HeyGen ElevenLabs v3 model_id per request vs Sume TTS

HeyGen's POST /v3/voices/speech now accepts settings.model_id with engine: "elevenlabs", so one request can pick eleven_v4, eleven_v3 or another listed model. Sume's TTS 1.0 does the opposite: its tool contract forbids provider model ids and any model picker. A separate TTS Router lists explicit models, and its v1 is Sonic only.

HeyGen facts are from its September 2026 changelog; Sume facts are from the OpenAPI description of the TTS Router and the TTS tool description in the Sume codebase, read 2026-09-30. See also the Public API reference.

What does HeyGen's model_id do?

The changelog lists eleven_v4, eleven_v3, eleven_multilingual_v2, eleven_flash_v2, eleven_flash_v2_5 and eleven_turbo_v2_5. The setting overrides the voice's saved model preference for that request only; omit settings to keep the preference. The older elevenlabs_v3 engine is deprecated.

How does Sume choose a speech model?

For TTS 1.0 (POST /v1/tts-1.0/generate, public model id sume/tts-1.0), you pick a voice, not a model. The tool description says there is no provider model id, credentials or model/model_id picker, and a request over 1200 seconds fails with tts_duration_exceeded. The TTS Router is a catalog of explicit pass-through models, invoked with the model in the body.

Can I get an ElevenLabs model on Sume?

Not from what the docs quoted here show: the router's v1 catalog is Sonic only, and TTS 1.0 exposes no model choice. If a pipeline depends on a named ElevenLabs model, keep that call outside Sume or adjust the voice and script instead.

What should I change when porting a speech call?

Drop model_id, send the transcript with a voice id, and read results by job. For the Sonic route, see ElevenLabs API alternative: text to speech with Sonic.