Real Clay and Digital Ice Cream at Uber Eck

When Kayak needed Greek-inspired ice cream statues to melt beautifully, Uber Eck skipped the one-take dairy gamble. The team sculpted the figures in clay, scanned them, refined them in ZBrush, then controlled the collapse in Houdini MPM and rendered the final cream drama with Redshift in Solaris.

Niklaus Hofer (LinkedIn | Behance | ubereck.com) is a Swiss director and motion designer based in Munich, working at Uber Eck. His work spans motion design, animation and 3D, with projects including Kayak Premium Ice Cream, TOCA ME Opening Titles, Discovery ID TIPS and DIS CONNECTED.

Sebastian Boyamba (LinkedIn | Behance | boymaba.io) is a Munich-based interactive designer and motion designer working on 3D animation, interactive experiences and installations. His tools of choice are Houdini, Cinema 4D, TouchDesigner and Unreal Engine.

DP: What was the weather like outside when you first started working professionally with ice cream?

Niklaus Hofer: The weather was still very wintry when we started in February 2025. We definitely weren’t in the mood for ice cream yet. Perfect Accident sent their request back in late 2024. The project took quite a while to complete; the main production period was from February to July 2025.

DP: Before we get into the how, let’s start with the what. What was the briefing from Kayak and Perfect Accident, and what was the core idea behind the spot?

Niklaus Hofer: Over the past few years, we’ve worked on many projects together with Perfect Accident, mostly show openers for Al-Hurra, an Arab-American network. We both really value this collaboration, where we start exchanging ideas early on and develop concepts together. 

The core idea was to build a Greek-inspired sculpture out of ice cream and deliberately cause it to melt. The Client requested two sculptures: one depicting a man gazing longingly into the distance, and the other a couple embracing closely.

Right from the start, there were plans to make the whole thing “for real.” Some initial attempts looked very promising, but they had one major problem: you only get one chance to melt it perfectly, and on top of that, ice cream isn’t exactly known for holding its shape under studio lighting :)

Sebastian Boyamba: The story came from the Greek agency. It was basically a “seize the moment before it passes”-hookline. The sculptures represented typical Greek moments, and we started with about five. The Client chose the two strongest.
The “shoot it in real time” approach had an even bigger issue. Not only was it impossible to take multiple shots, but the melting ice cream also looked unappetizing. Therefore, a more glamorous approach was the right choice.

DP: Before we discuss anything else, how many kilos of ice cream did you bring into the studio for research, how much ended up in front of the camera, and how much ended up inside the team?

Niklaus Hofer: Since the production of the commercials dragged on a bit due to approval loops and such, we were lucky that by the time we got around to creating the ice shaders, the temperatures outside had already warmed up considerably. I’d say we’ve eaten the equivalent of the amount of ice cream we would have needed for the actual sculpture.

For this project, we started by working with sketches; to do this, we created simple geometries in Cinema 4D (since we already knew we would be modelling the entire piece in clay) that were already the correct dimensions. The sculptures were also designed to be approximately 30-40 cm tall. We then created the sketches based on these simple geometries and worked with the Client to determine the direction we wanted to take.

Once we had settled on a sketch and direction, we started building small clay models, about 8 to 10 cm tall. This allowed us to move on to the next round of feedback with the Client. They were quickly satisfied with the direction, so we were able to start working on the actual sculpture relatively quickly.

Since I (Niklaus) am currently taking a sculpture class at the “Volkshochschule,” we quickly decided to build the sculpture “IRL”. This has the advantage that the sculpture will be physically accurate (since it is a physical object), and, to be honest, because I’m actually almost faster with clay than with ZBrush. Plus, the clay we’re using already has a texture that’s very reminiscent of ice cream.

Sculpting the first figure took about 10-12 hours (including Client feedback). Due to the scale, we decided not to fully model the faces. Instead, we left the facial features flat and added them later in ZBrush. In the first sculpture of the seated man, for example, the head is about 3 cm high, which makes it very challenging to add the finest details.

The sculpture was then placed on a turntable, and we set up a scanning studio here in the office and created a 3D model using Reality Capture based on approximately 200 to 300 photos. While Sebastian began remeshing and cleaning up the first sculpture to prepare it for melting in Houdini, I had already started working on the second sculpture.

Once we had the first cleaned-up model, we imported it into ZBrush, where I adjusted the sculptures’ facial expressions and fixed minor details at the Client’s request. At the same time, Sebastian was mapping the sculptures in Houdini with heatmaps to melt them in a targeted, controlled manner. This workflow allowed us to continue modifying and optimizing the sculpt at a late stage.

The statue

DP: The sculptures were first built physically out of clay. That sounds wonderfully analogue. Why did you choose that step?

Niklaus Hofer: One reason we chose to build the sculpture out of clay is that we quickly realized these sculptures work (or look realistic) if they are physically accurate. After all, real sculptures also feature incredibly fascinating techniques that people have been using for hundreds of years to depict figures in complex poses. For example, the technique involving a tree trunk or a pedestal, or the use of a sword. Many ancient sculptors found ways to depict an outstretched arm without it collapsing under its own weight.

If we had created this as a purely 3D-sculpted figure, there would always have been a risk of depicting things that a real stone figure could never do, and in our opinion, that would have ruined the illusion.

The other thing was that, especially in this age of AI, we just really wanted to do it the real way. It was also this hybrid workflow, combining a physical model with 3D-scanned geometry, and the ability to stage a melt in Houdini perfectly. And we’re simply passionate motion designers; we see the art of our craft precisely in using different media and applying them creatively.

Not to mention that this project involved many iterations, with feedback from clients and agencies. And when you’re producing professionally, you have to control every single detail, no matter how small.

For the sculpture, we used PRGM refractaria ‘Gris’ clay. We used the same clay in the course; it has a fairly coarse chamotte content. This improves the drying and firing properties of the ceramic. For our purposes, the sculptures were not fired, as we only needed them as a base for our 3D scan. If you want to fire them, you need a temperature of 1240–1300°C. 

DP: If you build a figure in clay, it is no longer infinitely editable. Was that a limitation, an advantage, or just part of the process?

Niklaus Hofer: Funnily enough, there are actually tons of opportunities to start over, even when sculpting by hand. I learned this the hard way when my teacher simply tore off the head of the first sculpture I made in class, after I’d already spent 3 to 4 hours working on it. Unlike stone, clay is an additive material, meaning we can always add to and remove it. With stone, of course, that would be a different story.

What fascinates me about sculpting is that the techniques were developed tens of thousands of years ago and haven’t changed all that much to this day. Even when you create a sculpture in ZBrush, for example, you start with the basic form. I would say that’s where you should invest most of your time. That’s where you lay the foundation for the sculpture, especially in terms of anatomy. 

Only once the foundation is in place should you move on to the next level of detail, and then to the next desired level of detail. But a sculpture doesn’t become better simply because you’ve modelled every toenail, or haven’t. Ultimately, these are design decisions; the foundation must be right for every figure, whether it is realistic or abstract.

We did run into some physical limitations, such as with the faces. I want to point out that the decision to make the faces as detailed as they are in the final commercial wasn’t ours. I think they work quite well as they are, though I could have done without them, because they only contribute to the sculpture to a limited extent. But as I said, those are design decisions. And clients are still clients :D

At the same time, it was obviously great for our production process to be able to work with a digital sculpture relatively quickly, since our final product will also be entirely digital. I’d say that if it had been a purely physical sculpture, I would have spent another 3 to 4 hours on it before calling it “final” and firing it. Since ZBrush also gave us options, such as slightly moving the foot, we were about 80% done at that point.

Photogrammetry 

DP: At some point, the clay figures for the statue were standing on the table. What exactly did you do next?

Sebastian Boyamba: There isn’t much to say about this process because it’s so good and easy nowadays. Reality Capture was the tool of choice because I’ve been doing a lot in Unreal for the last few years, and it was already installed on my computer. 

We used a turntable and a quick studio light setup with a full-frame DSLR, then took way more photos than we needed at a much higher resolution, and let it run.

DP: After RealityCapture, you presumably had a point cloud, a mesh, and a texture. What happened next in ZBrush?

Sebastian Boyamba: The mesh accuracy was already pretty good from Reality Capture, but they were extremely high poly. Since ZBrush has no problem with that, we made basic adjustments and added some details in ZBrush. Then, we remeshed and cleaned up the models in Houdini. With proper UVs, we could refine the model again in ZBrush on a texture basis.

For remeshing and UVs, we used native Houdini and SideFX Labs tools. You can automate this very well in Houdini. First, close the holes procedurally. Then, since the ice cream is very organic, I used VDBs for remeshing and smoothing. I deleted the small islands by measuring the area attribute. I used AutoUVs from SideFX Labs in combination with the curvature attribute, which gave me good enough results. In a few spots, I had to scale the UVs to make them work, but that was also only semi-manual, with some selections I did by hand.

So it wasn’t linear. It was a scan into ZBrush with basic adjustments and add-ons, then cleanup and remeshing plus UVs in Houdini, then back to ZBrush for final details, and finally back to Houdini.

Let’s Sim! 

DP: Now we have the model in ZBrush. For the melting, the project presumably moved into Houdini. What did that handover look like?

Sebastian Boyamba: We prepared the meshes in Houdini. For exchange formats, we used OBJ and Alembic when we had attributes. Nik provided me with rough poly groups from ZBrush. I then painted about an additional five detailed attribute maps in Houdini using attribute paint to create melting variations and mix materials later on.

We had to find the right mesh density for the simulation while also maintaining the details. In Houdini, you can always transfer attributes, including the UVs, which helps a lot. 

DP: Ice cream melts. Sounds simple. What was your first thought when setting up the melting simulation?

Sebastian Boyamba: In the end, it was an MPM solver, but I did a lot of tests early on, including with Vellum Fluid and Flip. The melting proof of concept was done even before we decided to sculpt in clay. First, I melted a rubber toy and a sculpture model from the Internet.

As mentioned earlier, it was a mix of many Attribute maps. I did the hand-painted ones, as well as some maps with the Pyrosource spread in a volume. To control everything, you have to combine them. Since the MPM is very fast, you can also iterate very quickly with low point counts.

There was also a test shooting with melting ice cream that was not very appetising, but it was a reference. We could see that ice cream melts from the top down (as the sun usually is in the sky, studio lights in the tests as well), from thin to thick, and from the outside in.

Also, it’s not only the mesh melting but also the melting in the shader. When ice cream becomes liquid, it loses its roughness and becomes more glossy as the water in the mass forms a thin, coating-like layer. Done with Attribute Maps and a Shader mix.

In some simulations, the physically plausible was too morbid, like falling limbs and heads, so we had to control the melting process there.

DP: To make the statue melt not only physically, but appetisingly, what does Houdini’s setup look like in the end?

Sebastian Boyamba: We ended up using the MPM, which handled the chunks much better than Flip. Temporal smoothing was challenging because we could not afford very high-resolution VDBs.

The control of what melts and when is entirely determined by attribute maps. Some are hand-painted, some are simulated, and some come from the topology, like curvature.

One thing to keep in mind is that you always work with two meshes: the sculpted one and the one from the MPM simulation. You can mix them and use the original, distorted mesh up to a certain melting point, and the smoothed mesh from the melted state. Since you always have the UVs and all the attribute maps, you have full control over which mesh you use for which state or in which place because you can mix them and blend the seams (there was also a seam attribute), and the shader mix helps as well.

DP: Were there differences between the ice cream flavours? Does chocolate melt differently from strawberry, or is that just the kind of question people ask after staring at simulations for too long?

Sebastian Boyamba: The flavours with chunks melted a bit differently. This was important to the Client, so we did a secondary simulation of some chunks after the main simulation. However, we avoided using a sorbet flavour because it would have melted differently. I mean, it wouldn’t have been a problem with the setup because it’s highly controllable with an attribute map, but at some point, the budget has to fit the effort.

Ra-Ra-Ra-Rendering

DP: How many iterations were needed, and which parameters did you adjust until the melting felt right?

Sebastian Boyamba: The feedback was not homogeneous. Some people thought that realism was extremely important and would have liked air bubbles to form, for example. However, this was too unappealing. Others wanted more droplets, which would have been better done in Flip. It was a diplomatic balancing act to strike the right balance between realism, appeal, and effort (budget). This resulted in a lot of feedback loops, so I could not add some things I would have liked to in the fine-tuning.

DP: Once the statue existed digitally, with model, material, texture, and simulation, how did the project move toward turntable and rendering?

Sebastian Boyamba / Niklaus Hofer: The Client edited the final commercial together at their request. To speed up the feedback process, we exported a low-poly version of the melting scene in Cinema 4D as an Alembic file. We created a variety of camera angles using hardware rendering. Cinema 4D is still our tool of choice for quick iterations in camera animation. 

The Client chose about 8 or 10 shots per sculpture, each lasting about 2 to 3 seconds.

We then reimported the cameras as Alembic files in Solaris (Cinema does not handle animated USD very well). Still, Solaris can handle Alembic files too, and this was where everything came together.

Lighting was also handled in Solaris, which was very important to the look, so we adjusted it for each scene. It was 95% area lights and 5% dome light with noise for some variation. We used HDR only in the process, not in the final rendering.

DP: What did you render with, and what does Uber Eck’s current rendering workflow look like?

Sebastian Boyamba: We don’t have a house renderer. We are very pragmatic about it and choose the one that fits the project and, most importantly, the budget. Nowadays, we sometimes also render in Unreal. I hope for usable AI renderers. We also experimented with all the known AI models, and with Seedance you can see how close we are.

For this project, we tested Octane, Karma, and Redshift for rendering. We ended up using Redshift in Houdini Solaris because it handled subsurface scattering very well and was a bit faster than Karma, which also produced nice SSS. Octane was too unstable in this case due to the high polycount, and the SSS looked weird (though that could have been my test shader).

Conclusion

DP: How did Kayak and Perfect Accident react to the final work, and where are the spots shown?

Niklaus Hofer: The ads are currently running in Kayak ice cream shops across Greece. The Client is very satisfied, and so far we’ve received nothing but positive feedback, as has Perfect Accident. We believe that, precisely because there’s so much AI-generated content popping up everywhere right now, something handmade like this really stands out even more. And everyone loves ice cream!

DP: Now that the ice cream has been eaten, what is next for Uber Eck?

Niklaus Hofer: The motion design market has changed significantly in recent years, for a variety of reasons. We’ve been around in our current form as Uber Eck since 2012, so we’re slowly coming out of our “teenage years.” Lately, we’ve been focusing heavily on interactive installations, but of course we’re still pouring our hearts into classic motion design. It’s worth mentioning here that we’ve founded our digital artist collective, Extrabright.art.

This is where we started (in 2016, under the name UberEck&Motomoto), creating visual installations in physical spaces. We teamed up with our colleagues at Motomoto, and in recent years, we’ve been investing more and more time in these projects. 

Without getting too deep into the topic, AI certainly plays a role. In the sense that we feel like a lot of the cards are being reshuffled right now. What we’ve definitely realised, especially realised the Kayak Spot, is that we’re simply fascinated by the process and how things come together. 

That’s where our strength and passion lie. So while we’ll probably never become master prompters, if anyone’s looking for someone willing to get their hands dirty, we’re here.

Sebastian: Although we had an unusual pipeline coming from Clay, Kayak was also a bit of a goodbye for me to the classic 3D rendering pipeline. While working on the project, I felt that the effort of a full 3D pipeline would not pay off in the future. The expectations in speed shifted too much with AI. But with a bit of healthy grief, it’s not a sad goodbye. I’m also excited about what we can achieve in the future by staying pragmatic. We won’t let ourselves be forced into a fixed workflow. 

With brave clients like Perfect Accident and Kayak, the new possibilities can transform the production journey with a more evolutionary feeling.

DP: And now, just between us: heatwave or not, summer of the century or not, can you still eat ice cream after this project, or is the topic finished for this year?

Niklaus Hofer: It actually just kept getting worse. Since we’ve been staring at ice cream for months, I feel like even the sound of the graphics card whirring sets off a craving for ice cream.

Here in Munich, there’s an ice cream shop called “der verrückte Eismacher.” Our tip is to go there and order the vanilla. It’s pretty crazy.