**[MiniMax Open-Sources Music3: Single Generation of Complete Songs up to 5 Minutes Long, Outputting 32kHz, 16-bit Stereo Audio]** According to monitoring by Beating, MiniMax has officially released the MiniMax-Music3 model weights, along with an online demo and ComfyUI workflow. It can generate complete songs up to 5 minutes long in a single pass based on lyrics and music descriptions, outputting 32kHz, 16-bit stereo audio. The standout feature of Music3 is its controllability. Lyrics can be arranged directly into song structures using tags like `[Intro]`, `[Verse]`, `[Chorus]`, `[Bridge]`, `[Solo]`, `[Outro]`. Prompts can specify musical style, BPM, key, emotional shifts, vocal timbre, harmonies, instruments, and the arrangement of each section. Moreover, this time users can truly download the model and run it locally. According to the official statement, full precision requires less than 24GB of VRAM, while offloading to CPU requires about 22GB. With further layer-by-layer loading, it can even fit into an 8GB GPU. ComfyUI already offers native support and provides an INT8 model for low-VRAM usage. Among open-weight music models, Music3 has been regarded by many users as the strongest option currently available. Following H3, MiniMax has once again released a heavyweight generative model directly to the public. [Original Link]