Miris CPO Will McDonald on Streaming 44 GB of 3D in Under a Second

Miris opened its public beta on March 24, 2026, a little over a year after it first showed spatial streaming in public. The pitch has stayed consistent: upload a 3D asset, and Miris streams it to any device as radiance field data that reconstructs on the client, rather than as a decimated mesh download or a rented cloud GPU rendering frames. What has changed since March is that the beta is now running in production on other people's websites. The 3D marketplace CGAxis streams its catalog previews through Miris, and the collaboration platform Cavrnus is integrating the streaming layer into its shared review spaces.

I sat down with Will McDonald, Miris's Chief Product Officer, in a back room at SIGGRAPH 2026 in Los Angeles. McDonald joined Miris in February 2025 after running engineering for Unity's game engine group as Vice President of Engineering and a stint as a General Manager at AWS. Last year at the same show, I got the first Miris demo on a phone. This year the demo included a 44 GB asset that showed up before I could register it loading.

The conversation below has been lightly edited for length and clarity.

Michael Rubloff: I know you just started accepting glTF, but I'd love to catch up on what's been the latest at Miris.

Will McDonald: Since you last talked to Sean, we've been up to quite a bit. Namely, we launched beta back in the late March, April timeframe. That's been really exciting, because we've taken the approach where you're not going to our site and hitting the "contact sales" button. Instead, you can sign up today and actually use our software and give it a try yourself. That's been really illuminating for us, just to see some early patterns in how people are using the software, the type of content they're bringing in, what's resonating and what's not. The whole purpose of beta is to learn a lot from customers. We've gotten quite a few organic customers coming in, as well as key partners and customers we can continue to work with, some of which have gone public already, even with beta software out there.

Michael: Like Cavrnus?

Will: Yes, Cavrnus is a partner of ours. Another one quite recently is a 3D asset marketplace called CGAxis. You can go to their website now and, as you scroll, see a litany of the assets they sell actively streaming using Miris behind the scenes. That's been great for them.

Generally, we've been pretty excited about the 3D marketplace play, because a lot of these assets are rich in detail but ultimately trapped in traditional single-frame renders or video turntables. What we've been hearing from a lot of these marketplaces, and the creatives submitting to them, is: we want to show off all this work, but right now the people making buying decisions in these marketplaces are left to review the image, and that's often not enough. Being able to say "click on this asset, see it in real time, interact with it, zoom in, check out all the detail, and then make a buying decision" is a really nice bridge into why we have a lot of conviction in retail as somewhere we can make a big impact. What retailers and marketplaces are seeing is that if you can get people into a 3D experience quickly, they see three to five times the conversion to sale relative to 2D experiences, and 20% lower returns. It makes a really material difference if you can get people into the 3D experience.

Michael: You're giving them the capability to make much more informed buying decisions. But the other half of that is that people have a general association with 3D as being a heavy item, something that takes a bit more to do properly. I wanted to see if you could talk about how Miris creates something that defies that expectation.

Will: It's a few things. The heaviness of 3D is a real thing, on both the creation side and the delivery side. Miris is focused on delivery, but in order to distribute and deliver things properly, you have to understand what's happening upstream in the creative process.

Historically, delivering 3D has been one of two things. Either I take the content I have, in Maya, Houdini, whatever, convert it into glTF or GLB, and deliver that to the web. The drawback is that converting it to GLB and having it still look good relative to the source is not an automated process, and probably won't be for quite a while, if ever. It requires highly trained artists to massage that output into something that holds on to all the detail and material properties of the source. That's a lot of cost for people trying to deliver 3D, because there's a lot of labor-intensive work to get to something that looks good.

The second way of delivering 3D content is pixel streaming: I spin up infrastructure in the cloud, GPUs, I have a GPU per user, and I stream the output to the user. The issue is, as you might have seen in the news, these GPUs are not easy to find and are pretty expensive. Scaling that to millions of people in parallel isn't viable. Even if those GPUs existed, you'd be bankrupt before you got to see the value of it. It just doesn't work at that scale. Right now it's better for about a hundred users and under.

And going back to glTF, the other thing is that once you do get it into glTF, downloading it when you're trying to have the 3D experience can take several seconds, upwards of ten seconds. What we've heard from retailers and marketplace folks is that if you can't show the 3D asset within about two seconds, people churn out of the experience. We're a dopamine nation right now. If you can't get someone into an experience very quickly, they're going to move on.

So for Miris, that's meant we've focused on how to get really rich, high-detail 3D, and how to get it to people really fast. Sometimes it's better to be lucky than good. The timing of all the exciting research around radiance fields, Gaussian splatting, Radiant Foam, and so on has really come to a head over the last couple of years, as you know better than anyone. We've been able to leverage a lot of that research, plus a lot of our own proprietary approaches, to hit a really interesting value proposition, where we can deliver this content fast, 200 milliseconds or less to first view, but also retain all the high quality you often lose with glTF. We're doing that because we're not delivering things as traditional meshes anymore. We're delivering, in effect, a volumetric output. But from a user's standpoint, as I look at things on the web, I'm not looking at it and saying "this is different from the 3D I see today." It's meant to look effectively the same, even if the format is different.

The other exciting thing about all the progress in the radiance field field, if you will, is that it's afforded us a lot of interesting things around compression and incremental delivery. It's been really exciting to see all the innovation, particularly in our R&D group, in creative ways to deliver this content, even at multi-gigabyte scale. I'll show you a few examples where we have one asset that's 44 gigabytes that we can deliver in under a second, very reliably, on cell phone networks, on low-power devices. That's a very new thing. It's been really exciting talking to all these brands, retailers, marketplaces, even physical AI companies, about breaking down a lot of these legacy barriers and delivering 3D content this way.

Michael: It really needs to meet the general population's expectation. If it's not better, or at the very least as good as 2D, they won't make the effort to switch to something fundamentally new.

Will: That's right. glTF has always had a bit of this problem, and you've probably seen it countless times. You go to a marketplace or somewhere with a 3D experience, you look at the photos or renders of the object and think, okay, that looks good. Then you click into the 3D version and it reads differently, because they've had to decimate it so much and the materials look more plasticky.

What we've heard from a lot of brands in particular is: we know 3D brings a ton of value, all the things I said about conversion and lower return rates, they see all that and have confirmed it. But because their brand can't be accurately represented in 3D using legacy technology, they've stuck with photos and video. If I'm Louis Vuitton, or a manufacturer of a large-scale digital twin or a turbine, it's so important that it's represented the right way. For them, brand is everything. If their leather looks like plastic, it's a total no-go. The same is true for the Nikes and Kate Spades of the world. We've heard this over and over again.

The other part of your question is upstream of delivery. In the small bit of travel I've done on the SIGGRAPH floor so far, you'll notice there are a lot of generative 3D AI companies on the floor this year. That's a relatively new thing over the last year or two. There's going to be a set of 3D that will always be craftsmanship, hand-built, and that continues forever. I don't see that being downshifted in any way. But what we're seeing at the same time is that the barrier to entry to create 3D is dropping to near zero, and that is a very new thing. I went to a four-year college to learn all these tools, Blender, Maya, and so on, because creating high-quality 3D was a very technical task. Now you're seeing capture companies like XGRIDS, where you can use devices to capture really high-fidelity 3D, and then generative AI tools like Meshy, Tripo, TRELLIS, the list goes on, where you can prompt or send images and create pretty compelling 3D. There's still room to improve for a lot of these models, but the whole point is that the barrier to entry is now "can you type what you want?" That's a pretty amazing place to be.

What we're seeing is a tidal wave of 3D already coming to its crest. Because there's so much 3D, and the complexity keeps rising, what usually follows a drop in the barrier to content creation is distribution. What I often tell people is that video went through this in the early to mid 2000s. It used to be very hard to create video. Then camcorder hardware and video editing suites came down in price and became easy to use, and what followed were things like YouTube, Netflix, Twitch, TikTok. I think 3D is having that moment now. The barrier to entry is low, and now it needs a distribution channel. That's why I always say we're positioned downstream of the tidal wave that's coming, where you need to distribute all of this at scale.

Michael: You're the last-mile delivery.

Will: That's right. Honestly, a big part of why I was excited to join Miris is everything I learned at Amazon about how critical last-mile delivery is. I was at Unity for a while looking after engineering for the game engine group. Distribution and delivery, how you get content in a variety of forms to where it needs to be, ends up being the most critical bit of all of this. What good is content if you can't get it where it needs to be? The saying goes that history doesn't repeat, but it rhymes. 3D is rhyming with video and what it went through. You start to see these patterns, and I think Miris is going to be in a really good spot to catch this wave.

Michael: Could we take a look at some of the files?

Will: [Opens a demo player on his laptop.] Here's a little player we created to showcase some of these assets. This one is a ZBrush asset. It's fairly large and detailed, but as I zoom in, all this detail is being dynamically streamed in. One of the benefits of Miris is that we're only streaming what you need to see at any given time, which lets us be memory-efficient and performance-efficient. Normally, loading an asset of this detail into memory all in one go, you'd be sitting waiting a while for it to download, and your machine, particularly a low-power mobile device, would be chugging on it, so you wouldn't get smooth playback.

That 44-gigabyte asset I mentioned is this one right here. You'll notice it showed up effectively instantaneously. It does not feel like this was 44 gigabytes, and that's a bit of the magic. As I zoom in, you actually can't see the detail streaming in, because that's again part of the magic.

A key to this is the memory efficiency, which is really important for things like physical AI use cases. Imagine a warehouse where I'm training robots virtually, a 3D warehouse that's a kilometer's worth of data. Jamming all of that into VRAM is painful, sometimes not possible, or you're taking up all the VRAM for that single asset. Being able to stream really large environments like that lets people running training get more simulations out of a single GPU, because VRAM is no longer the bottleneck. As a robot traverses the world, we can stream only what it needs to see at any given time and evict the rest. That's been really interesting to talk about with robotics companies and physical AI companies at large. It's unlocked by the adaptive nature of what Miris provides.

Here are a few other examples, on the digital twin side. This is about 3.5 gigabytes of data. We can take really complex, highly detailed machinery, engines and so forth, and represent them in high-quality detail. You can zoom in and see the individual parts. It's all about providing 3D as an experience without compromise, while benefiting from the fact that we can deliver it super fast and at billion-user scale, which for retail in particular is key.

Michael: Thank you for telling me more about Miris. It's been awesome to follow along with the progress. Literally last year at SIGGRAPH I got the first demo from Joe and Sean on a phone, so it's cool to see how far it's come.

Will: It's been a long way since then. Those were the very early days. Fast forward to today, and the sheer number of assets we can process is what stands out. Miris has taken a different approach. I know you in particular have been quite involved with a lot of the capture work, which I think has a huge opportunity to become bigger and bigger, and we're really interested in that space too. We started on the other side of it. All these companies sit on synthetic data already and keep generating more, either through AI or by hiring well-trained craftsmen to build 3D content. For us, it's always been about how we unlock all the value of what people are sitting on today, and then how we take all of these photos and videos and let people get more value out of those as well. It'll be really interesting as we continue to dig into how we can help capture companies get more value and deliver more of that content to users. It should be an exciting journey coming up.

Michael: I'm very excited to follow along with it.