MiniMax Music3: Free AI Music Gen, 5-Min Japanese Vocals

China's MiniMax has released a new music generation AI called "Music3" for free. This model can create songs up to five minutes long with Japanese vocals from text instructions. It outputs music files as 32kHz, 16-bit stereo. MiniMax announced the release on its official X account, calling it a "Next-Generation Open-Weights Production-Ready & Versatile Music Model." The model weights are available on Hugging Face and GitHub. There's also an AI skill that generates long lyrics and structured prompts. The demo app lets users input prompts in Japanese, like requesting rock music with specific phrases. The system then generates lyrics, a structured prompt, and a music track with a video file. While the demo app uses a free inference service with limitations, "Music3" supports local generation via ComfyUI. This means unlimited song creation is possible in local environments. What's interesting is this follows MiniMax's late-July announcement of its video generation AI, "MiniMax H3." That model can generate videos up to 1080p resolution and 15 seconds long with native stereo audio. The bottom line is these free and locally runnable AI models offer new creative tools for music and video production.