Recently we talked about MusicGen, a model published by Meta for generating music with artificial intelligence and text. Its online demo works quite well, but I imagine many users out there want an offline solution. That brings us to Samplab's portal, which offers TextToSample.
The dance of algorithms and models is not focused exclusively on image creation. Slowly, the cannons begin to point toward the territory of music. For better or worse, technology often has a huge influence on it (e.g., Auto-Tune), and artificial intelligence doesn't want to be the exception. All that is needed to take the first step is the right tool, and the people at Meta believe they have a good solution in their MusicGen model. So how can we access it? There are two options: its official online demo, and the TextToSample client from Samplab.
TextToSample, a tool for creating music with AI locally
After installing TextToSample, you need to enter the first prompt so the program downloads the MusicGen model automatically. The official page confirms it's the "small" variant, but you'll still need patience (more than two gigabytes of files). Once that process is complete, you can keep entering prompts, use the previous sound to "continue" with it, or directly load your own sounds via drag and drop.
The main controls in TextToSample are duration and context. The first has a maximum of 20 seconds, and the second helps us define how far back the model should “look” to continue a melody. Regardless of these parameters, the final result is saved in the TextToSample/generated folder inside Documents, in WAV format. As for hardware, TextToSample recommends using a computer, but it used the 16 threads of our processor without hesitation.
TextToSample is free. Download a copy!
Official site and download: Click here
