Over the past year, a particular production case has been circulating in AI and virtual production circles. It deserves more attention from producers, showrunners, and production managers in Germany and Austria than it has received so far.
Promise, a GenAI production studio based in Los Angeles, built something specific for Harlan Coben's Final Twist, a CBS true crime series. The production required the host to stand inside crime scenes that no longer exist physically. Not in front of a green screen with composited images behind him. Inside the rooms, walking through them, interacting with the space as though it were still there. Ten separate crime scenes, all reconstructed from archival photographs.
What they built tells us something important not just about this one show, but about an entire category of storytelling formats that are now available to producers for the first time.
What Promise actually built
The technical details carry the creative implication, so they are worth walking through precisely.
Stage 1: AI upscaling
They started with degraded archival photographs of crime scenes — old images, often low resolution, sometimes damaged. The kind of material that would normally sit as reference photos on a production office wall: useful for context, far too poor in quality to show on screen.
AI upscaling is a process where a machine learning model analyzes a low-resolution image and generates a higher-resolution version, filling in detail the original photograph does not contain. The model has been trained on millions of images and makes intelligent predictions about what fine details should look like: textures of walls, grain of wood, reflections in glass. The result is a plausible, visually convincing version at broadcast resolution.
Stage 2: 3D reconstruction via Gaussian splatting
Promise used tools from World Labs, founded by Fei-Fei Li, a leading researcher in computer vision at Stanford. The specific technique is called Gaussian splatting, a method for converting flat photographs into three-dimensional environments a camera can move through.
Instead of building a 3D model polygon by polygon as a traditional VFX artist would, the system creates a cloud of overlapping volumetric elements that together produce a convincing sense of depth and space. What comes out is an environment you can navigate, not a simple panorama.
We work with production companies on AI-augmented pipelines, from development through post.
Book a conversation →Stage 3: LED volume display
An LED volume is a large curved wall of high-resolution LED panels that surrounds the performers on set. Instead of removing a green screen background in post-production and replacing it digitally, you display the environment in real time on the wall behind and around the performers.
The performers can see the space they are supposed to be in. The lighting from the wall falls naturally on their faces and bodies. The camera captures both performer and background together, so reflections, color temperature, and perspective all match without digital correction. Promise used a 16K resolution wall.
The host could walk through reconstructed crime scenes that felt spatially real: not because the rooms had been physically rebuilt, but because AI had generated them from fragments and they were displayed around him in real time.
Promise — Using Hybrid Production to Reconstruct Crime Scenes
How this would have been done before
To understand why this matters, it helps to think through how a production team would have approached the same creative challenge five years ago.
Physical set reconstruction
You would hire a production designer and a construction team to build the rooms from scratch based on reference photographs. For ten distinct crime scenes, this means ten separate sets or a series of redresses. Each requires research, design, construction, set dressing, and period-appropriate props. The budget for this on a true crime series, where the rooms appear briefly as atmospheric context, is difficult to justify. Most productions in this genre simply would not do it.
Location shooting
You would find real rooms that approximate the crime scenes and modify them to match the archival references. This requires location scouting, travel, permits, and compromises, because you are unlikely to find a room that matches closely enough to feel authentic. For crime scenes from decades ago, the original locations may have been demolished or renovated. This is why so many true crime documentaries fall back on talking-head interviews and B-roll rather than spatial reconstruction.
Traditional VFX
A skilled VFX team could build photorealistic rooms, but the cost per environment is significant and the timeline long. Each room might take weeks from multiple artists. For ten environments, this represents a substantial VFX budget competing directly with other production needs.
The key point: none of these options would have allowed the host to physically stand inside the reconstructed space and interact with it in real time. A performer who can see the space they are in gives a different performance than one who is imagining it. AI did not simply reduce cost here. It made a specific creative choice viable that was essentially off the table before.
What this means for German-speaking productions
The Promise case demonstrates something broader than crime scene reconstruction. The underlying capability is this: you can take minimal visual source material, sometimes a single photograph, and generate a navigable three-dimensional environment from it. World Labs released a tool called Marble in late 2025 that can produce an editable 3D environment from one image in under five minutes.
LED volumes are no longer exclusive to major American studios. They operate in London, Munich, Budapest, and dozens of other cities. The virtual production market sits between 2.8 and 3.8 billion dollars in 2025, with 14 to 20 percent annual growth projected through 2032.
European production companies are already working with this pipeline. Paprika Studios in Budapest used AI-generated illustration sequences for The Ruthless One, an RTL+ true crime series reconstructing a 1990s Hungarian murder case. VG Norway built Norske forbrytere with AI-animated reenactments, earning a Gullruten (Golden Screen) nomination in 2024.
The competitive advantage is shifting. Two years ago, the question was whether you could access these tools at all. Today, the question is whether you know what to do with them. Better AI generates better environments faster. It does not tell you which environment to build, what the camera should reveal, or how a performer should move through the space. The technology is becoming a commodity. The creative application is not.
What distinguishes the productions that work is a clear understanding of which creative choices the tools make newly viable, and a methodology for executing those choices in service of the story. This expertise has a name. It is directing. And it is the part of hybrid production that does not become easier as the tools improve.
Promise — Reconstructing Crime Scenes: AI VFX Breakdown
Three questions worth asking
If you are a commissioner, producer, or production manager evaluating AI-augmented production for the first time, these are the questions that cut through the noise.
Ask the format question first
Not which AI tools does this production partner use, but what formats or creative choices does this approach make available that were not viable before. Every genre has equivalents: creative choices that producers have historically ruled out on grounds of cost or feasibility that AI is now putting back on the table. Start there, not with the tool list.
Ask about methodology
Anyone with the right tools and sufficient budget can generate a 3D environment from a photograph. The question is whether the team has a systematic approach to deciding which environments to build, what level of detail they require, how they serve the story, and how performers interact with them. This is what separates a technology demonstration from a series that holds up over a full run.
Ask what they chose not to build
When you can generate any environment from minimal source material, the risk is building too many, or building environments because you can rather than because the story needs them. The best hybrid production work comes from teams that impose constraints deliberately. Ask them what they ruled out, and why.
"The technology is becoming a commodity. The creative application is not."
The format question
The industry conversation about AI in production has been dominated by the technology question: can AI do this, how much does it cost, is the result convincing enough. These questions mattered when the tools were immature. They matter less today, because the answer to most of them is yes.
Promise's crime scene reconstructions are a specific, small answer to the more important question. A host walking through rooms that no longer exist, experiencing them spatially rather than looking at photographs of them. A format choice that was not available at this budget level before, and that changes the way the audience experiences the material.
The producers who will define the next generation of formats are the ones who move past this question entirely and ask what was never possible before. Not cheaper rooms. Rooms that could not have existed. Not the pipeline. The formats it enables.
