Gaussian Splatting and the next step in 3D capture
Capturing real environments and turning them into 3D models is not new. Techniques like photogrammetry have been used for years to reconstruct locations from images. What is changing now is how detailed and usable these models are becoming in real production contexts.
Gaussian Splatting is one of the techniques currently being explored to do this. It allows scenes to be reconstructed from camera footage and used as navigable 3D environments, with a level of detail that becomes especially visible at closer distance.
Why this matters in practice
For Red Bull Media House, this builds on a longer evolution of working with 3D environments. Earlier event venue animations were fully created in 3D and took a lot of time to produce. This later shifted towards photogrammetry, where environments are captured using drones and converted into 3D models. This made it possible to visualise real locations, such as race tracks, at a larger scale.
Red Bull Cerro Abajo, 2022
At the same time, these models had clear limitations. The level of detail was not always sufficient, especially when zooming in, and postproduction required additional clean-up work.
Gaussian Splatting builds on this by allowing more detail to be captured, especially at a closer distance. In comparisons between earlier photogrammetry captures and more recent Gaussian Splat models, the difference becomes clearly visible.
How it works
Gaussian Splatting works differently from traditional 3D reconstruction methods. Photogrammetry creates models using meshes, built from points, edges and surfaces that define objects in space. These models are structured and can be edited using existing tools.
Gaussian Splatting does not build objects in the same way. Instead, it represents a scene as a collection of Gaussian splats in space, each with a specific position, size and colour that changes depending on the viewing angle. Together, these splats form an image that looks like a real scene.
Challenges and limitations
Like any emerging technology, Gaussian Splatting also comes with a number of limitations.
One of the main challenges is that these scanned environments are difficult to edit afterwards, since they don’t come with clear objects that can be easily selected or changed. This makes small adjustments in a production setting more complex. Take for example lighting adaptations: because the scenes are captured as they are in reality, it is difficult to change or improve the lighting afterwards.
On top of that, integrating these environments into existing 3D workflows is not seamless, which makes it harder to combine them with other assets or use them in tools like game engines or virtual production setups. Performance is another limitation, as the models can become resource-intensive and require a lot of computing power, especially in more complex real-time environments such as LED walls. For now, most applications remain in a testing phase.
To address these current limitations, new approaches are being explored, such as combining splats with mesh structures so they can be handled more like traditional 3D objects and used in existing tools like game engines.
How broadcasters are experimenting with it today
At VRT, Gaussian Splatting is currently being explored in different test trials within the VIVACE project, a VLAIO-supported research project involving partners such as EMG, Uncanny, Datameister, Digital Arts and Entertainment and IDLab (imec).
One experiment focusses on scanning studio environments and creating 3D representations that can be used in virtual productions, particularly for smaller formats such as vodcasts where sets are repeatedly built and taken down. By scanning a set once, it can be reused and placed on the LED wall.
Other experiments also look at larger productions, for example scanning VRT’s 89m high tower using a drone, with the intention to use it in media formats on the LED wall. Since the image of this national landmark extends beyond the edge of the LED wall, a virtual set extension is added so the full image can be shown.
Extending formats beyond the broadcast
In the VRT programme Ons Huis, Nieuw Huis, Gaussian Splatting was used in a different way. Together with 3-DEE, virtual tours of renovated homes were created as an extension of the programme, allowing viewers to explore the house beyond the TV episode and engage with the content in a more interactive way.
For this project, the scanned houses are turned into interactive environments where viewers can move through the space and access additional content such as product information, before and after images of the renovation and other contextual elements. This also creates a direct link between content and commercial elements, where products can be connected to clickable and engaging information inside the environment instead of only appearing during the broadcast.
This shifts the format beyond pure viewing, as it combines storytelling with more integrated forms of partner visibility and activation. Instead of relying only on exposure within the programme, the environment itself becomes a space where additional layers of content and value can be added. Partners can take on a more active role within the experience, positioning themselves as experts rather than only being visible as a brand. They can add tutorials, product explanations or practical guidance within the context of the home, making their presence more relevant and useful for the viewer.
At the same time, this also creates new monetisation opportunities. The environment can remain accessible beyond the original broadcast moment and be reused across different channels, allowing partners to extend their presence and activate the content further in their own communication. This can include elements such as exclusive offers or discount codes, as well as simple activations that encourage further interaction. In this way, the environment becomes not only a storytelling layer, but also a reusable commercial asset that can generate value over a longer period of time.
It also becomes possible to measure how viewers interact with the experience, for example how long they explore the tour or which elements and partners they engage with, providing more insight into performance and audience behaviour over time.
What media companies can take from this
These cases around Gaussian Splatting show how quickly this technology is moving from experimentation into real media workflows. From production to distribution, it opens new ways of working but also raises new questions.
A few takeaways:
- Start with reusing physical sets
Treat scanned sets as assets that can be reused across productions instead of one-off builds. This is especially useful for recurring formats where speed and flexibility matter more than rebuilding physical sets each time. - Accept that workflows are still fragmented
Scanning, virtual production and LED workflows are developing in parallel, but they are not yet a fully integrated pipeline and still require manual adjustments in practice. - Look at new monetisation layers inside content
3D environments can extend beyond the broadcast and function as reusable assets across multiple channels and timeframes. Instead of traditional product placement, partners can take on a more active role within these environments. - Match the right 3D method to the use case
Different projects require different levels of detail and flexibility. The choice between photogrammetry, Gaussian Splatting or hybrid approaches depends on how the content will be used in production.
Looking ahead
Gaussian Splatting is not yet a standard tool in media production, but it’s already being tested in different parts of the industry. It’s moving from experimentation towards more practical use, especially in how environments are captured and reused across production workflows.
At the same time, it is starting to connect production and audience experiences in a more direct way, where scanned environments are not only used during production, but also become part of the final content as interactive spaces that audiences can explore and engage with beyond the broadcast, creating additional monetisation opportunities.
For media organisations, the main question becomes where it actually fits in their workflow and what it can realistically be used for, whether that is in production, virtual sets or audience-facing applications.
