AI Music Generation with AI Agents
Hand an AI agent your lyrics or a one-line style description and let it call music-generation tools to produce a full song with vocals — then compare models and pick the best fit for short video, games, or marketing.

Tool names in this guide are for reference only and may have changed or been renamed. Please use the latest tools in the tool marketplace when implementing.
Key takeaways
- Turn lyrics or a style description into a full song with vocals
- Switch between multiple music models on one platform and compare
- One chain from lyrics to finished track for video, games, marketing
Tools used
Why making a song you can actually use is hard
Short videos need custom BGM, games need theme tracks, campaigns need a melody that carries the brand. The old options were either commissioning a writer and singer — slow and expensive — or digging through stock libraries, where licensing limits you and everyone ends up with the same clip. The real pain isn't producing sound; it's doing three things in one flow: getting a full song with vocals, specifying a style, and comparing a few versions to pick the right one.
Connecting music-generation tools to an AI agent turns it into one sentence: you give lyrics or a style description, the agent picks a model, generates the track, lets you compare, and you finalize — no shuttling between sites, no hand-tuning dozens of parameters.
The workflow in five steps
- Give the description: hand the agent lyrics or a style brief in natural language.
- Generate: produce candidate songs with a music-generation tool.
- Compare models: run the same brief through several models for different styles and vocals.
- Pick the best: choose the version that fits your scene — length, vocals, style match.
- Produce the track: export the final cut and move it into editing or distribution.
How to do each step
Step 1: Give lyrics or a style description
Say what you want: paste full lyrics if you have them, or describe the style, mood, tempo, and use case if you don't. The agent builds the music tool's input from this.
use ae to generate a 30-second upbeat pop song with female vocals about a sunny road trip
Step 2: Generate with a music tool
The agent calls a music-generation tool for candidates. For standard generation use ae_ace_step or ae_cassetteai_music_generator; for a higher-quality finish use ae_lyria3_pro, and for everyday generation ae_lyria3 works too.
use ae to compose it with ae_ace_step, then also try ae_lyria3
Step 3: Compare across models
Run the same lyrics or brief through several models and get different arrangements, vocals, and textures. Music is subjective — comparing side by side finds the right direction faster than blindly tuning one model.
Step 4: Pick the best for your scene
Choose from the candidates: short video cares about length and a catchy opening, games about whether the loop feels natural, marketing about vocals and brand fit. You can also give the agent your criteria and have it rank the options first.
Step 5: Produce the finished track
Export the chosen version and feed it into editing, voiceover, or distribution. Once the flow is stable, the same brief can batch-generate several alternates.
Tools by goal:
| Goal | Tool (example) |
|---|---|
| Standard music generation | ae_ace_step |
| Standard generation (alternate) | ae_cassetteai_music_generator |
| Higher-quality finish | ae_lyria3_pro |
| Everyday generation | ae_lyria3 |
Tuning one model vs comparing many:
| Hammering on one model | Comparing models on one platform |
|---|---|
| Tweak parameters, guess the direction | Run the same brief, hear the difference |
| Switching models means switching platforms | Swap models with one sentence |
| Easy to get stuck in one style | Spread styles and vocals out, then choose |
How to get started
- Sign up — new users get 1,000 free credits.
- Connect in-app from an agent like WorkBuddy or Trae by searching for "AgentEarth", or paste a one-line MCP config into Claude Code, Cursor, or VSCode.
- Paste your lyrics or style description and let the agent generate, compare, and finalize.
FAQ
Can it make a song with vocals? Yes. State in the description that you want vocals, plus gender, language, and style, and the music tool generates a sung version accordingly.
Can I specify a style? Yes. Put genre, mood, tempo, instruments, and reference feel into the description, and have the agent run different models when you want to compare.
Who owns the copyright of a generated song? That falls under the platform's and each model's terms of service; before any commercial use, refer to the platform and the relevant tool's licensing terms.
