← sources

Minimax H3 audio sounds robotic despite giving it clear audio references

source: reddit (reddit:comfyui)published 2026-09-26 08:31ingested 2026-09-27 02:40

retained text

As the title says. I am giving Minimax H3 clear multiple reference audios, and yet the output sounds robotic and not human.

I am running with RTX 5060 Ti 16GB, and 48GB of RAM.

As for the files/nodes I'm running:

* Model: either Q8 GGUF or **ref2va\_pruned\_int8\_convrot**
* VAE: Kijai's video VAE and the official audio VAE
* Text encoder: Qwen 32b
* Sampling shift with video:12, audio:6.
* For VRAM savings, the following nodes:
   * MiniMaxChunkFeedForward
   * Low VRAM Attention
   * Comfy-Kitchen
   * Block/Sparse Attention Speedup for Single Pass
* No turbo LoRA

I searched for this issue, some people (though a small amount) reported this as well. They've been advises to switch the sampler to either eular or heun, but none of this worked for me. It **maybe** improved the audio quality a bit, but not enough to matter.

Any ideas?

document fetched 2026-09-27 00:38 · canonical https://reddit.com/r/comfyui/comments/1wqlblx/minimax_h3_audio_sounds_robotic_despite_giving_it