Runway Integrates ElevenLabs v4 Narration and Dialogue
Runway announced that ElevenLabs v4 is now available on its platform, claiming that the model's expressiveness and natural tone in narration and character dialogue surpass previous versions, and it is available alongside its image and video models for creators.
On the same day, ElevenLabs released Eleven v4 and the low-latency version v4 Turbo. The company stated that the new architecture improves control and cloning speed, allowing approximately 10 seconds of audio to be cloned; in blind tests, about 75% of listeners preferred its expressiveness and naturalness compared to Cartesia Sonic 3.6, Inworld TTS-2, and Google Gemini 3.8 Flash-Lite TTS and Flash TTS. The Artificial Analysis voice ranking showed v4 Elo at approximately 1319, higher than Sonic 3.6 at about 1276 and Gemini 3.8 Flash TTS at about 1267. The number of languages expanded from about 70 in v3 to over 90, with the greatest improvements in Japanese, Brazilian Portuguese, Mandarin, and Cantonese. Professional voice cloning is now integrated into v4. Enterprise voice calls now account for over half of revenue.
Creators can use bracket tags to layer emotions and sound effects, such as laughter, anger, French accents, light rain, and phone vibrations, with the model executing them in order; the stability of IPA pronunciation has improved. The dialogue mode supports multiple speakers, interruptions, and tone changes. v4 is positioned for content, audiobooks, and character voiceovers, with a single instance supporting about 10,000 characters; Turbo has a median inference latency of about 100 milliseconds and initial voice package latency of about 150 milliseconds, aligning with pauses in conversation. The API promotion runs until October 12 at approximately $0.022 per thousand characters (list price $0.08), and Turbo at about $0.011 (list price $0.04).
Runway has already integrated character videos with ElevenLabs agents: voice recognition, inference, and synthesis are handled by ElevenLabs, while Runway manages lip-syncing and visual layers. Incorporating v4 into the same canvas means that generated videos can directly match the new generation of dialogue without needing to export and paste from external tools. Google is also promoting Gemini voice studio capabilities around the same time, as video platforms compete for voiceover entry points.
In market mechanics, this is about distributing creative tools, not box office settlements. The buyers are video teams that need a workflow to produce videos with visuals and dialogue; the sellers are Runway selling subscriptions and computing power, and ElevenLabs selling characters and cloning. Funding is shifting from editing outsourcing and dubbing studios to character-based billing interfaces. Beneficiaries are platforms that can lock in voice lines for long videos; those under pressure are independent dubbing software and pay-per-line human voiceovers. Model preference tests placed Google voice in the control group, with a price war window open until mid-October.
Supplementary data: v4 documentation examples cover task dialogues, audiobooks with fighting jokes, and multi-turn emotional switches. The Runway Voices interface can generate character voice lines using text descriptions or samples. Universal Music Group and ElevenLabs have a multi-year strategic agreement as of September. Quality rankings and promotional prices are based on the public page on the day of release and may change thereafter.
Source: Public Information
ABAB AI Insight
Runway is not just selling another button; it aims to lock in the flagship voice lines released that day into its timeline, preventing users from leaving the canvas for dialogue. ElevenLabs uses blind testing and Elo to position Google’s new voice in the second tier of narrative, and then employs ten-second cloning and tag directing to turn "performable" into a product differentiator. Turbo’s 100 milliseconds is for seats and game NPCs, while v4 is for finished pieces; Runway’s choice of the latter indicates it is selling works rather than phone calls. The expansion to 90 languages allows the same video track to be released in multiple languages, which is closer to a distribution system than just adding another filter.
The capital path involves platform commissions layered on interface billing. Runway drives traffic through video entry, while ElevenLabs monetizes characters; the Universal agreement provides a layer of insurance for voice line copyrights. The promotional price reduces trial costs to a quarter of the list price, aiming to capture workflow before Google’s studio narrative takes hold. Enterprise calls now account for over half of revenue, with consumer-level creation being the second growth curve. Whoever becomes the default voice layer will occupy a line in the video generation billing.
Comparable examples include DAWs binding to specific synthesizers and YouTube making subtitles and voiceovers a platform feature. The difference is that voice lines can be cloned and tagged for directing, and character assets can be reused across collections. The industry phase is moving from multi-modal kits that can produce visuals to those that can generate distributable dialogue. Google is pursuing a creative studio approach, while Runway is integrating the strongest third-party voice lines available that day; both paths are competing for the same timeline.
Structural judgment falls under technological substitution. The mechanism is: voiceover costs shift from studio time to characters, and directing power shifts from actor rehearsals to tag sequences. Once the visual model is on screen with flagship TTS, the friction of outsourcing dialogue disappears, and pricing power shifts from dubbing companies to generation suites. Natural tone is merely the ticket; locking in workflows is the rent.
ABAB News · Cognitive Laws
Once voiceover enters the timeline, exporting becomes friction.
Ten-second cloning turns voice lines into reusable assets.
Rankings are valid on the day, and locking in workflows is what generates rent.