FishAudio — S2.1 Pro Launch Video | 2D Animation

Fish Audio builds open-source text-to-speech. S2.1 Pro is their production model: 83 languages, voice cloning, roughly 70ms to first audio, and natural-language control over emotion and delivery written inline with the text.

They launched it with free unlimited API access, into a market where every competitor's best voice sits behind a paywall. The news was the model. The problem was showing it.

Sound is a hard thing to advertise. You can't screenshot a voice, and a talking-head demo flattens exactly what makes this model interesting - the speed, the range, the control.

Most product films go picture-first and let sound support the visuals. This one inverts. The product is the sound, so every visual decision had to stay out of the audio's way and still hold attention.

We built the motion to track the voice rather than compete with it. Waveforms respond to the actual output. Latency reads as the gap it is - short enough the animation barely catches it. Language switches and inline emotion tags are events the audio drives, not decoration laid over the top.

When the model shifts from calm to excited mid-sentence, the frame shifts with it. The rule throughout: if the voice isn't causing it, it doesn't move.

The result is a film where the model demonstrates itself. You hear what it does while you watch what it means.

Film end to end: concept, design, and animation. A launch asset and a demo at once - and it holds up on a second watch, which is the real test for anything built to live on a product page.