Runway Act-Two

One-line summary: Runway's 2025 closed/commercial performance-capture model — the successor to Act-One — that transfers a webcam-captured full-body, hand, and face performance onto a character reference image or video.

What it is

A driving-performance-video + character-reference to animated video model. Designed for filmmakers, game developers, and creators who don't have access to studio mocap.

The lineage:

• Act-One (2024, on Gen-3 Alpha) — facial-performance transfer only: eye-lines, micro-expressions, pacing, delivery from a driving video to a character reference image. Optional voice integration.
• Act-Two (2025) — extends Act-One with full hand, head, and body tracking from a single webcam capture. Accepts character image or character video as the reference.

Why it matters to ai-video-generation

Currently the closest closed-source equivalent of "puppet your avatar from your webcam" — the most natural workflow for performance-driven avatar creation. It's the named human-eval competitor that wan-animate benchmarks itself against. From 2026-05-07-ai-avatar-motion-mimicking-models-survey.

Key facts

• Vendor: Runway.
• Underlying model family for Act-One: Gen-3 Alpha (Runway provides no further architectural detail publicly).
• Inputs: driving performance video (webcam-grade is fine) + character reference (image or video for Act-Two).
• Optional: voice integration.
• 2026-08-18 in-app price (Runway Academy): Act-Two is 5 credits per second of generated video — the same meter prices Gen-4.5 at 12/sec and Kling 3.0 Motion Control at 13–17/sec. From 2026-08-18-autoresearch-best-ai-video-generation-tools.

Capabilities

• Act-One: facial performance transfer, including eye-lines, micro-expressions, pacing, delivery. Works across characters with proportions different from the source actor.
• Act-Two: adds full-body tracking, hand tracking, head tracking on top of facial performance.

Strengths

• Simple single-camera capture workflow — no studio rig required.
• High-fidelity face transfer including dialogue-heavy scenes.
• Marketed safety: detection of public-figure generation attempts and voice verification.

Weaknesses

• Closed / proprietary.
• Underlying architecture not publicly documented.
• Self-reported quality; competing claim from wan-animate paper that Wan-Animate beats it on human eval (author-reported).

Sources

• 2026-05-07-ai-avatar-motion-mimicking-models-survey
• 2026-08-18-autoresearch-best-ai-video-generation-tools

Related

• ai-video-quality-versus-cost
• what-are-the-best-tools-the-best-performing-most-cost
• wan-animate — open-source competitor.
• heygen-avatar-v — different commercial niche (talking-head, audio-driven rather than performance-video-driven).
• dreamactor-m1
• independent-avatar-benchmarks