Seed Audio 1.5 vs 1.0: what changed
The short answer: video
Seed Audio 1.0 already did more than read text aloud: one prompt could hold voices, effects and ambience. Version 1.5 keeps all of that and adds the thing 1.0 could not do: take a video as input. Dubbing, same-voice translation and sound for silent footage all follow from it.
We run Seed Audio 1.0 in Doitong's text-to-speech tool and tested 1.5 in early access.
What we saw running both
1.0 directs a voice from text
In our text-to-speech tool, Seed Audio 1.0 takes a free-text description of the voice: accent, mood, pace. One prompt can also carry effects and ambience around the speech.
1.0 would not take video
In September 2026 the 1.0 API already validated a video reference field, then answered that it was not implemented. Video work had to wait for 1.5.
1.5 translates whole scenes
With 1.5 we translated a 9-minute film into 12 languages, each track the length of the original, by sending the video itself.
Dubbing by hand in 1.0
Before 1.5 we dubbed a film into Spanish line by line with 1.0 and a voice reference. The voice held, but each line had to be timed and checked by hand. Video translation removes that loop.
Feature by feature
From ByteDance's Seed Audio 1.5 overview, which compares the two versions with a typical text-to-speech engine.
| Mode | Seed Audio 1.5 | Seed Audio 1.0 | Typical TTS |
|---|---|---|---|
| Text to speech | Yes | Yes | Yes |
| Voice cloning | Yes | Yes | Yes |
| Text to audio (voice, effects, music) | Yes | Yes | No |
| Text + reference audio | Yes | Yes | No |
| Text + video → audio (dubbing) | Yes | No | No |
| Text + video + voice reference | Yes | No | No |
| Music from text, up to 6 minutes | Yes | No | No |
| Capability | Seed Audio 1.5 |
|---|---|
| Video translation in the same voice | English or Chinese → 30 languages |
| Stem separation | Up to 6 tracks |
| Voice references per request | Up to 6 |
| Audio per request | Up to 6 minutes |
| Timestamp control | Voice, effects and music |
| Songs with lyrics | English, Chinese, Japanese, Korean, Cantonese |
Source: ByteDance, “Seed Audio 1.5 — Everything To Hear Is Here” and the early access integration guide, October 2026.
FAQ
What is the main difference between Seed Audio 1.5 and 1.0?
Video input. Seed Audio 1.5 accepts a video and produces audio that matches it, which enables video dubbing, same-voice video translation and sound for silent clips. Seed Audio 1.0 works from text and reference audio only.
Is Seed Audio 1.0 still useful?
Yes, for text-to-speech and text-driven sound design, where no video is involved. Doitong's text-to-speech tool runs on it.
Does 1.5 support more languages?
Seed Audio 1.5 speaks 30 languages, and video translation goes from English or Chinese into any of them.
Is Seed Audio 1.5 the same as Seed Audio 2.0?
Some sites list the new model as 2.0. ByteDance's early access documentation names it Seed Audio 1.5.
Related
Try the version that listens to your video
Translate, dub or score a clip with Seed Audio 1.5.
Start in Doitong