In a modest workspace near Sony’s historic Culver City lot, an actress runs in front of white walls. There is no field, no cave and not yet a world for her character to inhabit. Yet on the director’s monitor the performer already appears in windswept grass; on the next take, the same space becomes an underground cavern. This is the set of Touch Grass, a horror film produced by Promise through a workflow in which capture and generation begin to converge.
The scene described by the Guardian is not simply a technical demonstration. Touch Grass has a budget in the low millions, a real performer and an organized production. Promise is backed by investors including Andreessen Horowitz and has a partnership with Google, whose tools are integrated into the studio’s MUSE production system. Generative video is leaving the individual test phase and attempting to become a set methodology.
The most interesting element is not the disappearance of scenery. It is the movement of the moment when scenery takes form. In traditional cinema, the environment precedes performance: it is built, selected or prepared before the actor enters. In LED-wall virtual production, the digital space is displayed on set and photographed by the camera. Here the environment is composed from the captured body and can change almost immediately.
During Touch Grass tests, the actress’s movement was integrated in real time with backgrounds produced through Seedance 2.5. The director could see the relationship between figure and landscape, check whether a gesture crossed the space correctly and replace an open field with a cave. It is not yet the completely live generation of every final detail, but it dramatically shortens the distance between capture, previsualization, compositing and creative decision.
For visual-effects work, that is a major shift. A shot normally passes through capture, tracking, rotoscoping or keying, environment construction, integration, lighting, compositing and review. The new system tries to return a credible first version while the performance is still happening. Problems of scale, perspective, eyeline and interaction can be discovered before the set is gone and the performer is no longer available.
Immediacy does not eliminate post-production. An image sufficient to guide a choice on a monitor may not hold up on a large screen. Continuity, hands, contacts, hair, shadows, reflections, depth and consistency between shots still require control. The risk is confusing the speed of previsualization with master quality. Seeing the world immediately does not mean it is finished.
The most delicate transformation concerns performance. Actors construct their work through space: they measure distance, feel surfaces, respond to light and establish relationships with objects and bodies. If the environment exists only on a monitor, part of that stimulus must be imagined or translated by the director. The technology can reveal the result instantly without giving the performer the same sensory experience.
A minimal stage should therefore not become an impoverished stage. Even with synthetic environments, productions need physical references, floor marks, volumes, interactive light, wind, tactile materials and a precise account of the scene’s conditions. The more the environment is generated later, the more direction must construct it earlier in the performer’s mind. Reducing structures does not reduce direction; it makes direction more abstract and responsible.
The production designer’s role changes as well. It does not disappear; it moves from building one environment to designing a system of possibilities. Architecture, materials, period, wear, palette, proportions and rules must be defined so the model can generate coherent variations. A field is not simply “grass”: it has species, climate, wind direction, terrain history and a relationship to the character. Without those decisions, the world may look rich while remaining generic.
The same applies to cinematography. If the background is produced after capture, light and optics must anticipate an environment that is not yet stable. A source on a face must correspond to the sky or cave that will appear; depth of field, camera movement and motion blur must belong to the synthetic space. The cinematographer does not illuminate only what exists, but defines the conditions that the not-yet-existing image must obey.
Promise describes a more agile production model built around small teams and project-specific tools. This can create real opportunities for independent films that could not access large sets or locations. But lower cost does not automatically mean greater freedom. If a few models and platforms become dominant infrastructure, aesthetics, contract terms and tool availability may depend on companies outside the production.
New continuity documents are therefore necessary. It is no longer enough to record camera, lens and take. Productions must preserve model version, references, instructions, control images, settings, licenses and approved outputs. If an environment is regenerated weeks later, the team must be able to reconstruct the conditions that produced it. Synthetic scenery is variable; that is precisely why it needs rigorous technical memory.
Touch Grass shows a possible future of the set not as an empty place, but as the meeting point between physical presence and a calculated world. The actor’s body remains the unrepeatable event around which everything is built. AI can change the field, cave, sky and scale of the location; it cannot automatically recover the intention of a gesture that was poorly directed or captured.
Scenery that arrives after the actor does not make film immaterial. It makes what must exist even clearer: performance, time, bodily weight, gaze and directorial decision. The world can be replaced between takes. The human presence before the camera remains the point from which that world acquires scale.