Use authored video prompts when you already know what every scene should show. Write the prompt yourself or ask any AI assistant to prepare it. Paste the result into Studio and choose Generate media. Asteria Engine uses your scene directions without running the video designer.
The normal Design tools remain available for projects where you want help developing the idea.
Before you begin
Select Video in Studio and choose Gemini Omni in Output. This format supports landscape (16:9) and portrait (9:16), with a total running time up to five minutes, subject to your workspace and destination limits. Generation uses your normal video credits.
Your timeline controls duration, orientation and audio for this generation. It does not use a previous design or a locked music track. Choose the video model's own audio or separate text-to-speech (TTS) narration. References and end-card settings still apply; check them before generating. An end card can add time after the authored story; turn it off for an instalment that should end without one.
Copy this 30-second example
Copy the first line and the complete JSON below into the Studio prompt. Do not include Markdown backticks.
ASTERIA VIDEO 1
{
"prompt": "A short documentary-style story about a wooden bowl, from the workshop bench to a customer's hands. Keep the same craftsperson and clothes throughout. No on-screen text.",
"aspectRatio": "16:9",
"audio": true,
"scenes": [
{
"durationSeconds": 10,
"prompt": "The craftsperson turns a wooden bowl on the bench in a bright workshop while saying: Every piece begins with a small idea. The camera moves slowly closer as they speak.",
"transition": "continuity_bridge",
"endState": "The bowl rests on the bench, with the craftsperson's right hand beside it."
},
{
"durationSeconds": 10,
"prompt": "Continue from the same bench. The craftsperson lifts the bowl towards the camera while saying: We shape that idea into something you can use every day. End with the finished bowl in view.",
"transition": "continuity_bridge",
"endState": "The craftsperson holds the finished bowl facing the camera."
},
{
"durationSeconds": 10,
"prompt": "A market stall on a quiet high street, a new setting and shot. The craftsperson hands the same finished bowl to a customer while saying: Every piece finds its home. The customer turns it over in their hands and smiles.",
"transition": "editorial_cut",
"endState": "The customer holds the bowl at the market stall, smiling, with the craftsperson beside the stall."
}
]
}
The first two scenes stay on the same bench, so each uses continuity_bridge to continue the ongoing shot. The third scene deliberately moves to a different place, person and shot, so it uses editorial_cut instead. Use continuity_bridge for an unbroken moment and editorial_cut whenever the story jumps to a new setting, time or shot — mixing them up produces visible, unwanted jumps or stalled pacing.
The format
Use ASTERIA VIDEO 1 on its own first line, followed by the JSON object. The same object is also recognised without the header. Pasted smart quotes and line breaks inside text are accepted, but missing commas, duplicate fields and incomplete strings still need correcting.
Required at the top level: prompt, aspectRatio, audio and scenes. Required fields have no automatic substitutes: a missing or invalid value is explained before generation starts. Optional fields use the defaults below. Omitting a field is different from setting it to null; only narrationVoice accepts null to inherit. Blank narration direction inherits, while blank scene narration deliberately leaves that scene without speech.
With audio: false, the audio source and all narration controls are ignored. With audioSource: "video_model", TTS-only controls are ignored. Unknown field names are rejected in all modes, so a typo cannot silently change your request.
- prompt: shared direction for the whole video: people, setting, style and continuity.
- aspectRatio:
16:9or9:16. - audio:
trueenables audio;falserequests silence and overridesaudioSourceand narration. - audioSource (optional):
video_modeluses the video model's speech and sound, as before.ttsgives speech to a separate narrator while the video model supplies background sounds. If omitted,video_modelis used, so existing scripts continue to work. - backgroundAudio (optional, TTS only):
trueby default. The video model supplies ambience, sound effects and instrumental music described in your scene prompts. Setfalsefor narration without background sound. - narrationDirection (optional, TTS only): natural-language delivery instructions, such as language, accent, tone and pacing. These guide delivery and are not spoken. Voice characteristics depend on the selected TTS provider and voice; exact accents are not guaranteed.
- narrationVoice (optional, TTS only): a voice selection containing
provider,voiceand, for Google, optionallanguageCode. This is the default voice for scenes without a more specific choice. Without it, your Studio/workspace TTS voice applies. - speakers (optional, TTS only): an object of up to 100 named speaker profiles. Each profile may contain
narrationDirectionandnarrationVoice. Names may use any language; each scene must reference the exact name, including case. - scenes: an ordered list of 1–100 scenes. Their durations are added together; the total cannot exceed 300 seconds.
Each scene needs these four fields, plus narration when using TTS:
-
durationSeconds: a whole number from 3 to 10. Allow enough time for speech and action together.
-
prompt: what happens during this scene, including camera movement, speech and sound where wanted.
-
transition:
continuity_bridgeto continue the previous scene, oreditorial_cutwhen a deliberate change of setting or shot is appropriate. -
endState: what should be visible at the end, helping the next scene continue consistently.
-
narration (required for TTS): the exact words to speak during this scene. Use
""for a scene with no narration. Keep voice instructions innarrationDirection, and visual action inprompt. This field is ignored in video-model and silent modes. -
speaker (optional, TTS scene): the exact name of a profile in
speakers. Reuse the same name whenever that narrator returns. -
narrationDirection (optional, TTS scene): replaces the speaker's or video's direction for this scene. Omit it or use
""to inherit. Include all accent and delivery instructions you want to retain when overriding. -
narrationVoice (optional, TTS scene): replaces the voice selection for this scene only. Omit it or use
nullto inherit. This does not change the named speaker's later scenes. -
narrationSpeed (optional, TTS scene): a number from
0.25to2.0, default1. For example,0.9requests slower delivery and1.15requests faster delivery. This is the starting rate; scene fitting may increase it, and some selected or fallback voices have narrower speed limits. -
narrationStartSeconds (optional, TTS scene): seconds after this scene starts before narration begins, default
0. Use1to leave a one-second opening sound cue. The maximum isdurationSeconds - 0.5; the delay reduces the time available for speech.
Use any language for descriptions, speech and end states. Keep the field names and fixed values unchanged. Include no comments, extra fields or trailing commas. Escape quotation marks inside strings as \".
Use a separate narrator
Choose your TTS provider and voice in Studio's voice settings before generating. For Gemini TTS, select the Google TTS option and a supported Gemini voice/model. Your selected voice settings apply to TTS narration; choosing audioSource: "tts" takes precedence over keeping the video model's voice.
ASTERIA VIDEO 1
{
"prompt": "A calm documentary about a craftsperson shaping a wooden bowl. No on-screen text or logo outro.",
"aspectRatio": "16:9",
"audio": true,
"audioSource": "tts",
"backgroundAudio": false,
"narrationDirection": "A warm British English documentary narrator. Speak clearly and fluently, without unnecessary pauses.",
"scenes": [
{
"durationSeconds": 8,
"prompt": "Close-up of the craftsperson turning the wooden bowl on a bright workshop bench. The camera moves closer while the work continues.",
"narration": "Every piece begins with an idea, shaped with care into something we can use every day.",
"transition": "continuity_bridge",
"endState": "The bowl rests on the bench beside the craftsperson's right hand."
}
]
}
This example sets backgroundAudio: false for narration only. TTS provides off-camera narration, not character lip-sync. For characters speaking on screen, use audioSource: "video_model".
Combine narration with cinematic sound
For documentaries and explainers, try TTS narration with the video model's ambient sound. Describe heartbeats, room tone, environmental effects or instrumental music in the scene's prompt; put all spoken words only in narration. The background is lowered while narration plays. The video model is instructed not to speak or sing, but generated audio still needs listening to before publication.
This independent 16-second example opens on a star field with a heartbeat, then moves into the bloodstream. Speech starts after the first second. The visual opening is explicitly written into the prompt: choosing TTS does not decide what appears on screen.
ASTERIA VIDEO 1
{
"prompt": "An imaginative science-documentary opening, moving from a star field to anatomically accurate deep-red blood cells. No on-screen text, logos or end card.",
"aspectRatio": "16:9",
"audio": true,
"audioSource": "tts",
"backgroundAudio": true,
"narrationDirection": "A warm British English documentary narrator, clear and fluent, with a sense of wonder.",
"scenes": [
{
"durationSeconds": 8,
"prompt": "Begin with a wide star field. During the first second, hear a soft heartbeat and atmospheric instrumental music. Move smoothly towards a red point of light that becomes a deep-red blood cell inside a vessel. Keep the heartbeat gentle under the scene.",
"narration": "Inside every one of us, a remarkable journey is happening.",
"narrationSpeed": 1,
"narrationStartSeconds": 1,
"transition": "continuity_bridge",
"endState": "A single deep-red blood cell fills the foreground inside the vessel."
},
{
"durationSeconds": 8,
"prompt": "Continue following the same blood cell through the vessel, revealing many biconcave red blood cells. Add gentle flowing ambience and a restrained instrumental score.",
"narration": "Red blood cells carry oxygen through the body, helping our organs work every day.",
"narrationSpeed": 1.1,
"narrationStartSeconds": 0,
"transition": "continuity_bridge",
"endState": "The cells continue flowing through the vessel as the camera follows."
}
]
}
Keep spoken words out of the visual prompts in TTS mode to avoid conflicting instructions. For an intentional scene without speech, set narration to ""; its background sound can still play. audio: false overrides both narration and background sound.
Narration is fitted to each scene's available time without rewriting its words. Keep lines short enough to speak clearly. If narration cannot be applied safely, the usable video is returned with a warning rather than discarded. A shorter video uses only the available timeline. Resuming a partial video rebuilds narration against the assembled timeline, so earlier narration may be synthesised again.
Give recurring narrators their own voices
Define a speaker once in speakers, then reference that name from each scene. A name is a reusable label, not a voice clone or an instruction to imitate a real person. To make narrators audibly distinct, select different voices; direction alone may change delivery without changing the underlying voice.
Inheritance is resolved separately for direction and voice: scene → named speaker → video default → Studio/workspace settings. Blank direction inherits. An omitted or null voice inherits. A voice selection is replaced as a whole, so a scene override must include both its provider and voice. Scene overrides never change the speaker's other appearances. Speed and start time remain per-scene controls.
The following complete 24-second example uses a British host, a French-accented English narrator, then returns to the original host:
ASTERIA VIDEO 1
{
"prompt": "A documentary about the people behind thoughtful work. Show documentary footage, not on-screen speakers. Gentle instrumental music and natural ambience. No captions, logos or end card.",
"aspectRatio": "16:9",
"audio": true,
"audioSource": "tts",
"backgroundAudio": true,
"narrationDirection": "Clear, warm documentary delivery.",
"narrationVoice": {
"provider": "google",
"voice": "Kore",
"languageCode": "en-GB"
},
"speakers": {
"Host": {
"narrationDirection": "A warm female narrator speaking natural British English.",
"narrationVoice": {
"provider": "google",
"voice": "Kore",
"languageCode": "en-GB"
}
},
"Étienne": {
"narrationDirection": "A calm male narrator speaking English with a natural French accent. Fluent and thoughtful, without caricature.",
"narrationVoice": {
"provider": "google",
"voice": "Charon",
"languageCode": "en-GB"
}
}
},
"scenes": [
{
"durationSeconds": 8,
"prompt": "Track through a sunlit workshop as hands shape a wooden bowl. Gentle workshop ambience.",
"speaker": "Host",
"narration": "Every object begins with people: their ideas, their patience, and their care.",
"narrationSpeed": 1,
"transition": "continuity_bridge",
"endState": "The finished bowl rests on the workshop bench."
},
{
"durationSeconds": 8,
"prompt": "Cut to a designer's sketchbook beside a Paris studio window. Turn a page to reveal a new drawing. Quiet paper sounds and distant city ambience.",
"speaker": "Étienne",
"narration": "For me, good design begins by listening, and understanding how someone will use it.",
"narrationSpeed": 1.05,
"transition": "editorial_cut",
"endState": "The camera settles on the completed drawing."
},
{
"durationSeconds": 8,
"prompt": "Return to the original workshop. Follow the same bowl being wrapped carefully in paper. Soft paper sounds beneath the music.",
"speaker": "Host",
"narrationDirection": "The same female British English narrator, now gently hopeful and reflective.",
"narration": "Different perspectives, shared purpose. That is how thoughtful work finds its way into everyday life.",
"narrationSpeed": 1,
"transition": "editorial_cut",
"endState": "The wrapped bowl rests on the bench, ready for its next journey."
}
]
}
Choose a voice
provider: "google"selects Gemini TTS using Asteria's default Gemini TTS model. Examples of supported voice names areKore,Charon,Puck,Aoede,LedaandAchird. Voice names are case-sensitive. OptionallanguageCodespecifies the spoken language, such asen-GB,fr-FR,hi-INorja-JP. Put regional accent and delivery preferences innarrationDirection.provider: "openai"or"azure_openai"selects an OpenAI-family preset voice:alloy,ash,ballad,cedar,coral,echo,fable,marin,nova,onyx,sage,shimmerorverse. Availability depends on the configured service/model. UsenarrationDirectionfor language and accent; omitlanguageCodefor these providers.- An explicit voice overrides the workspace's preset or custom voice for that scene. If you omit all explicit voice selections, existing workspace voice behaviour applies.
Write the narration itself in the language you want spoken; setting a language or accent does not translate the script. Reusing a speaker selects the same voice and direction, but expressive delivery can vary between generated takes. Exact accent reproduction is not guaranteed.
Explicitly selected voices are not silently substituted with another provider if synthesis fails. Studio keeps the usable video and shows a narration warning. Fix the selection or try again. Invalid speaker names, unsupported voice names and malformed fields are reported before generation is queued.
Each scene has one TTS narrator. For alternating voices, use successive scenes. This format does not support overlapping speakers or several timed dialogue turns within one scene, and it does not synchronise TTS to on-screen mouths. Use native video-model dialogue for on-screen performances. audio: false disables all speech and background sound; named-speaker settings are only used in TTS mode.
Ask another AI assistant to write it
Copy this instruction together with the format and example above:
Create an Asteria Engine authored video prompt for my idea. Return only ASTERIA VIDEO 1 followed by valid JSON in the documented format. Keep every scene between 3 and 10 whole seconds, with a total of [your requested duration] seconds. Include prompt, aspectRatio, audio, audioSource and scenes. Use audioSource [video_model or tts]. Each scene needs durationSeconds, prompt, transition and endState. For TTS, also include each scene's exact narration and a shared narrationDirection. Set backgroundAudio true for model-generated ambience, effects and instrumental music, or false for narration alone. Keep speech out of visual prompts in TTS mode. Optionally set scene narrationSpeed (0.25-2.0) and narrationStartSeconds (0 to durationSeconds minus 0.5) for deliberate timing. Choose native audio for on-screen dialogue, or TTS for a separate narrator, according to my request. For recurring narrators, define named speakers with narrationDirection and narrationVoice, then reference speaker in each scene. Use different explicit voices for distinct narrators. Scene direction and voice override speaker defaults; omitted values inherit. Keep one narrator per scene. Keep the cast and physical state consistent between scenes. Fit speech into the time available while actions happen. Write descriptions and spoken lines in [your language]. My idea is: [your idea].
You can also ask Asteria's own AI assistant to write an authored script from your idea, choose the appropriate audio approach and preview the generation cost in chat. After you confirm, it can start generation and check progress without requiring you to complete the action in Studio.
Review the output before spending credits. An AI assistant can still make JSON or timing mistakes.
Generate and revise
When your workspace supplies a logo reference, it guides the appearance of branding that belongs naturally in the scene, such as an office sign. It is not a request for a logo shot between scenes. Request a standalone logo shot explicitly in the scene where you want it. Studio's automatic end card is a separate setting; switch it off for a film that should continue into another instalment. Generated branding can still vary, so review the result.
For precise colour, describe the subject's colour and lighting together. For example: “Deep crimson cells under clear neutral illumination, with strong local contrast and minimal atmospheric haze.” Soft volumetric lighting and hazy cinematic grading may produce a paler look even when you request red subjects. This is creative direction for that video, not a colour filter applied to all videos.
- Open Studio, select Video and open Prompt.
- Paste the complete authored prompt.
- Check References, Output and end-card choices.
- Choose Generate media, not Design or Edit selected media.
- Review the finished video. To revise, change the relevant scene text in the prompt and generate a new version.
Invalid formatting is reported before a generation is queued. Correct the indicated field and try again. If the timeline is too long for the selected destination or workspace, shorten it or choose a supported destination.
Scene directions are instructions to a generative model, not a guarantee of exact motion, speech, identity or timing. Longer videos are connected from multiple sections and can have visible transitions. If the model returns less footage than requested, Studio can deliver the usable shorter result with an expandable explanation. Review the result before posting.