Jul 1, 2026 · 42 min · 14 segments
Martin Piala, CEO and co-founder of Blurry, joins the show to break down what it actually takes to host, share, and scale Gaussian Splatting models on the web. With a background in building…
Martin PialaGuest
Gabriele RomagnoliHostGaussian splatting and the web, I think it's a marriage made in heaven, right? So we all agree, I think it's the place where we are seeing this new 3D paradigm landing and having the most success.
Yet there are different ways in which these models, environments, scene, object can be displayed.
What are the parameters or things that I need to consider when publishing, sharing, and hosting these splats, this scene on the web?
Well, if you want to host and share your goals and spotting models on the web, well, I would, I would advise you to pick your viewer or pick the part that you trust very, very, very well.
There's like a lot of different viewers out there and well, the biggest and the, you know, the most important factor you might want to consider is whether the viewer of your choice can even support the files that you that you are producing in terms of their size and quality because you might have a ginormous files uh which not every browser or every viewer can handle and there's only really a handful of viewers out there that can handle really really big files especially if you are scanning like a whole factory or a warehouse or you know parts of a city essentially so i think that that is that is the main one yeah
When we're talking about ginormous, what are we talking about? Because I think that there's also another aspect that is like the ginormous is like file size, splat count, because there are different elements to it.
So in a sense, they are related, even though the splats might be organized a little bit differently in the file, which makes the file smaller or bigger.
But really, what do we care about is the splat count, because the splat count is what makes it really difficult to render in browser, especially on, you know, older smartphones or laptops or, you know, weaker devices, essentially.
So when we then go back to giant hormones, when things get challenging to be taken into account, are we talking about 1 million, 3 million, 10 million splats? Can you give us some benchmarks so that people can understand what they are dealing with?
I would say anywhere above, you know, two and a half million, that's where we are starting to enter dangerous territory.
This dangerous because it will work very nicely on the machine of the creator, because usually people who create splats have really busy computers, so it will work well.
But it will not work as nicely on, you know, smartphones or tablets of your clients and you might be sending your spots to.
It really depends on the viewer, but around two, two and a half million is where I would start thinking
So you mentioned indeed, does it support these large file sizes? That's been, of course, one of the parameters to take into account.
What else? And how does that maybe the different viewers, like, for example, Blurry or Supersplat maybe differs or maybe many more, right? Right now, there are so many solutions to host it.
Gaussian splatting and the web, I think it's a marriage made in heaven, right? So we all agree, I think it's the place where we are seeing this new 3D paradigm landing and having the most success.
Yet there are different ways in which these models, environments, scene, object can be displayed.
What are the parameters or things that I need to consider when publishing, sharing, and hosting these splats, this scene on the web?
Well, if you want to host and share your goals and spotting models on the web, well, I would, I would advise you to pick your viewer or pick the part that you trust very, very, very well.
There's like a lot of different viewers out there and well, the biggest and the, you know, the most important factor you might want to consider is whether the viewer of your choice can even support the files that you that you are producing in terms of their size and quality because you might have a ginormous files uh which not every browser or every viewer can handle and there's only really a handful of viewers out there that can handle really really big files especially if you are scanning like a whole factory or a warehouse or you know parts of a city essentially so i think that that is that is the main one yeah
When we're talking about ginormous, what are we talking about? Because I think that there's also another aspect that is like the ginormous is like file size, splat count, because there are different elements to it.
So in a sense, they are related, even though the splats might be organized a little bit differently in the file, which makes the file smaller or bigger.
But really, what do we care about is the splat count, because the splat count is what makes it really difficult to render in browser, especially on, you know, older smartphones or laptops or, you know, weaker devices, essentially.
So when we then go back to giant hormones, when things get challenging to be taken into account, are we talking about 1 million, 3 million, 10 million splats? Can you give us some benchmarks so that people can understand what they are dealing with?
I would say anywhere above, you know, two and a half million, that's where we are starting to enter dangerous territory.
This dangerous because it will work very nicely on the machine of the creator, because usually people who create splats have really busy computers, so it will work well.
But it will not work as nicely on, you know, smartphones or tablets of your clients and you might be sending your spots to.
It really depends on the viewer, but around two, two and a half million is where I would start thinking
So you mentioned indeed, does it support these large file sizes? That's been, of course, one of the parameters to take into account.
What else? And how does that maybe the different viewers, like, for example, Blurry or Supersplat maybe differs or maybe many more, right? Right now, there are so many solutions to host it.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.
