HeyGen break tag: 5-second pause max, and a longer pause on Sume

HeyGen's changelog says text-to-speech on POST /v3/models/audio/tts and /stream supports <break time="0.5s"/> with professional voice clones, with pause values in seconds and a maximum of five seconds per pause. On Sume there is no tag: an avatar video pause is a silence scene in video_inputs, whose duration is required and can be up to 60 seconds, inside the 4-60 second video window.

What does HeyGen's September 2026 entry say?

The entry is titled "Professional Voice Pause Tags" and dated September 2026. It clarifies that the break tag works with a professional voice clone. It also says other SSML tags are unsupported, and that break tags for instant voice clones are unsupported.

That makes the tag a voice-clone feature: whether a pause is possible depends on which voice type you use, not only on the endpoint.

How do I write a pause on Sume?

Use ordered video_inputs instead of one script. A silence scene needs type: "silence" and a duration; script and input_text are not allowed on it. Spoken scenes use type: "text" with exactly one of script or input_text.

The total planned length must still land in 4-60 seconds, so a long pause uses up the same budget as speech. The docs also say current execution supports one avatar per final video and expects scene backgrounds to resolve to one shared scene.

{
  "avatar_handle": "product_host",
  "video_inputs": [
    {"id": "line1", "voice": {"type": "text", "script": "Here is the old setup."},
     "background": {"type": "prompt", "prompt": "Bright studio"}},
    {"id": "beat", "voice": {"type": "silence", "duration": 8},
     "background": {"type": "prompt", "prompt": "Bright studio"}},
    {"id": "line2", "voice": {"type": "text", "script": "And here is the new one."},
     "background": {"type": "prompt", "prompt": "Bright studio"}}
  ]
}

What is different about a pause as a scene?

Writing a pause as a scene changes how you plan it.

• It is a separate non-speaking scene, so the avatar is not asked to speak during it.
• There is no per-tag cap of 5 seconds; the cap is the scene duration maximum of 60 and the 4-60 second video window.
• For a short pause inside a sentence, split it into two text scenes with a silence scene between them; the schema allows a duration as small as 0.001 seconds.
• Send an Idempotency-Key on the create call so a retry does not start a second paid job.