Debbie  Well-spotted 😊
Sergio  Finally, someone spotted that... 🙂

Finding out about all the support that poured into this shortfilm, reminds us  what good relationships can help achieve. Going back to the process of HOME and how it sparked the genesis of Hydra Studios - what happened after the film was released?

DEBBIE
HOME went live in April 2020 and we registered Hydra the following month. We started our first live project in July 2020, so it was very quick. Hydra was founded during Covid lockdown, so we couldn’t have a physical studio space during that time. We always intended to have a virtual set up anyway – we produced HOME with no physical space and a virtual pipeline. I think that made it less daunting and we didn’t require any investment to create the studio. Initially the studio was us.

Thinking back, it was quite crazy. We didn’t even meet when we formed Hydra and I eventually saw Sergio twice in 2 years - he lives less than 4 miles from me!

How did you arrive at your current pipeline? Why the decision to use Houdini?

DEBBIE
Houdini was always a must. No Houdini = no Sergio – they’re one and the same. Over the 15 years of using Houdini, Sergio has developed procedural tools that we use as part of our regular pipeline and they have become a backbone of Hydra’s workflow. We use Maya for animation and assets can be created using most of the standard software – characters in ZBrush, cloth in Marvelous Designer on occasion, etc. Anything after animation goes into our Houdini pipe, rendering in Mantra with compositing in Fusion. We’re currently updating our render and lighting pipeline software to Solaris.

Where have the crew been prior to Hydra?

DEBBIE
Most of our senior crew held senior positions at Axis Studios prior to Hydra, so we have a lot of experience in creating hi-end cinematics and other short form material. We’ve all been doing this for a while 😊 Of course, we also bring in new blood to Hydra too – talent and a good attitude are key. Experience is a bonus of course, but experience without the other two doesn’t do it.

So. Beckoning. Hydra rock tool and terrain generation. It looks like your tool handles erosion while also inserting and creating rockface detail at the same time! Did we get that correctly?

SAVA
It’s pure magic to me, simple as that:) I won’t linger here too much as this is Sergio’s domain, but I’ve always been inspired by what technology and different tools can bring to the creative process and how much they can streamline the production. In the early days of exploring Unreal Engine’s procedural terrain systems, I came to realise that while they’re most certainly powerful, they’re lacking the nuance and level of detail when it comes to large overhangs and cliff faces. At the time, I was aware of the tool Sergio was developing at Hydra and thought that Beckoning would be a good testing ground for it. I love the iteration process and how flexible the tool is with targeting a specific type of rock formation, and the fact that the texturing can be driven procedurally in Unreal Engine is just a cherry on the top!

Rock Synthesizer

SERGIO
Yes, the tool grows rock while eroding it at the same time. I can think of at least two good reasons for doing that. The main one is performance, as in you don’t want to generate a lot of rock surface area only to try and erase a lot of it as a separate process afterwards. It gives the user better feedback because the first few frames of the simulation are pretty much real time, it will have built up the overall rock volumes including erosion within ~15 real-time frames, so you know more or less what you will get if you wait longer to resolve the details. Thirdly, it might even be analogous to natural processes over geological time scales, at least on average. For example, as a seam of rock gets pushed through the ground you expect erosion to have taken place throughout the process. There are obvious outliers of course, like a load of freshly laid basalt.

The motivation

The tool came about because I was interested in creating rock formations that have a sense of history over geologic time scales (staring at interesting cliff faces for way too long does that). The Earth's crust is constantly churning rock, exposing it to the elements as it surfaces, spewing from volcanoes, folding it, crumbling, sedimenting, exposing and eroding again, and erupting or flowing again over any combination thereof.

But one can only go so far replicating those processes. Even if I could, I want the simulation to take less than geologic timescales to complete, ideally a few minutes at most. So a lot of shortcuts have to be taken.

I was also interested in working out a process for realistic erosion of 3D shapes. After all, this was 2020 at the time, why are CG terrain tools still based on simple 2D heightfields? Erosion is a 2D-only thing as far as DCC software goes (although anything is possible in Houdini, of course). But nothing has really changed in the field for a very long time. The competing tools are, let's face it, much of a muchness, just different UIs with more or less the same capabilities and ingredients.

For the older folk like me that played around with arcane tools like VistaPro/Terragen/Bryce 20 years ago, has terrain authoring really changed that much beyond adding 2D erosion?

It is also partly motivated by the observation that even if you start with accurately scanned rocks, if you take them too far out of their natural context and say - stack them into a large cliff face - it will unsurprisingly tend to look like rubble from a distance, because it lacks the cohesion of a naturally large cliff face. There is no shortage of UE5 videos to illustrate this. A tremendously larger library is needed to regain some sense of natural placement, and originality, because we are definitely seeing that same rock that everyone likes everywhere.

The core idea brings together a few key ingredients; growing while re-meshing, Voronoise, proper strata, erosion, and deposition. I will try to give an overview on how it works.

Growing Voronoise

We take any starting shape mesh, and on every iteration, we re-mesh the geometry, then displace it along some vector. This tool actually started life as VDBs rather than meshing, but it became apparent that this would be much harder to scale. For example, the ability to work on open meshes and limit the computations strictly to the surface, is a significant performance advantage when things get big.

We animate the re-meshing edge length to start coarse and finish at the desired resolution, which is linked to the step length of the displacement. This means the mesh grows volume quickly over the first few frames, hence what I said earlier about having more interactivity than you might expect.

The displaced amount is set by a feature-rich procedural noise based around Voronoi. When you run Voronoi iteratively, you can use the previous iteration's cell id as a random seed to drive say, a transformation of subsequent iterations or any other parameter you can imagine a use for, like gap width between cells, or how about chamfering the top of rocks according to the slope. A wide variety of looks can be created with it.

The mesh grows towards the local minimum in the noise value or, it stops if a collision is detected. The effect of collision avoidance is what forms deep cracks - a feature you don't often see in scans. 

Additionally, the displaced direction is controlled by various "Grow Weights", analogous to turbulence/wind forces we use to control fluid simulations. Those are: direction along a vector, curl noise, tangent, or normal. It's possible to grow branching coral-like structures from it without too much effort.

Strata

This is one of the simplest and most important aspects. Most procedural attempts at this involve a simple ramp or steps based on Y and domain warping it. But the problem with this is that the number of strata is constant everywhere. This is too simplistic. In the real world, when you look at a cliff face, particularly active or interesting areas - you often see strata tapering in and out of existence. You could have say two main strata, and a pocket in-between with a bunch of strata in it. Point being the topology of strata is often much more

interesting and worthy of putting a bit more effort into. That is what we aim to achieve with our strata shader. It's simply a shader that accumulates the noise-modulated height of each layer (thus displaced by the height of the prior layer), in a for-loop, and then domain warped. We get stack numbers for each layer which we can use to randomize things like Voronoise size, offset, roughness, erodibility, as well as a thickness value that we use to ramp colours.

Erosion

I found early on that masking the growth according to the slope of the surface can go a long way to mimicking the effect of erosion on the top of the surface. In fact, before we do any erosion, we will have already done everything possible with Voronoise (e.g., rounding the tops of accessible Voronoi cells to taper the gap between cells). But we don't want every rock to do this, because we want some proportion of them to stay above ground to be the outcrops, freshly exposed, or recently transported rocks. This is where the Voronoi cell ids can interject a seed to randomize parameters. The main benefit is that it avoids adding a lot of material only to remove

it right after, therefore making it easier for the erosion processes to do their thing without needing substeps. Now that we have avoided adding material where we 'guessed' it might have eroded away already, we follow with material removal and transportation steps. We have various methods that create different effects, including slump, plateau, pool forming, thermal, and water. Water is the most recent addition and in that case, to transport water over long distance and dig persistent water channels; one does need substeps but, I have found ways to achieve the effect efficiently.

Deposition

This is the process of adding sedimentary material like soil/snow/sand. It’s obviously weaker than rock, much more affected by the presence of wind and water, you can see some wind drifts in one of the examples. It grows everywhere within the slope threshold, unless a Voronoi random seed tells it this is a 'recent' rock, it tends to fill in concave areas, and it will grow beyond its slope threshold

if a concave area allows it to cling more easily. Water is good at removing it, leaving behind trails of rubble, because the Voronoi shader is specifically designed to accept a rubble modifier which transitions the rocks into smaller and smaller pieces, while avoiding to place them in impossible places like the side of a cliff.

An additional feature is Drip. Where deposition has occurred and water flows, stalactites can form.

Attributes painted on the source meshes can be used to affect the strength of the main effects say, erosion, rock growth rate, deposition rate, etc. You can also choose to stop growing the surface and then only use the deposition feature on top. Everything happening everywhere all at once is optional.

Will you release this tool for public use?

SERGIO
I haven't really thought about releasing it. It's still too much fun to mess with. Every time I look at it, I end up refactoring a good chunk of it. But I could find the time to make it releasable if there is enough interest.

Really incredible set of tools and controls for making rich landscape! Glad we can study all the details here, which we were not able to observe in Beckoning, with such clarity ;)

SAVA
Blame it on my love for fog:) That damned thing always obscures 90% of the assets, but when it’s essential to the story, there’s no way around it. Suffice to say, we’ll see a lot more landscape details and cliff faces in the feature length version of Beckoning:)

Can you tell us how you created the terrain?

SERGIO
The terrain started with the meshes that Sava used in UE5 to previz. I imported them into Houdini and projected a new low-detail seamless mesh encompassing the disjoint pieces, and then created a seamless connection to a much larger distance terrain. This was converted to a heightfield, for which I used standard terrain tools to shape it further. The attributes that the standard 2D erosion tools export are used to seed an initial state for the debris, water, in the following 3D solver.

At the end of that process, we get a low polygon mesh which serves as the canvas for what the rock tool (now effectively a terrain synthesizer tool), can then do its thing. 

Wow, the results speak for themselves.
For overall direction of the studio, where would Hydra like to be in the future?

DEBBIE
We’re currently working on a co-production passion project with a creative partner we’ve been excited to collaborate with for some time. Hopefully we’ll have a teaser to share before Summer. Sorry to be cryptic, but it should be worth the wait!

Our commercial work for 2023 includes both game trailers and hopefully some short form narrative films in the entertainment space. We’re still to be completely sure what the full year holds!

If content is creatively inspiring, we’re on the hook. If it encourages us into new areas of exploration, we’re ready for the challenge. If it has good people behind it, sign us up!

Debbie, Sergio, Sava - thank you for the chat.

SERGIO
Here are some Network Editor bonus shots ;)

hydrastudios.com