Sim. Make sure to check it out on the XR AI Spotlight channel.
The Insta360 X6 can turn a 2 min 360° video into a usable Gaussian Splat, with the Insta360 app offering hat delivers surprisingly good results, while Scaniverse produces sharper, cleaner and metrically accurate reconstructions at a cost.
For best results, record in 8K/30fps, move slowly, and capture two passes at different heights. The main limitations are fine details, plain white surfaces and floating artefacts, but with the right processing pipeline, a consumer 360 camera can quickly produce a usable 3D environment.
Last but not least, running a robotic simulation in Isaac Sim on the #DellProMax T2 Tower equipped with an #NVIDIA RTX 6000 was a breeze. Special thanks to #DellTech for providing such a capable rig
The Insta360 X6 can turn a 2 min 360° video into a usable Gaussian Splat, with the Insta360 app offering hat delivers surprisingly good results, while Scaniverse produces sharper, cleaner and metrically accurate reconstructions at a cost.
For best results, record in 8K/30fps, move slowly, and capture two passes at different heights. The main limitations are fine details, plain white surfaces and floating artefacts, but with the right processing pipeline, a consumer 360 camera can quickly produce a usable 3D environment.
Last but not least, running a robotic simulation in Isaac Sim on the #DellProMax T2 Tower equipped with an #NVIDIA RTX 6000 was a breeze. Special thanks to #DellTech for providing such a capable rig
1
3
12
1,188
The Interaction SDK has a new gaze simulation sample I highly recommend you test. #VisionPro clearly demonstrates how powerful eyes are as an input system.
For interaction with the UI it is incredibly handy (even though sometimes a bit inaccurate if fit or calibration are not s to object manipulation, I always wanted to "physically" grab, move around and explore and on AVP I always feel like I am interacting "remotely".
Quest has a much more "physical" approach and I hope we'll be able to get the best of both worlds soon 😎.
#Quest #MetaHorizonPartner
For interaction with the UI it is incredibly handy (even though sometimes a bit inaccurate if fit or calibration are not s to object manipulation, I always wanted to "physically" grab, move around and explore and on AVP I always feel like I am interacting "remotely".
Quest has a much more "physical" approach and I hope we'll be able to get the best of both worlds soon 😎.
#Quest #MetaHorizonPartner
2
0
21
1,751
My video on the Insta360 X6 just dropped 👉 youtu.be/S2ZLmeZl2E8
Watch or read the summary 👇
The Insta360 Enterprise is a surprisingly capable #3D scanner, especially when paired with the right processing workflow.
I tested two options:
1. @insta360 Spatial Capture: Built 360 app, it turns a 2 min 360° video into a Gaussian Splat for free, with 20 monthly reconstruction credits. The results are impressive for a completely free cloud workflow and a 2-hour turnaround time. Fine details, plain white surfaces and floating artefacts can be problematic.
2. @NianticSpatial: @scaniverse processes the original X6 footage in the cloud, producing significantly sharper details and cleaner reconstructions (but prepare to pay around 20 USD for a 5 min reconstruction). Especially if you add a tag at the beginning of the scan, the result is metrically scaled and quite accurate based on my tests. You also get an USDZ to export where the splat and the collision mesh are packaged together, opening up applications such as robotics simulation. I did a test on Isaac Sim on my #DellProMax T2 Tower equipped with an #NVIDIA RTX 6000 and I even managed to get a robot dog going down the stairs.
A few practical tips: record at 8K/30fps, walk slowly (like if you are holding a full glass of water that you don't want to spill), and prioritize two passes at different heights over covering too much ground in one go.
A consumer 360 camera, a short walk and the right processing pipeline can take you from video footage to a usable 3D environment.
Watch or read the summary 👇
The Insta360 Enterprise is a surprisingly capable #3D scanner, especially when paired with the right processing workflow.
I tested two options:
1. @insta360 Spatial Capture: Built 360 app, it turns a 2 min 360° video into a Gaussian Splat for free, with 20 monthly reconstruction credits. The results are impressive for a completely free cloud workflow and a 2-hour turnaround time. Fine details, plain white surfaces and floating artefacts can be problematic.
2. @NianticSpatial: @scaniverse processes the original X6 footage in the cloud, producing significantly sharper details and cleaner reconstructions (but prepare to pay around 20 USD for a 5 min reconstruction). Especially if you add a tag at the beginning of the scan, the result is metrically scaled and quite accurate based on my tests. You also get an USDZ to export where the splat and the collision mesh are packaged together, opening up applications such as robotics simulation. I did a test on Isaac Sim on my #DellProMax T2 Tower equipped with an #NVIDIA RTX 6000 and I even managed to get a robot dog going down the stairs.
A few practical tips: record at 8K/30fps, walk slowly (like if you are holding a full glass of water that you don't want to spill), and prioritize two passes at different heights over covering too much ground in one go.
A consumer 360 camera, a short walk and the right processing pipeline can take you from video footage to a usable 3D environment.
0
3
9
923
The Meta Ray-Ban Display are available today in France, Italy and Germany.
Should you get one?
If you have been hesitant about Meta because of its policies, privacy concerns, or the idea of wearing a camera on your face, the release of these glasses will not make ANY have a camera and, if tampered with, can potentially be used to record people without their consent. This is a serious issue that I feel we should tackle as a society as more and more wearable devices reach the market (whether they are pushed by Meta or other companies).
So if you were already against this category, I don't think the Meta Ray-Ban Display will change your mind.
For developers, though, I think this is a really interesting moment. There is an opportunity to start building and learning on a device that combines displays, cameras, audio and different forms of input in a wearable form factor, on a device owned by millions of people.
You can build both web and native experiences and start experimenting with what actually makes sense on these glasses. For me, that is probably the most interesting part: seeing what developers will do with them now that they are becoming more widely available.
Should you get one?
If you have been hesitant about Meta because of its policies, privacy concerns, or the idea of wearing a camera on your face, the release of these glasses will not make ANY have a camera and, if tampered with, can potentially be used to record people without their consent. This is a serious issue that I feel we should tackle as a society as more and more wearable devices reach the market (whether they are pushed by Meta or other companies).
So if you were already against this category, I don't think the Meta Ray-Ban Display will change your mind.
For developers, though, I think this is a really interesting moment. There is an opportunity to start building and learning on a device that combines displays, cameras, audio and different forms of input in a wearable form factor, on a device owned by millions of people.
You can build both web and native experiences and start experimenting with what actually makes sense on these glasses. For me, that is probably the most interesting part: seeing what developers will do with them now that they are becoming more widely available.
3
1
8
694
Trying new apps and experiences, and seeing how developers and creatives tackle this new medium, always fascinates me.
In this lens on SPECS, you “scan” an object or surface and turn it into a fictional material sphere that you can leave around for others to find.
When I think iations of this idea, the ability to leave a message, object or note on top of the real world for someone else to discover is fascinating.
I hope more developers experiment with this idea and that we figure out a good moderation layer 🤔
In this lens on SPECS, you “scan” an object or surface and turn it into a fictional material sphere that you can leave around for others to find.
When I think iations of this idea, the ability to leave a message, object or note on top of the real world for someone else to discover is fascinating.
I hope more developers experiment with this idea and that we figure out a good moderation layer 🤔
7
4
36
2,111
Interact with #GaussianSplatting on the web right here: louenpottier.github.io/lagsplat.html
LaGSplat combines #3DGS with a learned model of how an object moves.
It learns the object’s motion from a few seconds of video and represents that motion in a compact latent space. The Gaussian from that state in real time.
This means you can grab or push the object, and the system calculates how it should respond and updates the 3D scene.
The researchers have also put the result into an interactive web viewer, giving us a glimpse of how 3DGS could evolve on the web: scenes that you can look around and interact with directly in the browser.
LaGSplat combines #3DGS with a learned model of how an object moves.
It learns the object’s motion from a few seconds of video and represents that motion in a compact latent space. The Gaussian from that state in real time.
This means you can grab or push the object, and the system calculates how it should respond and updates the 3D scene.
The researchers have also put the result into an interactive web viewer, giving us a glimpse of how 3DGS could evolve on the web: scenes that you can look around and interact with directly in the browser.
6
15
96
5,080
These are crazy times to talk about the intersection of #XR and #AI.
Come join me Wednesday the 7th, at 9 am for the opening keynote at the @nextappdevcon in Berlin.
I will be sharing a framework I put together that against all my expectations, is holding up despite how fast ng 😉
Come join me Wednesday the 7th, at 9 am for the opening keynote at the @nextappdevcon in Berlin.
I will be sharing a framework I put together that against all my expectations, is holding up despite how fast ng 😉
0
0
0
370
Don’t trust (#AI) ads.
@andrewpprice aka #Blender Guru, has recently posted a video where he recreated "viral ads" of Gen AI products, and the gap was HUGE.
Some of the major #3D and AI video Players ended up apologising for how misleading the ads were and ended up removing is one of the reasons you should be very conscious about what you see, and I spend so much time testing the technology I talk about. It is easy to share a trailer video from a new feature but until I see it working in practice i am always skeptical.
Before I share a tool, I want to know:
- Does it actually work?
- Does it work as advertised?
- Is it reliable?
- What are its limitations?
It is overwhelming. There are so many tools coming out, and so much content showing what they can supposedly do.
But you can't take all of that at face value. So my advice is simple: watch the Blender Guru video, and find other humans you trust to do the discovery and testing for you.
Because right now, perhaps more than ever, there is a big difference between seeing something in an ad and actually using it.
That's also why I keep testing these tools myself and sharing what I find every week.
If you want to follow along, follow me.
@andrewpprice aka #Blender Guru, has recently posted a video where he recreated "viral ads" of Gen AI products, and the gap was HUGE.
Some of the major #3D and AI video Players ended up apologising for how misleading the ads were and ended up removing is one of the reasons you should be very conscious about what you see, and I spend so much time testing the technology I talk about. It is easy to share a trailer video from a new feature but until I see it working in practice i am always skeptical.
Before I share a tool, I want to know:
- Does it actually work?
- Does it work as advertised?
- Is it reliable?
- What are its limitations?
It is overwhelming. There are so many tools coming out, and so much content showing what they can supposedly do.
But you can't take all of that at face value. So my advice is simple: watch the Blender Guru video, and find other humans you trust to do the discovery and testing for you.
Because right now, perhaps more than ever, there is a big difference between seeing something in an ad and actually using it.
That's also why I keep testing these tools myself and sharing what I find every week.
If you want to follow along, follow me.
0
2
7
633
When it comes to findings, only 22% of people were looking at cloud-based solutions. The remaining 75% were more interested in local solutions, with a prevalence of Windows machines.
1
0
0
24
When it comes to input sources, there is a staggering number of people looking for ways to process 360 footage, followed by images and only 21% favouring video. This was quite an interesting finding, as I expected video to be the clear winner
1
0
0
62
Another of the parameters I selected was the budget preference. Half of the responders were looking for a free solution and there was a solid 24% who were open to paying the right price for the right tool (even without a prior trial period). I was expecting this ratio to change s that had different levels of technical expertise (asked in a separate question), but that didn’t have any effect on the budget preference.
0
0
0
65
I built a Quiz to help you find the best 3DGS tool for your workflow:
3dgstoolfinder.xraispotlight.com
More than 200 people responded and the findings of what people are looking for are in the 🧵
3dgstoolfinder.xraispotlight.com
More than 200 people responded and the findings of what people are looking for are in the 🧵
2
0
5
692
The filtering works in 3 steps
- First, it filters out tools that don't fit your setup, based on things like where you want to run the reconstruction and what input you have.
- Then I score the tools that remain based on technical depth (max 6 points), cost (max 6 points) and extra features (max 3 points).
- In case of a tie I picked a winner based on my own testing and experience. This luckly happened very rarely.
3dgstoolfinder.xraispotlight.com
- First, it filters out tools that don't fit your setup, based on things like where you want to run the reconstruction and what input you have.
- Then I score the tools that remain based on technical depth (max 6 points), cost (max 6 points) and extra features (max 3 points).
- In case of a tie I picked a winner based on my own testing and experience. This luckly happened very rarely.
3dgstoolfinder.xraispotlight.com
1
0
0
45