MiniMax H3 Audio Cloning Lacks SFX, Dev Explores Foley LoRA Solution
dev_ne · reddit · 2026-08-13
A developer on Reddit reported issues when using the MiniMax H3 model for audio cloning. While the model accurately clones character voices and delivers lines as intended, it has two major drawbacks: the speech rate is somewhat slow, and it lacks background music or sound effects (SFX), such as footsteps.
Addressing this pain point, the author asked the community if others have experienced similar issues and whether training a dedicated Foley LoRA to work alongside voice cloning could solve the lack of environmental audio.
More from Multimodal
- Flux 3 Klien first impressions: occasional body horror like SD3, otherwise solid — rionix88 · 2026-10-16
- XGRIDS builds world's largest 3DGS town: 23.5 km² and 6.2 billion Gaussian Splats — bilawalsidhu · 2026-10-16
- IMMERSION: AI film compilation links 12 shorts with one continuous camera move — AI_Andrew · 2026-10-16
- A song with vocals, lyrics and 3D head animation created entirely by Opus 5.5 — repligate · 2026-10-16
- Alibaba open-sources CommerceGen-Studio for commercial-grade e-commerce images — alibabagroup · 2026-10-16
- EVA-Cap: Fudan and OPPO align video captioning with atomic audiovisual events via GRPO — Fudan-University · 2026-10-16