Audio reference support for music beats and character voice consistency

[lvsas]

Audio reference support for music beats and character voice consistency

I'd like to request a feature that would unlock a lot of creative use cases: audio reference inputs beyond the current video/image refs.

Use case

With MiniMax H3 (and similar AV models), you can pass a reference audio track that guides:

• Music beats/rhythm — sync generated visuals to a soundtrack
• Character voice consistency — maintain the same voice timbre across generated dialogue
• Soundscape continuity — ambient audio that the generated video should follow

Currently, the only way to pass audio is through the source video's own track (via LoadVideoUI). There's no way to provide a standalone audio reference.

Proposed solution

Add one or more canonical audio roles such as:

Role	Purpose
(Input:audio)	Generic audio reference (beat, music, voice)
(Input:audio-beat)	Music/rhythm reference for visual sync
(Input:audio-voice)	Voice timbre reference for character consistency

These would map to a LoadAudio-style node in ComfyUI and be passed to the model's ref_audios input.

Why this matters

Models like MiniMax H3 already support ref_audios in their conditioning pipeline. Without a way to expose this through Calliope's role system, users have to manually edit workflows or fall back to the source video's audio track which limits creative control.

Would this be something you'd consider for a future release? Happy to provide example workflows or test cases.

Thanks!

[benjiyaya]

(Input:audio) already have it.

(Input:audio-beat) , (Input:audio-voice) , what is the data type of those?

[lvsas]

Hey! Glad you already have (Input:audio)
I must have missed it. I tried uploading an audio file earlier and hit a bug (it failed silently), so I assumed it wasn't supported. Just tried again and it works! My bad.

To answer your question: (Input:audio-beat) and (Input:audio-voice) were just proposed sub-roles/asset type for more specific use cases.

But honestly, they're not necessary. If (Input:audio) already works generically, that's all we need.

[delgado3j-art]

Bonjour,

Je rejoint l'idée de pouvoir mieux gérer l'audio surtout pour les personnages et les associer aux personnages de manière automatique quand on passe sur le script et ensuite dans les videos !

[lvsas]

I'm also curious if it would be feasible to have an audio source that could be trimmed across multiple clips

I'm attempting to replicate the process I use for music videos.