← sources

reference voice not working in minimax h3 ref2va

source: reddit (reddit:StableDiffusion)published 2026-10-24 19:27ingested 2026-10-25 02:38

retained text

I will try to keep it short as I have tried basically everything I could with the help of LLMs to try to troubleshoot, so just saying everything I tried I think is pointless >!(yes I fed the prompting guide to LLMs, yes I tried to rewrite the prompt with different LLMs, yes I tried different workflows, yes I tried with/without turbo loras, yes I tried to change the time gen, yes I tried to match the time of the audio reference with the time of the generated audio, yes I did many testing changing only 1 parameter at a time to try to isolate the issue properly spending an entire day on this and on and on with stuff such as using different audio formats)!<, trust me I tried basically everything possible.

**What I want:** *an audio file of a short sentence sentence working as a reference for what the voice of a character should be.* Example: Audio file A is a sentence saying "The eggs are boiled come and eat" with Voice A. Using a ref2va workflow + the proper prompting, character 1 says a different sentence "I enjoy having a walk" while having Voice A. 

**What actually happens**: *Voice A reference gets completely ignored and instead a random, unrelated voice X is being used when the character is talking.* Example: Audio file A is a sentence saying "The eggs are boiled come and eat" with Voice A. Using a ref2va workflow + the proper prompting, character 1 says "I enjoy having a walk" with a random unrelated voice X.   
  
The only 'fix' working: The ONLY thing that works is if the audio file is 'saying' the exact same sentence that the character will say, basically as if there is a pass-through. Example: Audio file A is a sentence saying "I enjoy having a walk" with Voice A. Using a ref2va workflow + the proper prompting, character 1 says "I enjoy having a walk" with Voice A. 

What is crazy is that the reference images work perfectly, but trying to use a reference audio in the same way for a voice seems to just not work at all. And it seems like I am not alone: [https://huggingface.co/MiniMaxAI/MiniMax-H3/discussions/64](https://huggingface.co/MiniMaxAI/MiniMax-H3/discussions/64) "I'm just responding to say that I don't think it works. I have been over the guide in every way possible, and the only reference audio that works for me is direct pass-through. I would love to be wrong and have someone explain how they did it (and not just say "read the guide")."  

document fetched 2026-10-25 00:29 · canonical https://reddit.com/r/StableDiffusion/comments/1xehip4/reference_voice_not_working_in_minimax_h3_ref2va