Higgsfield adds ByteDance's Seed Audio, with 18-language dubbing
By Steven Van ·
The audio model handles voice changing, text narration, and video dubbing, and is also reachable from Claude via the Higgsfield MCP.
ByteDance's Seed Audio 1.0 is now live on Higgsfield, bringing the Seed team's all-in-one audio model to one of the most popular AI video platforms. Announced by Higgsfield on June 30, 2026, Seed Audio 1.0 can change a voice, narrate from plain text, or dub a finished video into 18 languages, all inside the same workspace where you create and edit your footage. Even better, it's available two ways: directly on Higgsfield, and inside Claude through the Higgsfield MCP.
What Seed Audio 1.0 Actually Does
Seed Audio 1.0 is ByteDance's unified audio-generation model, built by the same Seed team behind its speech and video research. Unlike a plain text-to-speech engine, it generates the full spectrum of sound from a text prompt: natural human speech, original music, foley-style sound effects, and environmental ambience. Its ecosystem capabilities include multi-speaker dialogue with a distinct voice per character, zero-shot voice cloning from a short reference clip, portrait-guided character voices, multilingual synthesis for dubbing, and fine-grained controls over tone and pacing. Launched in late June 2026, it is rolling out across the major AI platforms, including fal, WaveSpeed, Runware, Runway, and now Higgsfield.
Change a Voice or Narrate From Text on Higgsfield
On Higgsfield, Seed Audio 1.0 slots naturally into the two jobs creators reach for most. The first is voice changing: swap a narrator or character voice for a different one, or replace it with your own cloned voice learned from a short sample. The second is narration from text: paste any script, choose a language, tune the tone and pace, and get a polished spoken track in seconds. Because Higgsfield is already a full video and image platform, that audio arrives on the same timeline as your visuals rather than in a separate tool, so voiceover, music, and effects stay in sync as you build.
Dub Any Video Into 18 Languages
The standout feature for a video-first platform is dubbing. Seed Audio 1.0 lets you take a finished clip and re-voice it into 18 languages, keeping the delivery natural instead of robotic. For anyone localizing a course, a product demo, an ad, or a short film, this collapses what used to be a multi-vendor pipeline, recording talent, translating scripts, mixing tracks, into a single step inside Higgsfield. Paired with the platform's existing lip-sync and voice-cloning tooling, dubbing becomes something you can iterate on in minutes, reaching new markets without hiring separate voice actors for each region.
Use It From Claude via the Higgsfield MCP
The distribution angle is what makes this release notable: Seed Audio 1.0 isn't locked to the Higgsfield website. It's also reachable from Claude through the Higgsfield MCP. MCP, short for Model Context Protocol, is an open standard that gives AI agents access to external tools, so connecting the Higgsfield MCP means you can drive Higgsfield's models, including Seed Audio, using natural language directly inside Claude. In practice you add the Higgsfield MCP server in Claude's connector settings, authenticate with your Higgsfield account, and then simply ask Claude to narrate a script, change a voice, or dub a clip, no API keys to juggle. It turns audio generation into a conversational step alongside your other creative work.
Sources: Higgsfield MCP, Higgsfield Voice Cloning, ByteDance Seed — Speech.