Expanding Creative Control, Runway Adds Custom Voices So Characters Can 'Sound Exactly Like You Want Them To'

In the fast-moving world of generative AI, what many call the LLM war has expanded far beyond text, now encompassing video, audio, and interactive experiences.

Companies race to deliver higher fidelity, better consistency, longer sequences, and more intuitive controls, each trying to carve out an edge in simulating realistic worlds from prompts alone. Big players like Kling, Google’s Veo, and others push boundaries in photorealism, native audio integration, or extended clip lengths, while some earlier entrants have scaled back or pivoted.

The result is a crowded field where raw capability often clashes with practical usability for creators who need tools that actually fit into storytelling workflows.

Runway has approached this space differently, prioritizing precision and creative control over sheer spectacle.

Rather than chasing the longest possible video or the flashiest benchmarks, the company has built a suite of interconnected features aimed at filmmakers, animators, and narrative builders. Its models excel at maintaining character consistency across shots, offering fine-grained camera directives, and allowing iterative editing that feels closer to traditional post-production than pure generation.

This focus has kept Runway relevant even as the broader market fragments, with users frequently citing its reliability for professional-grade output that can be stitched into longer projects without losing coherence.

The latest addition is custom voices for Runway Characters, which fits squarely into that philosophy.

Characters themselves are a relatively recent capability, letting users create persistent, conversational video agents with defined appearances, personalities, and behaviors that can respond in real time or be directed through scripts. Until now, giving those characters a voice often meant relying on preset options, recording audio separately, or syncing external clips.

The new feature changes that by letting anyone design a voice directly through text prompts, much like prompting for visuals or motion.

Users can just describe what they want, like "a gruff, battle-worn male voice with a deep, raspy texture and a thick Highland brogue" or "high-pitched, low-confidence, clumsy viking voice," and let the system generates it, complete with natural intonation, accent, and emotional range.

The integration is seamless.

Once created, the voice can be assigned to a Character, and lip synchronization happens automatically when the agent speaks, whether in a generated video clip or a real-time interaction.

A single Viking warrior, for example, can shift from friendly and warm to aggressive and smoky or nervously comedic simply by swapping the voice prompt, all while the facial animation and body language remain consistent with the underlying model. This prompt-driven approach mirrors the rest of Runway’s workflow.

The idea of having this feature is to reduce friction by keeping everything inside one environment, eliminates the need for separate voice actors or heavy post-production, and gives creators the same level of textual precision they already expect for visuals.

What stands out is how this update reinforces Runway's emphasis on usability within the larger competitive fray.

While some competitors focus on embedding audio natively into video generation or scaling to 4K with spatial sound, Runway has leaned into modularity and creator-centric tools, with Characters that feel like digital actors users actually can direct, edit, and evolve.

It's worth noting that the Runway Characters' custom-voice capability is far from being the most advanced audio model on the market. But at this time at least, it solves a specific pain point for people building stories, allowing them to prototype dialogue, test character dynamics, and iterate quickly without leaving the platform.

In a landscape where the next breakthrough is always one release away, Runway's steady accumulation of practical features suggests a different kind of strategy: one that values workflow depth over headline-grabbing specs alone.

As the war becomes fiercer, this kind of focused refinement may prove more enduring for those actually producing work than the raw power of any single model.