Pika previews Enact for directing AI video in real time
Pika’s Enact preview shows how creators could move actors, give notes and control the camera as scenes play, making AI video feel more like directing.

Pika previewed Enact on Nov. 18, an experimental AI video tool designed to work like a live film set. The company says users can drag actors around, give them spoken notes and move the camera while a scene keeps playing.
The preview featured in The Rundown is still a research project, with no consumer access yet.
How Enact works
According to Pika, users can grab a person or object and move it while a scene plays. The AI then adjusts the subject’s pose, lighting and shadow to fit the new position. Camera controls stay available during the live action, so users can change how the scene is framed.
Each actor is a separate AI agent with its own personality. Pika says actors respond to voice and text direction and retain feedback within a session, including notes from earlier takes. Users rehearse a scene, then call “Action” to start a fresh take.
Pika’s camera benchmark ran at 640×480 and 24 frames per second on eight B300 graphics processors. The company also edited its actor demo for length. How well all the controls work together through a full session remains unclear.
More control, with limits
Pika’s Oct. 29 H3 Flex announcement describes more ways for creators to shape a result. The model supports up to 12 image, video or audio references and five audio tracks that can be edited separately, according to the company. Its advertised time of seven seconds to generate 15 seconds of video measures only the model’s core computation.
The company’s technical report lists starting rates of $0.01 per output second for video made from text or an image, and $0.03 per second for video guided by references. Enact’s own pricing remains unknown.
Quality also varies by task. Pika’s own preference study favored competing models on physics, stories spanning several scenes and some action tasks.
Why it matters
AI video’s reputation for “slop” can obscure the deliberate work creators put into it. WIRED reported in October 2025 that filmmaker Josh Wallace Kerrigan wrote his scripts and performed his characters for Neural Viz. It also described repeated attempts to fix outputs that failed to follow his instructions. For creators doing that work, quality gains matter alongside control, speed and the cost of trying again.
Enact’s appeal is the chance to make those corrections while the scene develops. Moving an actor, changing the camera and giving a performance note could let creators judge staging and emphasis as they work. For filmmakers and small teams, fewer interruptions to request another clip could bring the process closer to directing and rehearsing on a set.
H3 Flex’s low starting rates could give small teams room to try more versions of an idea. Actual savings would depend on how many attempts it takes to get a usable shot and how much editing and finishing it needs.
An early role for Enact could be planning scenes before filming them. Interactive staging could let a director explore where actors stand and how the camera moves before committing to finished footage.
Google’s Nov. 18 Genie 4 announcement points to a similar move toward interactive creation, with changes to environments and action during a session. Google also identifies camera delay, physical interactions, audio precision and lasting scene consistency as research challenges.
Enact’s usefulness will depend on how reliably actors retain direction across takes and whether camera control stays responsive through a full session.
Sources & further reading
This story builds on reporting from The Rundown newsletter on November 19, 2026.