MiniMax H3 RefMod Audio Reference

https://youtu.be/H5L5sO7dRu8

Creator: AI-KSK · Collection: Video 327 / six-workflow MiniMax H3 RefMod episode

Workflow file: 03_音频RefMod_单素材制作.json · Cover file: 琥珀录音棚音频参考-3.png

Main features:

Instead of sending a voice recording through a visual image/video loader, this workflow encodes one audio source as a reusable H3 RefMod. It demonstrates the audio branch of the library and creates an asset that can be chosen in a later video generation graph. It is not a speaker-finetuning workflow, a text-to-speech engine or a guarantee of exact voice identity and lip sync.

The connected nodes are LoadAudio → MiniMaxH3RefModAudioExtract, with VAELoader providing the MiniMax H3 audio VAE. The output is inspected with MiniMaxH3RefModInspect and PreviewAudio. This graph does not use the image-folder loader from the character workflow.

Exact values from this JSON

- Original audio selection: 女主.mp3 — not bundled.
- VAE: minimax_h3_audio_vae_fp32.safetensors.
- Extraction: name=Rin_Voice_01, max_seconds=8, max_tokens=0, budget_policy=truncate, concept_type=voice, subfolder=voice, save=true.
- Inspect node: preview=stored, strength=0.5; the short preview is not the source extraction duration.

Suggested workflow:

Upload a rights-cleared audio recording and select it in LoadAudio. Prefer an isolated, clean voice sample and check the actual eight-second cap. Run the extractor, inspect the preview and locate the saved asset in models/refmods/voice/. In Generation A it appears as voice/Rin_Voice_01 in the fourth loader slot. The episode transcript warns that background music in an example source influenced later generation; this graph does not automatically strip music or guarantee identity-preserving speech.

Limits

A source recording may carry ambience, music or other speakers into the reference representation. The preview duration does not establish the amount of audio internally retained. The microphone-and-waveform cover is a graphic depiction of an audio workflow, not a recorded voice test.

Dependencies, tutorial and deployment

This project uses ComfyUI, MiniMax H3 weights and the upstream RefMod node pack; workflows involving VHS_ nodes require Video Helper Suite. See Comfy-Org's H3 model inventory for official distribution guidance. The exact int8-convrot video VAE and Turbo 4-step LoRA named in the creator's JSON need separate availability/provenance checks. We distribute this workflow JSON and its original cover, not external checkpoints or third-party media. Respect applicable rights and licenses.

Deploy AI-KSK's H3｜RefMod免训练超级参考 image on 优云智算. Referral disclosure: the deployment link includes an AI-KSK referral code and may support the creator. Check the cloud provider's actual billing and environment before deploying.

Watch the complete six-workflow YouTube tutorial. This project is one module in that episode. Description and settings were prepared by inspecting the source JSON and transcript, not from a new measured benchmark.

Bilibili tutorial: https://www.bilibili.com/video/BV1F2nZ6SEA7/