There are actually quite a few AI music generation tools, but the one that supports Chinese lyrics well is currently none other than Suno AI. Especially with the recent release of version 3, many media outlets are calling it the “ChatGPT” moment for AI music. While that might be a bit exaggerated, there’s some truth to it. I tried version 1 on Discord last year, which could only generate 30 seconds at a time, and there were significant issues with pronunciation and occasional loud electrical sounds. You can listen for yourself.
1. Access and Payment
Website: https://www.suno.ai/
Currently, there are three login methods. Discord and Google require a VPN, but Microsoft allows login with a domestic account, meaning a VPN is not necessary.
The free version currently offers 50 points per day, with each song consuming 5 points, allowing for 5 draws per day. The monthly plans are $10 and $30, which provide 2500 and 10000 points respectively, and support for domestic credit card payments. Paid subscriptions come with commercial rights.
2. Basic Usage Methods
1. If you don’t have lyrics, you can directly express your ideas, and Suno will generate lyrics and a musical style based on your thoughts.
2. If you have lyrics, open the Custom Mode and fill in your lyrics and desired style.
3. Turn on the Instrumental option to generate pure music. However, sometimes it may not be completely pure, so it’s best to listen to it before using it as background music~
3. Generating Music Longer Than Two Minutes
Currently, the first generation of songs is limited to under two minutes, and longer lyrics may not be fully sung. To continue generating, you can click the … symbol on your favorite song and select Continue From This Song, then fill in the lyrics you want to continue with (although sometimes the complete lyrics may continue, using segments is more controllable).
Additionally, after the recent update, you can sing from a specified position, greatly enhancing the controllability of completing a perfect song. This way, if the beginning is fine but the later part is unsatisfactory, you can draw again from a specified time.
In principle, songs can continue indefinitely. I haven’t tested the longest limit yet, but I’ve extended it 5 times without issues, with each extension lasting up to 1 minute. Generating a long song of 5-6 minutes is completely feasible.
After completing a song, click Get Whole Song at the final stage, wait a moment, and all your selected segments will be seamlessly connected. However, this requires an additional cost of 5 points, and the processing effect is quite perfect. I previously didn’t realize this feature and foolishly tried to splice them manually~~
4. Other Tips
1. You can add the following prompts in the lyrics to control the song’s rhythm. Although it doesn’t always adhere to the prompts, it’s better than not using them at all.
[Intro] [Verse] [Chorus] [Bridge] [Rap] [Outro]
Additionally, you can add repeat specified words in parentheses at the end of a sentence to achieve a duet feel, e.g., [Verse 1] The traces of time drift quietly with the wind (quietly drift) The direction of dreams, a faint smile (faint smile) Sweat must be shed to exchange for glory (glory)
2. Sometimes you may find that certain words just can’t be sung, indicating that Suno may not recognize them. You can use homophones to replace them.
3. You can write your requests in Chinese; while English might be more accurate, the drawing probability seems to be similar. The underlying model likely uses the GPT-4 model to understand intent.
4. In Custom Mode, you can try filling in more intentions for the style, such as: female fronted vocal, melodic death metal, metal, death metal, swedish, swedish metal, instrumental, alternative, 120 BPM. These are the main prompts for trying out music models.
5. You can also explore excellent works in the Explore section to learn how others write prompts. Additionally, you can directly use the Remix feature to copy others’ lyrics and prompts to try your own draws.
6. Although many styles and instruments are currently supported, there are still unsupported ones. For example, I wanted to generate background music primarily featuring the flute and xiao, but the probability of drawing that is quite low. I look forward to its continued evolution, as it has only been evolving for a little over six months.
7. Additionally, I asked GPT to summarize some information about music genres, styles, types, and instruments. Interested students can click read the original for reference.
Interestingly, songs I participated in generating sound so good that I want to listen to them on repeat, while songs shared by others feel “just this? It’s quite average.” This might be a case of the “effort justification effect.” However, this may not be a personalized trend, as in the future, the songs I listen to most on my playlist might just be the music I put effort into generating myself!