Skip to content
TheChatArchiveThe Chat Archive
gpt-4.1-mini-2025-04-14· OpenAIAi2 WildChat-4.8M (AI2)

Are there any music creation/editing AI's on huggingface, or anywhere else, that can…

  1. User

    Are there any music creation/editing AI's on huggingface, or anywhere else, that can take an existing music track and create an extended version of it?

  2. Assistant

    Great question! Extending an existing music track with AI involves generating new audio that seamlessly continues or elaborates on the original piece, which is a challenging task. Here’s a summary of the current state of AI music generation tools relevant to your request:

    AI Models on Hugging Face for Music Creation/Editing

    • Music Generation Models on Hugging Face (like OpenAI's Jukebox) can generate music given a prompt or description, but they mostly generate from scratch or based on style/genre tokens rather than extending a specific existing track.

    • Other models on Hugging Face include MusicVAE and DDSP (Differentiable Digital Signal Processing), which can interpolate or transform music but are mostly focused on MIDI or symbolic music, not raw audio, which limits their ability to directly "extend" an audio track.

    • Latent Diffusion Models for audio, found occasionally, focus more on synthesis or style transfer rather than extending a given track.

    External AI Tools for Extending Music Tracks

    1. OpenAI Jukebox

      • Can generate music continuations conditioned on an existing snippet. It’s a raw audio VAE model trained on a large dataset, capable of generating music continuations in various genres.
      • Limitations: High compute, output can be noisy or low-fidelity; works better with shorter inputs.
    2. Riffusion (Hugging Face)

      • Uses latent diffusion models to generate music/riffs from text prompts or an image of a spectrogram, not exactly extending an existing audio clip, but can generate new riffs in similar styles.
    3. AIVA, Amper, Soundraw (commercial AI music platforms)

      • Mostly focused on creating new compositions; some offer "variations" but not straightforward extension of arbitrary audio tracks.
    4. DDSP (Google)

      • This end-to-end differentiable synthesizer can do timbre transfer and interpolation of instrument sounds, but extending an arbitrary track would require significant manipulation and is not an off-the-shelf solution.
    5. Audio Source Separation + Remixing

      • You can separate stems from the existing track (using models like Spleeter), then use AI-generated MIDI accompaniment (from tools like MusicTransformer or Magenta) to create new accompaniment parts, then mix — but this is a multi-step manual process.

    Summary

    • Extending an existing arbitrary music track with AI, maintaining the exact audio characteristics while adding new content, is still mostly experimental and resource-intensive.
    • Models like OpenAI Jukebox are the closest existing state-of-the-art models capable of conditioned music continuation but require technical setup and significant compute power.
    • On Hugging Face, you won’t currently find an easy plug-and-play model that takes an audio file and extends it seamlessly. Symbolic music models help if you work with MIDI.

    Recommendations if You Want to Try

    • Investigate OpenAI Jukebox open-source code for audio continuation.
    • Use music transcription/analysis tools combined with AI symbolic music generation to create MIDI extensions and synthesize audio.
    • Experiment with audio segment interpolation using DDSP if you have a specific instrument track.
    • For a more user-friendly approach, try commercial AI-assisted music tools for creating variations or longer versions of generated tracks.

    If you want, I can help you find code, demos, or tutorials for any of these approaches!