multimodalart's picture
multimodalart HF Staff
Update README.md
1866d52 verified
|
Raw
History Blame Contribute Delete
1.61 kB
metadata
title: MiniMax Music 3 Workflow
emoji: 🎵
colorFrom: pink
colorTo: purple
sdk: gradio
sdk_version: 6.24.0
app_file: app.py
pinned: true
suggested_hardware: zero-a10g
hf_oauth: true

MiniMax Music 3 — diffusers demo as a gr.Workflow

Generates full songs from tagged lyrics + a structured caption using the MiniMaxMusic3Pipeline diffusers port, presented as a visual, node-based gr.Workflow canvas (workflow.json):

Lyrics + Global Metadata + Vocal Details + Arrangement (editable reference nodes, defaults per the official prompting guide) → generate_song (@spaces.GPU ZeroGPU worker: AR frames → windowed DiT decode → vocoder → wav) → Output Song / Seed Used / Stats → make_video (CPU ffmpeg visualizer) → Share Video.

Duration, seed, randomize-seed, flow-matching steps, guidance scale, and video title are also editable reference nodes. hf_oauth: true lets the owner edit the canvas; visitors can run the pipeline. The original Blocks app (with the live-streaming PCM player) is preserved as app_blocks.py — workflow fn nodes are plain callables, so chunk-by-chunk streaming is replaced by a final audio subject.

  • Weights: MiniMaxAI/MiniMax-Music3
  • AoTI kernels: diffusers-internal-dev/MiniMax-Music3-aoti (compiled on RTX Pro 6000, matching ZeroGPU hardware)
  • Generation streams chunk by chunk with a configurable playback headroom. The 8B language-model stage runs eager on ZeroGPU (its JIT StaticCache ladder needs a persistent process); AoTI-exporting the LM decode step per cache bucket is the follow-up that brings the extra ~1.9x.