standard Minimax H3 Ref2v workflow fails to output ref audio.

### Custom Node Testing

- [ ] I have tried disabling custom nodes and the issue persists (see [how to disable custom nodes](https://docs.comfy.org/troubleshooting/custom-node-issues#step-1%3A-test-with-all-custom-nodes-disabled) if you need help)

### Your question

I ran the same inputs (image, video, and prompt) through the Minimax H3 Ref2v API and locally through the attached (standard) workflow. The local workflow gives no audio output (just an occasional click) and random lip motion. The API worked perfectly. ISame bad result with/without turbo Lora.

### Logs

```powershell

```

### Other

_No response_

Start by running the attached standard Minimax H3 Ref2v workflow with the same image, video, prompt, and optional Turbo LoRA settings used against the API. Compare the local result with the API output and inspect why reference audio is absent and lip motion is inconsistent. Done means the local workflow produces the expected reference audio and matching motion.

Written by the indexing model from the issue text.