The Most POWERFUL AI Storytelling Tool of 2024 is Here.

Runway ML’s Act One is positioning itself as a fast, actor-driven way to generate expressive character performances from real facial acting—without the usual motion-capture and rigging pipeline. The core workflow takes up to 30 seconds of an actor’s video (recorded on a phone is enough), detects facial features, and then transplants those expressions onto a chosen character image—ranging from 3D animated models to photos and custom creations—producing outputs that can look surprisingly cinematic and, in some cases, close to non-AI animation.

The most important practical claim is that emotion and nuance can survive the transfer. Traditional facial animation often requires multi-step setups: motion capture hardware, multiple reference angles, and manual face rigging to preserve the subtleties of a performance. Act One aims to bypass that complexity by using an AI-driven approach where a single performance can drive a character’s acting. In demos and tests, characters can turn their heads and maintain convincing facial detail, while the background tends to be more limited—camera movement is generally constrained, and scenes often rely on steadier or stylized environments (burning buildings, swaying trees) rather than dynamic tracking shots.

Quality is a recurring theme. Many generated results are described as watchable and not “uncanny,” with some outputs resembling high-end renders or even footage-like realism. Still, the system shows clear failure modes. Blinking can be inconsistent, especially with cartoon or stylized characters, where the model may blink only parts of the eye rather than the full eyeball. Face detection is another friction point: some cartoony or stylized inputs trigger “unable to detect a face,” and the model may require a more human-like nose/face geometry to lock on. Longer clips also increase generation time, though the wait is framed as manageable—typically a few minutes.

The transcript also highlights how creators can extend the pipeline beyond Act One. Users can generate characters via idiogram, run performances through Act One, then use 11 Labs to translate the actor’s speech into a different voice—turning acting into dialogue with a new vocal identity. There’s also experimentation with post workflows, including combining Gen 3 video outputs with Act One to push toward more fluid, fuller-body animation.

Access and cost matter for adoption. Act One is described as available for everyone to try, but it runs on a credits system rather than being fully free. Copyright and ownership questions come up in community discussion, with claims that generated assets are owned by the user, though the transcript also notes practical limitations like head-and-shoulder focus and limited consistent character control across multiple angles.

Community reactions emphasize the workflow shift: animation that once took forever can happen in minutes, enabling lunch-break experimentation and new creative directions. The remaining bottlenecks—full-body tracking, consistent multi-angle character identity, and more reliable face detection—are framed as the next steps needed for Act One to become a truly dependable production tool for longer, story-driven projects.