Gemini Omni audio reference: unsupported; Sume models that take audio

Gemini Omni Flash does not take an uploaded audio file as a reference: Google's limitations list says "Uploading audio references is unsupported in the current version of the API." Sume's Video Router page agrees for its Omni row: no reference_audio_urls, and native audio is always on. On Sume, audio references go to other models; the Video generation page says Seedance 2.x, Wan 3.0, MiniMax H3 and MiniMax H3 Max honor audio and video references.

Google's statement is from its Gemini API: Generate and edit videos with Gemini Omni Flash, read 2026-10-02. Sume's is from Video Router and Video generation, same date. The OpenAPI request schema has a reference_audio_urls field in general, so the limit is per model, not per route.

How do I steer Omni's audio, then?

With the prompt. Google's guide says that by default the model tries to generate an appropriate audio track, and you can describe the audio you want, especially music. Its examples include "Include calm background music" and "The video has a high energy techno beat". It also suggests simple negatives such as "No dialogue" or "No extra sound effects".

A video reference does not carry sound either: Google says any audio in a video reference is ignored.

Which Sume models take reference audio?

The Video generation page states which families honor audio and video references and which accept video only. Check supported_input_references for the model in GET /v1/videos/models before you send an audio URL; a model that does not list audio_url will not use it.

What if I need my own voice track under an Omni clip?

Generate the clip first and add your own sound as a separate step, outside the Omni request. Sume's docs list audio references only for the models in the table above, so do not send an audio URL to Omni.