Stable Audio 2.5 Prompt Guide
With the release of our latest audio model Stable Audio 2.5, we’re sharing an updated roundup of our best practices for prompting.
Stable Audio 2.5 is the first audio model designed for enterprise-grade sound production. That means the model introduces capabilities like improved musical structure, faster inference at less than 2 seconds on a GPU, and support for audio inpainting.
The new model is ideal for generating sophisticated, fully developed songs and iterating quickly to create the right sound for every production. With effective prompting techniques, you can get the most out of Stable Audio 2.5’s capabilities for use cases like advertising, game soundtracks, and short-form video.
Read on for our latest tips, and try these techniques out with Stable Audio 2.5 now:
Using Musical Vocabulary
Consider your use case to ensure creative alignment
Before building your prompt, think about what exactly you want to create. Are you producing audio for an ad campaign and your brand has a certain mood you’re going for? Are you scoring the final boss battle in your game and you need the tempo to be fast and exciting?
Each use case will have different musical requirements, from mood to tempo. Start with the end result in mind, and you can build your prompt to get there.
Begin with the three basic building blocks
Start with the essentials of music. Stable Audio 2.5 is designed for generating up to three-minute tracks, so your prompt should include everything you need for a full composition.
The below three building blocks are the simplest – but most important – context to give the model. If you don’t specify these things in the prompt, you’ll get more random results.
| Building Block | Examples |
|---|---|
| Genre/Style: Define the category for your track. Pro tip: Make sure to clarify your genre references with more specific classifications or subgenres. |
Genres:
|
| Tempo (BPM): Define the speed to match your intended use case and energy level. Pro tip: If you know the general ranges for your genre of music you want to produce, the model can adjust to the tempo you want the song to be. |
General tempo:
|
| Overall Mood: Establish the emotional foundation of your track. Pro tip: Use clear, descriptive emotional terms that go beyond simple words like “happy” and “sad.” |
|
Example prompt: "A relaxing Bossa Nova instrumental, perfect for an elevator. The mood is smooth, patient, and relaxing, 115 BPM."
Layer in more musical elements to build depth
Specify the important sound sources in your track, then go beyond that with more detail on texture and production characteristics.
A common challenge is generating a track that sounds too generic or flat. Add more specific musical elements to get a richer, more dynamic composition.
| Musical Element | Examples |
|---|---|
| Primary instruments: These instruments carry the melody or main musical theme. They’re the focal point that listeners typically follow, and often play the hook or most memorable parts. |
|
| Supporting instruments or elements: Add supporting instruments to complement the main melody. They’re usually mixed in the background or at moderate levels. |
|
| Rhythm components: Define the percussive elements that drive your track. |
|
| Texture elements: Add in layers that create sonic depth and atmosphere. |
|
| Production characteristics: These elements describe how the track is recorded, mixed, and produced. Pro tip: You don’t always want a polished studio sound. For example, a lo-fi hip hop track calls for a relaxing, nostalgic feel that requires a vintage recording quality. Play around with different production styles to get the feel you’re going for. |
Recording quality descriptors:
|
Example prompt: “An exciting breakbeat instrumental perfect for fast-paced video games, featuring funky electric guitar chords, steady break drums, smooth electric piano, and supporting bass. The mood is fresh, modern, and adventurous, 105 BPM.”
To recap, your prompt should have the following musical elements for best results:
Specific genre classification
Sophisticated mood descriptors
Appropriate BPM for the genre
Detailed instruments, including primary and secondary instruments, rhythm, and texture
Additional production characteristic
Prompt Construction Techniques
Now that you have the elements of your track, let’s look at how to structure your prompt.
Order matters in your prompt
Start with the core style and genre, then the key musical elements. The order of information you give your prompt should look like:
Core style/genre
Key instruments and musical elements
Mood and emotion
Specific details
Additional instructions
Good structure:“Cinematic outlaw country instrumental perfect for a long drive featuring a blues pedal steel guitar, a rustic mandolin, a fiddle playing call and response, a tape-driven rattly drum-kit, an autoharp, and a soaring accordion solo. The mood has southern soul, is raw, emotional, expansive, and full of the blues.”
Less effective: "Make a song that’s happy and has guitar, fiddle and drums"
To iteratively improve, start with the basic prompt and evaluate the result. Identify any missing elements, add in more specific instructions, and retry it as needed.
Advanced tips for refinement
1. Use titles to give direction
Adding a title to your prompt can provide a clear and immediate context for the mood and direction of your song. For example you can add "titled 'Rebellion'" or "titled 'On & On'" to give the model an interesting direction for the mood of the song. This simple addition can improve the musicality of your results.
2. Incorporate geographic context
If your desired sound is associated with a real-world location, include it in your prompt. This helps the model understand the cultural and historical influences of the music. For instance, if you're creating a House track, specifying "Chicago," "New York," "Detroit," or "Ibiza" will steer the model toward the distinct styles and characteristics of that region.
3. Define a use case
Clarify the intended use or scenario for your song. Adding a phrase like "perfect for opening credits" or "perfect for a long drive" gives the model a functional purpose for the music. This helps the model create a track that not only sounds good but also fits a specific narrative or mood.
Example prompt:“A luxurious indietronica instrumental perfect for a perfume advertisement featuring clean guitars, synthesizers, and a slow-tempo drum machine pattern.”
4. Use original and precise language
To get the most out of your prompts, use a variety of descriptive words. Avoid repeating the same adjectives. Instead, use a range of vocabulary that cohesively fits a musical context. For example, use phrases like "smooth electric piano" or "textural percussion" to provide rich, detailed information. This maximizes the context you provide and ensures every word counts.
It also helps to reference eras to describe musical and production elements, such as “80s gated reverb” or “90s grunge distortion.”
Example prompt: “90s garage rock instrumental with a grunge influence featuring poppy distorted guitars, frantic and energetic drums and tube-distorted bass guitar.”
5. Try inpainting
Stable Audio 2.5 supports audio inpainting. This is useful for when you want to preserve an original part of the song, but generate variations of the track. For instance, advertising professionals might generate different versions of a song to match different creative in an advertising campaign.
Inpainting allows you to:
Upload existing audio (Note: Our Terms of Service require that uploads be free of copyrighted material, and we use advanced content recognition to maintain compliance and prevent infringement)
Select starting points for generation
Let the model complete the composition using context
If you’re aiming for a specific section to be inpainted, tell it the start and end time. The more context you give before it, the more it takes that as an example. The prompt structure itself is the same structure you would use normally.
Example prompt:“A soulful tech-house song with beautiful and melancholic undertones and epic breakdowns, noisy granular synthesizer layers, and epic reverberating and echoing leads.”
Getting Started
Ready to put your new prompt knowledge to the test with our latest model? You can try Stable Audio 2.5 now at StableAudio.com.
Stable Audio 2.5 is also now available through the Stability AI API, as well as through partner platforms including fal, Replicate, and ComfyUI.
For enterprises interested in deploying our audio models on their own infrastructure, please contact us to discuss our Enterprise Licensing, with implementation support, customization options and professional services available. You can also visit Stability AI Solutions to learn more about customizing audio models and workflows for specific use cases.
Our flexible licensing options allow you to deploy our models in your environment with advanced customization and full control over your data.
