Launch Video Library · AI · Launch · 2026

Introducing Praxis-1

Runway introduced Praxis-1, an open-weight World Action Model designed to translate video pre-training into real-world robotic control.

What Runway shipped

Runway announced Praxis-1, an open-weight World Action Model that translates their extensive video pre-training infrastructure into physical control for robotics. By leveraging infinite video data over limited physical demonstration data, the model acts as a universal policy for developers and researchers.

Moving from generative video to embodied intelligence marks a significant expansion of the company's focus. The release signals a shift toward applying large-scale video models directly to hardware, creating a bridge between digital simulation and real-world environments.

How the motion works

For a launch bridging digital models and physical robotics, motion design can emphasize the translation between the two domains. A designer might use a match-cut to transition seamlessly from a digital video training simulation to a physical robot executing the exact same movement. This visual parallelism proves the core claim without relying on voiceover, grounding the abstract concept of a World Action Model in tangible hardware.

Pacing is critical when explaining complex AI infrastructure. Rather than overwhelming the viewer with continuous camera movement, a hard-cut between distinct physical environments can demonstrate the model's versatility across different embodiments. Holding on these static shots allows the viewer to process the complexity of the robot's actions before cutting to the next context, establishing scale through rhythm.

Typography should support the technical weight of an open-weight model. Using kinetic-typography to introduce key terms can anchor the narrative while the footage does the heavy lifting. A sudden punch-in on specific robotic joints or sensors as the text lands reinforces the precision of the real-time infrastructure, syncing the typographic rhythm with the mechanical precision of the hardware.

What to steal from it

  • Use match cuts to visually connect digital training data with physical real-world execution.
  • Anchor complex AI terminology with stark, kinetic typography to give abstract concepts physical weight.
  • Rely on hard cuts between distinct environments to quickly establish a product's versatility without lengthy explanation.
  • Let the camera linger on complex mechanical actions so the viewer can fully process the technical achievement.

Want a video like this for your product?

Impractical cuts one from a single prompt — the same motion craft, in about twenty minutes.