AI instructs AI to create music with Google Colab

I made music by suggesting what kind of music the chat GPT can produce with text to music. Recently, various AIs derived from text to image have appeared. text to music generates music from three tag information. In order to generate music based on this tag information, we first asked chat GPT to create a base sentence, decided tag information based on that sentence, and generated music. Chat GPT is provided by open ai, and is a next-generation AI that may be used instead of Google in Microsoft’s bing. I’m looking forward to the AI ​​era, and I tried to touch it ahead of time.

I asked AI for a café-like song

I visited “AI’s café-like song" and asked him to give me a concrete English sentence. I asked a question using the open AI API on Slack. Chat GPT can support Japanese and speed up. The article below introduces how to use chat GPT with Slack, so please read it if you like.

Confirmed in Slack to chat GPT.

He specifically taught me slow, light music. This time, we will use this character as an input character string to generate music.

text to music 「mubert」

Text to image is in vogue, but text to music is being created. You can create music from text. With mubert, you can generate music by specifying three tag information.

Please try it.

Using the Mubert API, get_track_by_tags determines 3 tag information from a sentence. The code was published on Google Colab. I copied and pasted the characters that chat GPT output to prompt and executed it.

generate music

I entered the characters output by Chat GPT in Prompt and executed it.

The tags of the cafe-like song were the following three.

  • jazz / funk
  • electro funk
  • sleepy ambient

I generated music from these three tags. duration is 60 seconds.


I made a café-like song concrete with chat GPT, converted it to 3 tag information, and generated non-existent music. As far as actually listening to music, text to music is still in its infancy, but I think it has great potential. By automating MIDI and lyric linkage, linking with NEUTORINO, etc., Vocaloid singing smoothly, etc., I think that if automation is developed, an integrated automatic music generation environment can be created. I’m looking forward to the world of automatic music generation in the future.

Please follow me on Twitter if you don’t mind .