gradio==5.35.0
transformers
torch
accelerate
bitsandbytes
soundfile
numpy
librosa

# LFM2.5-Audio end-to-end model
liquid-audio

# Qwen3-TTS (voice cloning + stock voice for narration)
qwen-tts>=0.1.1
huggingface-hub>=0.23.0

# FlashAttention-2: not in requirements (needs CUDA at build time).
# Code auto-falls back to eager/SDPA if unavailable.
# To enable: pip install flash-attn --no-build-isolation (at runtime)