US startup Suno specializes in AI audio generation from text. Its latest audio model generates some impressive songs.
In early September, Suno unveiled its latest text-to-song model, Chirp v1, which can generate music, including vocals, based on style and lyrics. The biggest improvement is that v1 can convert genres such as rock, pop, K-pop, and descriptions such as melodic or fast into music.
Lyrics can now be split into parts using commands like [verse] and [chorus] to give the generated songs more structure. Lyrics can either be typed in or generated directly in Chirp's interface using ChatGPT.
The startup notes that prompts with a specific artist reference are not supported, probably to avoid copyright discussions. This was the case when an AI-generated song featuring the voices of Drake and the Weeknd went viral. The song was subsequently blocked by Universal Music Group.
AI song generation on Discord
Chirp generation is fully integrated into Discord, similar to Midjourney. For each Chirp prompt, the model generates two variations, usually between 20 and 40 seconds in length. If you like a variation, you can generate more by clicking "Continue", which can add up to 30 seconds to a generation while continuing the style of the previous generation.
If you want to get inspired or get to know the Chirp's potential, just browse the Discord servers - more than 40,000 users make sure there is a constant supply of songs. You're bound to find a gem or two, like this politically motivated love song. Will we hear more of it soon?
Or you can take existing lyrics of well-known songs and let the AI generate them in a new musical style.
The complexity of the songs combined with the quality of the generated voices is sometimes impressive. There are no chart-toppers yet, but at the current pace of generative AI development, this could change quickly. Suno has posted some particularly good-sounding demos of the new model on a website.
Suno supports more than 50 languages, with English and rock music performing best in my tests. The style also seems to be influenced by the content or structure of the lyrics. Based on the lyrics, a matching background image is generated for each audio clip.
Free chirps on Discord
Suno offers 250 free credits per month on Discord, which is equivalent to 25 chirps. Chirps can be generated either on the public server or in the Discord DMs. A Pro plan offers 1000 credits / up to 100 chirps per month and costs $10 per month. You can purchase additional generations. More information about the payment models can be found here.
Last spring, Suno introduced Bark, a text-to-speech and audio model that is freely available on Github under the MIT license for commercial use. Bark is also available via Discord.