Auto-generate cinematic multi-scene videos from text prompts with n8n

Google Sheets row → GPT-4o narrative & storyboard → Fal.AI Seedance per-scene rendering → merged MP4 with ambient audio

Who

Content creators, short-form video marketers, educators, and n8n users with some automation experience

You get

Turn a spreadsheet row into a multi-scene cinematic video (MP4) with optional auto-upload to YouTube

You need

n8n · fal-ai · google-sheets · youtube

Steps

1. Prepare input and trigger the workflow

   Create a Google Sheet with columns: story (short prompt), number_of_scene, duration (per clip), aspect_ratio, resolution, model. In n8n, click 'Execute workflow' manually or set a Sheet-update trigger.

2. Generate full narrative

   An AI Agent node calls GPT-4o-mini to generate an immersive, emotionally arced short video story based on the story field. Output is constrained by Structured Output Parser to {story: string}.

   Execution prompt · Source text。AI Agent node 'Generate Full Narrative from Prompt' text parameter

   Copy prompt

   =Instructions:
   
   Create a captivating, vivid, and detailed short video story based on the idea provided below. The story should feel immersive, emotional, and cinematic — suitable for visual storytelling.
   Please ensure the following:
   
   Include a rich, visually descriptive setting (lighting, weather, architecture, colors).
   
   Describe the main characters clearly: their appearance, mood, clothing, and motivations.
   
   Build a sense of mood, tension, or emotional arc.
   
   Keep the tone aligned with the theme of the original idea.
   
   Format your output as long-form narrative text.
   
   Idea: {{ $json.story }}

3. Split narrative into scenes and verify count

   A second AI Agent node (GPT-4o-mini) breaks the narrative into exactly number_of_scene scenes, output constrained to {scenes: string[]}. A Code node then verifies sceneCount; if insufficient it loops back, otherwise proceeds.

   Execution prompt · Source text。AI Agent node 'Break Narrative into {{n}} Scenes' text parameter

   Copy prompt

   =<Instructions>
   
   Break down the following story into a script (for short video) with exacly {{ $('Get Data').item.json.number_of_scene }} scenes.
   
   </Instructions>
   
   <Story>
   {{ $json.output.story }}
   </Story>

4. Generate structured video description per scene

   A Split Out node separates the scenes array into items. Each item flows into an AI Agent node (GPT-4o) that outputs a structured scene description: characters, background, camera movement, object movement, and sound effects. Constrained by Structured Output Parser to {characters, scene_description, camera_movement, object_movements, sound_effects}.

   Execution prompt · Source text。AI Agent node 'Describe Each Scene for Video' text parameter

   Copy prompt

   =Prompt Template: Generate Detailed Scene (5s Clip)
   
   Instructions:
   Create a detailed visual breakdown of the following scene.
   Use the full story context to ensure continuity, tone, and emotional flow.
   
   For this scene, include:
   
   1. Characters in the Scene:
   - List all characters present
   - Describe their age, attire, appearance, posture, emotion, and current action
   
   2. Scene Background & Environment:
   - Describe the setting in detail (lighting, background elements, color palette, time of day)
   - Mention any objects in the environment that affect mood or action
   
   3. Camera Direction:
   - Describe camera angle, motion (e.g. dolly, pan, zoom), perspective, or transitions during the 5s scene
   
   4. Object & Character Movement:
   - Describe how characters or objects move within this 5-second clip
   - Use dynamic motion where appropriate
   
   5. Sound Effects & Ambience:
   - Describe background sound, music, or specific SFX
   - Include emotional tone and pacing of sound
   
   Scene Input:
   {{ $json.scenes }}
   
   
   Full Story Reference:
   {{ $('Generate Full Narrative from Prompt').item.json.output.story }}
   
   Output Format Example:
   
   Characters:
   - Anya (17): Wears a tattered blue coat. Snow clings to her hair. She breathes heavily, face flushed from the cold. Her eyes are locked ahead, determined.
   - The Old Man (60s): Frail, cloaked in furs, holding a lantern that flickers with a weak flame.
   
   Scene Background: A narrow trail carved through pine trees covered in frost. The sky is turning orange. In the distance, the silhouette of a crumbling tower. Mist coils low to the ground.
   
   Camera Movement: Slow push-in from behind Anya’s shoulder, transitioning to a side dolly as she steps forward and the tower comes into view.
   
   Movement in Scene: Snowflakes fall steadily. Anya trudges forward. The Old Man stops, raises his lantern slightly, casting light across the branches. A raven flutters from a tree.
   
   Sound Design:
   - Soft wind whistling through trees
   - Crunching snow with each step
   - Distant echo of a bell
   - Faint strings rise in tension as the tower is revealed

5. Call Seedance to generate video and poll status

   An HTTP Request node POSTs to https://queue.fal.run/fal-ai/{model} with XML-formatted Characters, SceneDescription, CameraMovements, ObjectMovements, plus aspect_ratio, resolution, duration from Sheets. SplitInBatches processes items one by one; a Wait node delays before HTTP polling of status_url; a Switch node routes on status: COMPLETED fetches the video, IN_PROGRESS/IN_QUEUE loops back to wait.

   Execution prompt · Source text。HTTP Request node 'Call Fal.ai API (Seedance)' bodyParameters.prompt field

   Copy prompt

   =<Characters>
   {{ $json.output.characters.map(character => `<Character>
     <Name>${character.name}</Name>
     <Description>${character.description}</Description>
   </Character>`).join('\n') }}
   </Characters>
   
   <SceneDescription>
   {{ $json.output.scene_description }}
   </SceneDescription>
   
   <CameraMovements>
   {{ $json.output.camera_movement }}
   </CameraMovements>
   
   <ObjectMovements>
   {{ $json.output.object_movements }}
   </ObjectMovements>

6. Add ambient audio to each video

   An HTTP Request node POSTs to https://queue.fal.run/fal-ai/mmaudio-v2 with video_url (from previous step) and prompt (from scene sound_effects). SplitInBatches loops per scene; Wait+HTTP polls completion status to retrieve audio-enhanced clips.

   Execution prompt · Editorial suggestion。HTTP Request node 'Start adding audio to the video' bodyParameters; replace with actual n8n expressions

   Copy prompt

   video_url: {{ $json.video.url }}
   prompt: {{ $('Describe Each Scene for Video').item.json.output.sound_effects }}

7. Merge clips into final video

   An Aggregate node gathers all audio-enhanced clips. An HTTP Request node POSTs to https://queue.fal.run/fal-ai/ffmpeg-api/compose with a jsonBody tracks array, each clip 6000ms sequential. Wait+HTTP polls merge status; finally retrieve the merged video URL.

   Execution prompt · Source text。HTTP Request node 'Start merging videos' jsonBody field

   Copy prompt

   ={
     "tracks": [
       {
         "id": "1",
         "type": "video",
         "keyframes": {{ JSON.stringify($json.data.map((item, index) => ({ url: item.video.url, timestamp: index * 6000, duration: 6000 }))) }}
       }
     ]
   }

8. Optional: auto-upload to YouTube

   If YouTube OAuth2 is configured, the final video can be auto-uploaded via the YouTube node, title pulled from the Sheets story field, category ID 28 (Science & Technology), regionCode TH.

You are setting up an n8n workflow that turns a Google Sheets row into a multi-scene cinematic video. First confirm you have: an n8n instance, Fal.AI Header Token, OpenAI API Key, and Google Sheets OAuth2 credentials. Then import the template, configure credentials, prepare the sheet columns (story, number_of_scene, duration, aspect_ratio, resolution, model), run it, and watch the queue-polling status.