Generate full-length songs from lyrics and a music description.

Done with

MiniMax-Music3

How

An 8B global model tracks structure across the whole song while a 0.6B local model fills in frame-level acoustics. Synthesis fuses their continuous hidden states instead of decoding from discrete tokens.

Result

Five-minute songs rather than short clips that must be stitched together.

Replaces

Stitching together short generated music clips.

Products from this video

Agent-Safe Pipeline Anti Slop apple-liquid-glass-webgl ASD-STE100 Skill BlueFerry book-to-skill Cinematic Zoom claudish-to-english Colite DeepSeek Harness dsh-anchored-standard Ember eve Software Factory Graft ha.mr Human Review legal-skills lumabri MCP-Memory mcptoon MiniMax Music 3 Multi-Agent Workbench OpenAnalytics oss-pr-reviewer pdfcn pgbot pi-transcribe Rakazo Skills for Designers and Engineers TermDOM tokentab ttfx ZSvirt

Related use cases

• Transcribe multi-instrument audio into note events or MIDI files.
• Generate up to 15-second videos with synchronized native stereo audio from text, image, video, and audio references.
• Local video generation
• Generate real-time multilingual subtitles with speaker labels and live translation on a CPU.
• Generate short videos with synchronized audio locally on Apple Silicon Macs.