Get started
Step by step
- 1
Describe the scene
First line: who speaks, in what tone and where. For example: "A crisp, calm news anchor opens the bulletin".
- 2
One line per cue
Put each line on a new row that starts with [start s:end s], such as [0s:4s]. Use the / command and pick the timestamp to insert it.
- 3
Size each cue to its line
Give every cue enough time for its words: about two to three words per second. Keep the cues in order and without overlaps.
- 4
Add a second voice if needed
Start the line with "Speaker A:" or "Speaker B:" after the cue for a dialogue.
- 5
Pick the language and generate
Choose one of 24 languages in the menu, set tone, speed and volume if you like, and generate. If the timing is off, move the cues and generate again.
Example prompts
Copy them into Audio Generation and swap in your own lines.
Product launch
A confident promo voice introduces a new product. [0s:3s] The new charger is here. [3s:7s] One cable for your phone, your watch and your earbuds. [7s:10s] Available from today.
Two-voice dialogue
Two friends plan a trip, relaxed and natural. [0s:4s] Speaker A: How about the coast this weekend? [4s:8s] Speaker B: Only if we leave early and stop for breakfast. [8s:11s] Speaker A: Deal.
Guided breath
A calm guide leads a breathing exercise with long pauses. [0s:6s] Breathe in slowly through the nose. [6s:12s] And let it go, shoulders dropping.
Podcast intro
A relaxed, warm podcast host opens the weekly show, conversational and unhurried. [0s:4s] Hey, welcome back to the show. [4s:9s] This week we're talking about the one habit that changed my mornings. [9s:12s] Let's get into it.
Recipe steps
A friendly cooking host explains quick steps, clear and upbeat. [0s:5s] Start by heating a little oil in the pan. [5s:10s] Add the garlic and stir until it turns golden. [10s:15s] Now in go the vegetables, and a pinch of salt.
Trailer build
A deep, slow trailer voice builds tension with long pauses between lines. [0s:4s] They said the signal was gone. [5s:9s] They were wrong. [10s:13s] This winter, listen closely.
The [2s:5s] syntax
A cue gives the second a line starts and ends: [2s:5s] means from second 2 to second 5. The studio's own Timing templates use the same format.
Cues are timing hints the model reads, not exact edits: the result can drift a little. For dependable timing, leave some room in each cue and generate two takes.
Write the scene line first
The first line of the prompt has no cue. It tells the model who is speaking, how they sound and where they are, and every timed line after it is read in that voice. A good scene line names a role, two or three qualities of the voice and the energy: "A warm, unhurried documentary narrator" or "A high-energy sports commentator, voice climbing".
Describe the voice in plain words. Age, pace, warmth and energy work better than technical terms, and one clear description beats a long list of adjectives that pull in different directions.
- Role: narrator, news anchor, podcast host, coach, commentator, storyteller.
- Voice: warm, deep, bright, husky, soft, crisp.
- Pace: unhurried, measured, quick, building.
- Energy: calm, confident, playful, tense, excited.
How many words fit in a cue
A natural read is about two to three words per second. Size each cue to its line, then leave a little room at the end so the words never have to be squeezed in. If a line feels rushed when you play it back, make the cue longer or cut words; the cue cannot make speech faster than sounds natural.
- A 3-second cue fits about 8 words: a short hook or tagline.
- A 5-second cue fits about 13 words: one full sentence.
- An 8-second cue fits about 20 words: a longer sentence or two short ones.
- For a pause, leave a gap between one cue's end and the next cue's start, such as [4s:8s] followed by [9s:12s].
Steering the delivery
The scene line sets the overall voice; the lines themselves carry the moment-to-moment feeling. Punctuation helps: a full stop gives a firm end, a question mark lifts the line, and short sentences sound more urgent than long ones. You can also add a few words of direction to the scene line, such as "slowing down with each line" or "intensity climbing".
Audio Generation also has Tone, speech rate and Volume sliders next to the prompt. Use them for overall adjustments after the prompt sounds right, rather than trying to describe the speed in words.
Start from a Timing template
The Audio Generation studio has ready-made Timing templates that already use cues: a countdown promo, a news bulletin, a match commentary, a guided breathing exercise, a bedtime story, a movie trailer and recipe steps. Open one, swap in your own lines and keep the timing structure. It is the quickest way to see how cue length and line length relate.
Text templates work without any cues, for when you only need the voice and the feel and the exact timing does not matter.
Other languages
Pick the output language from the menu: Audio Generation speaks 24 languages. Write the spoken lines in that language and keep the cues exactly as they are; the [start s:end s] format is the same in every language. Some languages take more syllables to say the same thing, so check the playback and lengthen cues where lines feel tight.
Using the audio in a video
Download the result and line it up with your video in the free video editor. Because the cues already match your plan, a voice line written for [3s:7s] lands on the shot you planned for those seconds. For a video made in the KreatorFlow studio, the voiceover guide shows how to plan the script around the clip lengths.
Troubleshooting
Most timing problems come from cues that are too tight or lines that are too long.
- A line sounds rushed: lengthen its cue or cut words.
- A line starts late or runs over: leave more room in the cues around it and generate a second take.
- The voice changes partway through: make the scene line more specific and keep the lines in the same register.
- Two speakers sound alike: describe each voice in the scene line, for example "Speaker A is a deep, slow older man; Speaker B is a quick, bright young woman".
- A cue seems ignored: check it uses the exact [2s:5s] format, with the s after both numbers, a colon between them and the cue at the start of its own line.
Common mistakes
- Overlapping cues, such as [0s:5s] followed by [4s:8s]. Keep them in order with no overlaps.
- Packing a whole paragraph into one short cue.
- Leaving out the scene line, so the model has to guess the voice.
- Mixing two languages in one prompt.
- Expecting frame-exact timing. Cues guide the timing; they do not cut the audio.
What @Audio 1 does
The / command also inserts @Audio 1, a tag the studio's Reference templates use for "the voice in reference clip 1". Audio Generation in KreatorFlow sends your text prompt only and has no clip upload, so describe the voice you want in words as well, for example "a warm, measured narrator".
The video studio is different: audio clips you attach there are numbered Audio 1, Audio 2 and so on up to 3, and the prompt can refer to them.
Frequently asked questions
Are the timestamps exact?
No. They are timing hints the model follows as closely as it can. Leave some room and generate two takes when timing matters.
Can I use a reference clip for the voice?
Audio Generation sends text only and takes no clip upload. Describe the voice in words. The video studio does take audio references.
What does it cost?
1 credit per started 10 seconds of audio (minimum 2 credits). The studio shows an estimate from your prompt before you generate, and the final charge follows the length of the audio it makes.
Do I have to use timestamps?
No. Cues are optional. A prompt without them still works; the model then decides the timing itself. Use cues when lines must land at set moments, such as over specific shots of a video.
How many words fit in a cue?
About two to three words per second: roughly 13 words in a 5-second cue. Leave a little room at the end of each cue.
How do I add a pause?
Leave a gap between cues, for example [0s:4s] for one line and [5s:9s] for the next. The second without a cue becomes a pause.
Can cues overlap?
Keep them in order and without overlaps. For two people speaking, give each line its own cue and start it with Speaker A or Speaker B.
Which languages can I use?
Audio Generation speaks 24 languages. Pick one in the menu and write the lines in that language; the cue format stays the same.
