Which Sume video models take an audio_url reference, which do not

Check supported_input_references on the model row in GET /v1/videos/models. The Sume docs show seedance-2 with image_url, video_url and audio_url, say MiniMax H3 Max reports image, video and audio references, and say Gemini Omni Flash 1.1 accepts image and video references with no audio. A model that does not list audio_url cannot take an audio reference.

What the docs list

The statements come from the Videos docs. The seedance-2 entry is the sample row in the docs, so confirm it against the live response before you rely on it.

Why it matters

An audio reference lets you steer a clip with a voice or a music track. If you send an unsupported reference type to a model that does not list it, expect a 400 rather than a quiet ignore, which is how Sume treats unlisted fields on its image route. The video route docs say the same of unsupported fields, so read the row first.

• Omni Flash 1.1 has native synced audio generation, so you may not need an audio reference at all.
• generate_audio defaults to the audio capability of the model.
• For a specific voice, generate the speech first, then use a model whose row lists audio_url.