Music & Adaptive BGM Generation¶
Audio
Open Source¶
-
ACE-Step Discovered by AI
2026
A highly efficient open-source music foundation model designed to run locally on consumer hardware, generating high-quality full-length songs with vocals in seconds.
local-aimusic-generationsuno-alternative -
HeartMuLa Discovered by AI
2026
An LLM-based open-source music foundation model family capable of synthesizing high-fidelity, long-form music conditioned on lyrics, tags, and reference audio.
foundation-modelllmlyrics-to-songmusic-generation -
MiniMax Music 3.0 Discovered by AI
2026
A high-performance open-weights music generation model capable of creating complete, structurally coherent songs up to five minutes long. It combines an 8B Global LLM for long-range musical structure, a 0.6B Local LLM for frame-level acoustic detail, and a continuous hidden-state synthesis system based on Flow Matching and Flow-VAE.
bgmmusic-generationtext-to-music -
Muse Discovered by AI
2026
A fully open-source system for long-form song generation with fine-grained style conditioning, designed to address the reproducibility gap in academic music generation. It includes a licensed synthetic dataset, training and evaluation pipelines, and an easy-to-deploy song generation model.
music-generationopen-sourcesong-generation -
Stable Audio 3 Discovered by AI
2026
A family of fast latent diffusion models for high-fidelity, variable-length audio and music generation, supporting text prompting and audio editing.
audio-editinglatent-diffusionmusic-generationGitHub · arXiv · Hugging Face · Paper
-
V2M-Zero Discovered by AI
2026
A zero-pair video-to-music generation framework that produces temporally aligned background music matching visual events without requiring paired video-music data during training.
bgmtemporal-alignmentvideo-to-music -
VIBE Discovered by AI
2026
A text-and-video-to-music generation model that ensures background music aligns with explicit user text instructions and visual cues through a depth-wise cross-layer conditioning mechanism.
bgmcontrollablevideo-to-musicGitHub · arXiv · Hugging Face
-
YuE2 Discovered by AI
2026
An advanced open-source music generation foundation model that unifies symbolic planning and audio generation. It uses a Mixture-of-Transformers (MoT) architecture to first generate a readable score (specifying melody and harmony) and then realize it as high-fidelity full-song audio with vocals, supporting zero-shot covers and agentic music editing.
bgmlyrics-to-songmusic-generationtransformer -
ACE-Step Discovered by AI
2025
An open-source music foundation model that generates high-quality, full-length songs with vocals and instruments from text prompts and lyrics, supporting local execution on consumer GPUs.
local-ailyrics-to-songmusic-generationopen-weightsProject · GitHub · arXiv · Hugging Face
-
ACE-Step 1.5 Discovered by AI
2025
ACE-Step 1.5 is a local open-weights music generation foundation model capable of producing full tracks with instrumentation and genre controls offline across CUDA, Mac, AMD, and Intel hardware.
bgmfoundation-modelmusic-generationtext-to-music -
DiffRhythm Discovered by AI
2025
An open-source, latent diffusion-based song generation model capable of synthesizing complete stereo songs (including both vocals and accompaniment) up to 4 minutes and 45 seconds long in just 10 seconds. It utilizes a non-autoregressive structure to achieve fast inference speeds.
bgmdiffusionlatent-diffusionmusic-generation -
DiffRhythm 2 Discovered by AI
2025
An end-to-end semi-autoregressive song generation framework based on block flow matching, designed for high-fidelity, controllable full-length song generation with precise lyric alignment.
bgmmusic-generationsoundtrack -
InspireMusic Discovered by AI
2025
An open-source music, song, and audio generation framework by Alibaba's FunAudioLLM team. It combines an autoregressive transformer language model with a super-resolution flow-matching model to generate high-fidelity, long-form music (up to 8 minutes) at 48kHz.
flow-matchinglong-formmusic-generationsuper-resolutionGitHub · arXiv · Hugging Face
-
JAM Discovered by AI
2025
A lightweight flow-matching song generator designed for fine-grained controllability and aesthetic alignment, enabling high-quality music and song synthesis.
bgmmusic-generationsoundtrack -
Magenta RealTime Discovered by AI
2025
An open-weights live music generation model developed by Google's Magenta team that produces a continuous stream of music in real-time with synchronized user control via MIDI, text, or audio prompts.
googlemidimusic-generationreal-time -
MusicGen-Stem Discovered by AI
2025
An autoregressive multi-stem music generation model that generates separate stems (bass, drums, and other) instead of a single stereo mix, allowing for high-quality music editing and adaptive mixing.
audiocraftmusic-generationstemsGitHub · arXiv · Hugging Face
-
SongGen Discovered by AI
2025
A single-stage autoregressive transformer designed for controllable text-to-song generation, producing high-quality full songs with vocals and accompaniment from text prompts.
lyrics-to-songmusic-generationopen-sourcetransformer -
YuE Discovered by AI
2025
An open-source foundation model for full-song music generation, capable of transforming lyrics into complete songs with high-quality vocals and instrumental accompaniment.
foundation-modellyrics-to-songmusic-generation -
YuE Discovered by AI
2025
An open-source full-song music generation foundation model based on the LLaMA2 architecture, designed to transform lyrics into complete songs (lyrics-to-song) up to five minutes long. It decouples vocal and accompaniment tracks and supports versatile style transfer and bidirectional generation.
bgmfoundation-modellyrics-to-songmusic-generation -
ChatMusician Discovered by AI
2024
ChatMusician is an open-source large language model integrated with intrinsic musical abilities, capable of understanding and generating music using text-compatible ABC notation. It treats music as a second language, enabling symbolic music generation and reasoning without external multi-modal neural structures.
abc-notationllmmusic-generationsymbolic-music -
FluxMusic Discovered by AI
2024
A text-to-music generation model that extends diffusion-based rectified flow Transformers to operate in a latent VAE space of mel-spectrograms, improving semantic alignment and audio quality.
diffusionmusic-generationtext-to-musictransformer -
Instruct-MusicGen Discovered by AI
2024
An instruction-tuned music language model based on MusicGen that enables controllable text-to-music editing, such as adding, removing, or separating musical instruments and stems, while preserving the original audio structure.
diffusionmusic-editingmusic-generation -
JASCO Discovered by AI
2024
A temporally controlled text-to-music generation model based on the Flow Matching paradigm. It supports both global text descriptions and fine-grained local controls (such as chords, melody, and drum tracks) using a novel conditioning method.
bgmcontrollableflow-matchingmusic-generation -
MetaBGM Discovered by AI
2024
A dynamic soundtrack transformation framework designed for continuous multi-scene experiences, enabling background music to adapt in real-time to environmental transitions and user interactions.
adaptive-musicbgminteractive -
Multi-Track MusicLDM Discovered by AI
2024
An extension of the MusicLDM latent diffusion model designed for multi-track music generation, allowing the synthesis of aligned individual instrument stems (e.g., drums, bass, melody) rather than just a single mixed audio track.
diffusionmulti-trackmusic-generation -
SONIQUE Discovered by AI
2024
A video background music generation model that leverages unpaired audio-visual data, combining LLM-based video understanding with conditional diffusion models to generate customizable, high-quality soundtracks.
bgmdiffusionvideo-to-music -
Stable Audio Discovered by AI
2024
A latent diffusion model architecture for generating high-quality, variable-length stereo audio and music at 44.1kHz, conditioned on text prompts and timing embeddings.
audiodiffusionmusicGitHub · arXiv · Hugging Face
-
AudioCraft Discovered by AI
2023
Meta's deep learning library for audio generation, featuring MusicGen for text-to-music and melody-conditioned generation, and EnCodec for high-fidelity compression.
audiocraftmusic-generationtext-to-music -
2023
A latent diffusion model for text-to-audio and text-to-music that uses a unified 'language of audio' representation to generate speech, sound effects, and music.
diffusionmusicopen-weightstext-to-audio -
AudioLDM 2 Discovered by AI
2023
A unified latent diffusion framework for text-to-audio and text-to-music generation. It introduces a language of audio (LOA) representation to bridge different modalities, enabling high-quality, controllable generation of music, speech, and sound effects.
diffusionlatent-diffusionmusic-generationtext-to-audioGitHub · arXiv · Hugging Face
-
2023
A universal high-fidelity text-to-music diffusion model combining autoregressive and non-autoregressive training to improve text-music alignment and audio quality.
diffusionhigh-fidelitytext-to-music -
2023
An LM-guided diffusion model (MeLoDy) for efficient music audio generation that reaches state-of-the-art quality while sharply reducing the number of sampling forward passes.
diffusionefficientlm-guidedtext-to-music -
2023
A long-context latent diffusion model for text-to-music that generates multiple minutes of high-quality stereo music from textual descriptions.
diffusionlong-contextstereotext-to-music -
2023
A single-stage controllable music generation model conditioned on text and melody, used for prototyping background music and adaptive scoring.
music-generationopen-weights -
2023
An open-source PyTorch implementation of Google's MusicLM, which generates high-fidelity music from text descriptions via hierarchical modeling of discrete audio tokens.
audio-tokenstext-to-musictransformer -
Mustango Discovered by AI
2023
Mustango is a music-domain-knowledge-inspired text-to-music system based on diffusion that allows users to control generated music using rich text captions. It supports specific instructions related to chords, beats, tempo, and key by incorporating musical features into the training process.
bgmcontrollablediffusionmusic-generation -
2023
Google's series of cascaded diffusion models that generate high-quality 30-second music clips from text, faithfully reflecting genre, tempo, instruments, mood, and era.
cascadeddiffusiongoogletext-to-music -
2023
Generates high-fidelity music via masked acoustic-token modeling with a bidirectional transformer, supporting synthesis, compression, inpainting, and variation.
inpaintingmasked-transformervariation -
2022
An adversarial multimodal framework that generates complex music conditioned on dance videos, using a vector-quantized audio representation for diverse dance-music styles.
ganmultimodalvideo-to-music -
2022
A Transformer for symbolic music generation that uses fine- and coarse-grained attention to model long sequences and capture musical repetition structure efficiently.
long-formstructuresymbolictransformer -
2022
A fast, user-controllable music generation system that synthesizes arbitrary-length waveform music much faster than real time on a consumer CPU.
fastganinfinitewaveform -
2022
A fine-tuned Stable Diffusion model that generates spectrogram images and converts them to audio, enabling near real-time text-prompted music generation and looping.
diffusionreal-timespectrogramtext-to-music -
SymphonyNet Discovered by AI
2022
A permutation-invariant language model designed for symbolic symphony music generation, utilizing a Multi-track Multi-instrument Repeatable (MMR) representation and 3D positional embeddings to generate complex orchestral pieces.
midiorchestralsymphony -
2021
Microsoft's research project on AI music covering understanding and generation, bundling models for melody, accompaniment, lyric-to-melody, and structured symbolic composition.
microsoftresearchsymbolictransformer -
Symbolic Music Generation with Diffusion Models
2021
Applies diffusion models to symbolic music by training over a continuous latent of note sequences, achieving strong unconditional generation and post-hoc conditional infilling.
diffusioninfillingsymbolic -
MMM (Multi-Track Music Machine)
2020
A Transformer-based system for conditional multi-track symbolic music generation, offering track- and bar-level control while handling long-term dependencies.
multi-tracksymbolictransformer -
2020
An open-source Python toolkit for symbolic music generation, providing dataset management, data preprocessing, representations, and model evaluation utilities.
datasetssymbolictoolkit -
2018
A deep learning model for style-specific polyphonic music generation, able to compose in the style of particular composers or eras with controllable dynamics.
lstmpolyphonicstylesymbolic -
2017
A convolutional GAN for symbolic melody generation in the MIDI domain, with conditional mechanisms and an extension to multiple MIDI channels.
cnnganmidisymbolic -
2017
A multi-track sequential GAN for symbolic music generation that models bass, drums, and other tracks jointly, capturing temporal dynamics and inter-track dependencies.
ganmulti-tracksymbolic -
2016
Google's open research project for music and art generation with machine learning, providing models (MelodyRNN, MusicVAE, and more) and tools for symbolic and audio music creation.
googlernnsymbolictoolkit -
2015
An early algorithmic music generation project that uses LSTM/GRU recurrent networks to generate raw audio waveforms sample by sample.
earlyrnnwaveform -
A lightweight Python tool that generates ambient background music from text prompts and runs without a GPU, suited to simple BGM prototyping.
ambientcpulightweighttext-to-music
Proprietary / Industry Tools¶
-
Udio Discovered by AI
2024
A state-of-the-art proprietary AI music generation platform that allows users to generate high-quality full-length songs, instrumentals, and vocals from text prompts. It supports advanced features like audio inpainting, track extension, and style remixing.
bgmmusic-generationproprietarytext-to-music -
Suno AI Discovered by AI
2023
Suno is a leading generative AI music platform that allows users to generate full-length, high-quality songs with vocals, lyrics, and instrumentation from simple text prompts. It features granular creation controls, stem extraction, and a web-based generative audio workstation.
bgmmusic-generationsoundtrackvocals -
Soundraw Discovered by AI
2020
A royalty-free AI music generator that allows creators to customize beats, instrumentation, and song structure bar-by-bar for video and game production.
bgmmusic-generationsoundtrack -
AIVA Discovered by AI
An AI music generation assistant specializing in classical, symphonic, and cinematic music composition. It allows users to generate original soundtracks in over 250 styles, upload MIDI/audio influences, and export tracks in MIDI and audio formats for games, videos, and podcasts.
bgmcinematicmusic-generationproprietary -
A commercial AI music service with an API that generates royalty-free, often adaptive/streaming background music from text prompts and tags for apps, video, and games.
adaptiveapicommercialroyalty-free -
A commercial AI music generator that produces royalty-free tracks from genre templates, aimed at creators who need quick background music for content and games.
commercialroyalty-freetemplates -
Stability AI's commercial platform for generating music and audio with a latent diffusion model conditioned on text and timing metadata, offering fast inference and length control.
apicommercialdiffusiontiming-conditioned -
A commercial music generation service producing full songs with vocals from text prompts, used for rapid soundtrack and BGM prototyping.
apicommercialmusic-generation
