MCP topview_generate_video rejects inputAudios despite Wan 3.0 Omni / Board support

[AI-KIMCHY]

Summary

topview_generate_video (omni_reference) cannot accept reference audio. The MCP schema has no inputAudios field and sets additionalProperties: false, so the gateway rejects the param before it reaches the API. The tool description also states reference audio is not a video submit input.

Why this matters

Wan 3.0 Omni on Board (and the public Omni Reference API) already supports reference audios for lipsync / timing. Music-video and audio-driven workflows need MCP parity with Board.

Evidence

• Backend / Board Wan 3.0 omni config: hybrid uploads allow audios (e.g. maxAudios, max total audio duration).
• Board UI submit with Image1 + Image2 + Audio1 succeeds for Wan 3.0 Omni.
• MCP call with inputAudios fails schema validation (inputAudios is not allowed).
• Canvas generation capabilities expose reference_image / reference_video but no reference_audio role.

Request

1. Add optional inputAudios (fileId + name) to topview_generate_video for taskType=omni_reference when the selected model supports audio refs.
2. Update the tool description: reference audio IS a valid video submit input for models that support it (do not route all audio refs to topview_generate_music).
3. Optionally expose a reference_audio input role on Canvas video capabilities for the same models.

Workaround today

Manual Board Omni UI submit with audio attached.