GitHub - affige/genmusic_demo_list: A List of Demo Websites for Automatic Music Generation Research GitHub - affige/genmusic_demo_list：自动音乐生成研究演示网站列表

TL;DR

• A comprehensive curated repository cataloging 100+ demo websites for automatic music generation research, spanning multiple architectural paradigms
• Models are systematically organized by approach: diffusion, transformer, flow matching, DiT, Mamba, GAN, VQVAE, and hybrid architectures
• The list covers diverse subdomains including music generation, singing voice synthesis, MIDI generation, and general audio synthesis
• References include recent 2026 publications alongside foundational works from 2020-2024, reflecting rapid evolution in the field

Why It Matters

This repository serves as an essential landscape map for researchers and practitioners navigating the crowded and fast-moving field of AI music generation. By consolidating demos, papers, and architectural classifications in one place, it enables quick comparison of approaches and identification of emerging trends, saving significant literature review effort.

Technical Details

• Architectural diversity: The list spans diffusion-based models (MusicLDM, Stable Audio, DITTO), transformer-based models (MusicGen, MusicLM, YuE), flow matching approaches (VersBand, MusicFlow), DiT architectures (FullDiT, EzAudio), and hybrid systems combining multiple paradigms (SongBloom: transformer+diffusion, MelodyFlow: transformer+diffusion)
• Subdomain coverage: Organized across music generation (full compositions), singing voice synthesis (TechSinger, FM-Singer, Everyone-Can-Sing), MIDI generation (MIDI-LLM, Text2midi), and general audio/sound synthesis (AudioLDM, Make-An-Audio)
• Temporal scope: References range from 2020 (JukeNox, dadabots) through 2026 (FullDiT, WanSong, LeVo 2, Agogic), capturing both foundational and cutting-edge work
• Notable systems: Includes well-known open-source models like MusicGen (Meta), Stable Audio (Stability AI), MusicLM (Google), and newer entries like FullDiT, LeVo 2, and Khala from 2025-2026

Industry Insight

• The dominance of diffusion and transformer architectures in recent 2025-2026 publications suggests these paradigms have become the standard, with flow matching emerging as a competitive alternative offering faster inference
• The proliferation of singing voice synthesis models (15+ entries) indicates vocal generation is a high-priority research area with significant commercial interest, likely driven by demand for AI-generated music content
• Practitioners should monitor the shift toward flow matching and DiT architectures for potential efficiency gains, and consider the hybrid transformer+diffusion approach as a promising direction for balancing quality and controllability

摘要

一个全面的精选仓库，收录了100多个自动音乐生成研究的演示网站，涵盖多种架构范式
模型按方法系统组织：扩散模型、Transformer、流匹配、DiT、Mamba、GAN、VQVAE及混合架构
列表涵盖多样化的子领域，包括音乐生成、人声合成、MIDI生成和通用音频合成
参考文献包括2026年的最新出版物以及2020-2024年的基础性工作，反映了该领域的快速演进

深度分析

核心要点

• 一个全面的精选仓库，收录了100多个自动音乐生成研究的演示网站，涵盖多种架构范式
• 模型按方法系统组织：扩散模型、Transformer、流匹配、DiT、Mamba、GAN、VQVAE及混合架构
• 列表涵盖多样化的子领域，包括音乐生成、人声合成、MIDI生成和通用音频合成
• 参考文献包括2026年的最新出版物以及2020-2024年的基础性工作，反映了该领域的快速演进

重要意义

该仓库为研究人员和实践者提供了AI音乐生成这一拥挤且快速发展的领域的重要全景地图。通过整合演示、论文和架构分类，它使快速比较不同方法和识别新兴趋势成为可能，节省了大量的文献综述工作。

技术细节

• 架构多样性：列表涵盖基于扩散的模型（MusicLDM、Stable Audio、DITTO）、基于Transformer的模型（MusicGen、MusicLM、YuE）、流匹配方法（VersBand、MusicFlow）、DiT架构（FullDiT、EzAudio）以及结合多种范式的混合系统（SongBloom：Transformer+扩散，MelodyFlow：Transformer+扩散）
• 子领域覆盖：按音乐生成（完整作品）、人声合成（TechSinger、FM-Singer、Everyone-Can-Sing）、MIDI生成（MIDI-LLM、Text2midi）和通用音频/声音合成（AudioLDM、Make-An-Audio）组织
• 时间范围：参考文献从2020年（JukeNox、dadabots）到2026年（FullDiT、WanSong、LeVo 2、Agogic），涵盖了基础性和前沿性工作
• 代表性系统：包括知名的开源模型如MusicGen（Meta）、Stable Audio（Stability AI）、MusicLM（Google），以及较新的条目如FullDiT、LeVo 2和Khala fro