The era of AI's grand entrance into the Chinese music scene has finally arrived, smoothly sweeping in with its silky touch!
On April 2nd, the curtain rose on the testing phase of China's first and only publicly available AI music generation model, Kunlun Wanwei's "TianGong SkyMusic," sparking a torrent of rave reviews!
With its seamless compositions, catchy melodies, and remarkably realistic, even emotionally charged vocals, the works generated by "TianGong SkyMusic" are so smooth that they make you wonder if they originated from an AI model. With "TianGong SkyMusic," even if you're a novice, you can effortlessly transform your inspiration into captivating songs.
So, how can you create your own exclusive song using "TianGong SkyMusic"? Follow our rhythm and let's dive into the world of music production together!
The image was generated by TianGong AI.
To access "TianGong SkyMusic," first, we need to obtain the download link through the official App Store or the Kunlun Wanwei Group official account, then install the "TianGong" app.
After installation, locate the "Music" entrance at the top of the app interface, and simply tap to enter "天工SkyMusic" for music creation. Moreover, on the homepage, you can explore great songs created by others, or find more music inspiration in the "Inspiration Guide" section.
It's important to note that currently, "天工SkyMusic" is only available for internal testing. However, fret not, as it will be officially launched on April 17th and open to the public for free. At that time, whether you're a professional music creator or just a music enthusiast, you can experience the joy of music production firsthand by downloading the "天工AI" app.
The first step in songwriting starts with crafting the lyrics. After preparing the simple "tools" mentioned above, we can now dive into songwriting using "天工SkyMusic". Click on the most prominent "Start Writing Song" button on the interface. Upon redirection, you'll find input fields for the song title and lyrics. If you already have a song title and lyrics ready, simply copy and paste them into the respective fields. Remember, when writing lyrics, ensure they have balanced structure and rhyme!
But what if you don't have lyrics, just a simple "inspiration"? Don't worry! You can simply fill in your thoughts and let the AI do the rest by clicking on "AI Lyric Writing" in the bottom right corner.
For example, if you're eating fried chicken in Times Square right now but feeling a bit sad inside, let's see how Tian Gong AI can help you express that in song lyrics:
Of course, we can also add structural labels in the lyrics to instruct the AI about the desired song structure. Currently supported labels include [Intro], [Outro], [Verse], [Chorus], [Bridge], and [Interlude]. When providing this information, be sure to enclose it in brackets!
Choosing a reference song to determine the style
Crafting a complete song involves four key steps of "lyrics and melody integration". Once the lyrics are ready, we'll rely on AI to assist us with the next three steps.
Before "SkyMusic" gets into full swing, we need to select a reference track for the song we're creating to establish its mood.
After finishing the lyrics for our new song "Feeling Blue in People's Square While Eating Fried Chicken," we'll navigate to the song creation interface and hit the "Choose a Song as Reference" button at the bottom. For this song, we can select "Rainy Day" by Nan Quan Mama as our reference track. Then, with a click on "Generate Song," we'll get three original compositions. We can pick the best one and download it to share.
When selecting a reference song, we can tailor our choice to match our different moods, feelings, and preferences. On the reference song selection page, simply tap on the filter icon next to "All" to sort songs by different emotions for inspiration.
And just like that, your very own song is ready to go!
Download, Share, and Edit Again
Once your creation is complete, you can find the song in "My Works". The song card allows for direct playback and provides options for easy downloading and sharing.
At the top right corner of each song title, there's an edit button. Clicking on it takes you back to the initial lyric composition and "Select Reference Music" page, allowing for further editing of the song.
One More Thing: The Inspiration Generator
Aside from its songwriting capabilities mentioned above, Tiangong SkyMusic may hold even greater significance in the realm of discovering and seeking musical inspiration. When adding reference songs, users not only have the option to include existing tracks but can also upload their own.
At this point, if you happen to have a guitar, piano, or even just a vocal recording exceeding ten seconds, Tiangong SkyMusic swiftly generates three new melodies based on your recent style. What's truly remarkable is its near-perfect ability to replicate the sound you just recorded.
Although Tiangong SkyMusic is currently in its early stages of internal testing, it effectively ignites the passion of numerous music enthusiasts for creative composition, enabling more users to explore the joy of music creation.
In the future, it will further integrate into the music ecosystem and continuously refine itself, offering users more innovative and professional features. Tiangong SkyMusic will continue to lower the barriers to music creation for every user and assist professional musicians in expanding the boundaries of music composition. Together, we will promote the fusion and innovation of musical styles, empowering everyone to create music that fully expresses their personality and emotions.
About Tiangong SkyMusic
Tiangong SkyMusic utilizes the Sora model architecture in the field of music audio, with Large-scale Transformer responsible for composing music, learning the contextual dependencies of Music Patches, and ensuring musical controllability. The Diffusion Transformer handles vocals, using LDM to restore Music Patches into high-quality audio, enabling Tiangong SkyMusic to generate 80-second, 44100Hz sample-rate stereo songs. This model architecture excels in processing video, audio, and music domains. The Kunlun Wavenet team will continue to iteratively add new capabilities, enabling the model to possess multimodal emotional understanding and expression.
