The anatomy of a good Seed Audio prompt
Seed Audio 1.0 renders a whole sound scene, so a strong prompt reads like a short brief for a sound designer. Cover four things: who is speaking, what they say, how they say it, and what surrounds them.
Put spoken lines in double quotes. Describe the voice before the line ("a calm, older male narrator"), and the setting after it ("rain on the window, a clock ticking"). Keep the whole prompt under about 900 characters for the most reliable results.
Example prompt
A calm, older male narrator speaks slowly: "Some places remember everything." Rain on a window, a grandfather clock ticking, and a fire crackling in the room.
Writing dialogue with several speakers
Introduce each speaker right before their line, in the order they speak. Short descriptions such as "a nervous teenager" or "a cheerful barista" help Seed Audio keep the voices distinct.
Add reactions in plain words: laughs, sighs, interrupts, whispers. They shape the performance more than any setting.
Example prompt
In a busy café, a cheerful barista asks: "The usual?" A tired regular laughs and answers: "Make it a double today." The espresso machine hisses and cups clink in the background.
Example prompt
Two astronauts on a quiet station. A young woman whispers: "Did you hear that?" An older man pauses, then answers: "Stay here. I will check the airlock." A low hum and a distant metallic knock.
Sound effects and ambience
Name the source, the material and the space. "A door" is vague; "a heavy wooden door creaking open in a stone hallway" gives Seed Audio something to render.
Use time words to build a sequence: "then", "after a short pause", "in the distance", "getting closer". For ambience, say "constant" or "steady" so the bed stays even.
Example prompt
Glass shatters on a tile floor, then a cat hisses and runs off, and a door slams somewhere upstairs.
Example prompt
Constant heavy rain on a tin roof, occasional distant thunder, and wind pushing against a loose shutter.
Voice cloning with @Audio1, @Audio2 and @Audio3
Upload up to three reference clips in the studio. Each one becomes a tag in order: the first clip is @Audio1, the second @Audio2, and so on. Use the tag where you would name the speaker.
Clean clips of 5 to 30 seconds with one speaker work best. Only use voices you own or have permission to use.
Example prompt
@Audio1 says warmly and clearly: "Thanks for watching. See you next week."
Example prompt
@Audio1 asks with excitement: "Did you see the results?" @Audio2 replies, unimpressed: "I saw them. I just do not believe them."
Image to audio prompts
With a reference image, the picture already sets the scene. A short prompt such as "the natural sound of this scene" is enough. Add direction when you want something specific: "add a documentary narrator describing the view".
Seed Audio 1.0 takes either reference audio or one image per take, not both.
Example prompt
The natural sound of this scene, detailed and immersive, with gentle wind.
Example prompt
A calm documentary narrator describes this place in two sentences over the natural ambience of the scene.
Common mistakes to avoid
Too many speakers in one take: two to four is the sweet spot. Very long scripts: split them into scenes. Conflicting directions ("whispering loudly"): pick one. Asking for a real person's voice by name: not allowed, and it will not work.
If a take is close but not right, change one thing at a time: the voice description, the pacing or the background.
Ready to try? Open the Seed Audio studio, or go straight to dialogue, sound effects or voice clone.