MiniMax Music 3: Text to Music
Create full-length songs from text prompts with MiniMax Music 3, generating complete 5-minute tracks with stable structure and high-fidelity audio. Input a structured caption describing lyrics, style, mood, and arrangement, and receive a finished song as output. Ideal for music producers prototyping demos, and content creators needing original background tracks.
试试一个示例
关于此工作流
This ComfyUI tutorial workflow turns structured text into full-length music using MiniMax Music 3. At its core is the MiniMax Music 3 generator node (UUID: ac99f841-a3de-4329-9564-953b81cf9e16), which accepts two key inputs: a Caption describing style, mood, tempo, instrumentation, and arrangement, and optional Lyrics formatted with section tags like [intro], [verse], [chorus], [bridge], and [outro]. The node renders a complete track—up to five minutes—with stable song structure and high-fidelity audio.
A MarkdownNote block in the graph documents the prompt schema so you can quickly copy, tweak, and version your captions and lyrics inside the UI.
Under the hood, you can choose between two model weights: minimax_music3_dit_fp16.safetensors (recommended when VRAM allows) and minimax_music3_dit_int8_convrot.safetensors (optimized for lower VRAM). The generator node produces an audio tensor that is written to disk via SaveAudioAdvanced, where you control filename, format, and sample rate. The result is a finished, arranged song suitable for prototypes, background scores, and content production without leaving ComfyUI.
常见问答
Use minimax_music3_dit_fp16.safetensors if you have enough GPU VRAM for best fidelity and stability. Choose minimax_music3_dit_int8_convrot.safetensors when VRAM is limited; it’s more memory-efficient with a potential slight quality trade-off.
Keep the Caption explicit: genre, mood, tempo/BPM, key instruments, production style, and arrangement (e.g., intro–verse–chorus–bridge–outro). For Lyrics, include clear section tags like [intro], [verse], [chorus], [bridge], [outro]. Keep lines concise and align section lengths with your requested duration.
If you provide Lyrics with section tags, the model will attempt to render vocal-like parts following your text and structure. If you leave Lyrics empty and supply only a Caption, it will generate instrumental music.
The workflow supports up to about 5 minutes (300 seconds). Set the duration parameter on the MiniMax Music 3 generator node and keep your arrangement (in Caption/Lyrics) sized to fit that length.