Automating the Pipeline with n8n
At volume, clicking through Runway and Kling by hand does not scale. n8n lets you wire the pipeline together: a script model writes prompts, the generators produce clips, and the results land in storage, all triggered by a single row in a sheet. This is how one person ships ten videos a week.
Step 1: Map the flow before you build it
A working automation is a clear sequence: trigger, generate prompts, call the video API, poll for completion, store the file, notify you. Draw it before you touch n8n so each node has one job.
[Sheet row added]
-> [Claude: write 4 shot prompts]
-> [Split into 4 items]
-> [HTTP: Runway generate]
-> [Wait + poll until done]
-> [Download clip]
-> [Upload to storage]
-> [Slack: 'batch ready']Step 2: Generate prompts with a model node
Use a Claude Sonnet 4.6 node for the prompt-writing step. It is fast and cheap enough to run on every row, and it returns clean structured JSON you can split into individual generation calls.
{
"shots": [
{ "id": 1, "prompt": "slow dolly push-in on a glass of cold brew, morning light" },
{ "id": 2, "prompt": "overhead macro of milk swirling into coffee, slow motion" },
{ "id": 3, "prompt": "close-up of beans pouring into a grinder, warm light" },
{ "id": 4, "prompt": "wide shot of the finished drink on a cafe table, soft bokeh" }
]
}Step 3: Call the generator API and poll
Video generation is asynchronous. You submit a job, get an id, then poll until the status is done. An n8n Wait node plus a polling loop handles this without you watching the screen.
Result: add a row, walk away, come back to four downloaded clips and a Slack ping. The creative judgment stays human; the clicking is gone.