Build an Image-to-Video Workflow in n8n, From Asset to Review Queue

Connect asset validation, prompt assembly, async generation, retries, and human review into an operable pipeline.

Key takeawayAutomation is not about removing one click. It is about making every job traceable, replayable, pausable, and owned.

Start with a data contract

Before dragging nodes, define the fields for every job: asset_url, prompt, aspect_ratio, duration, route, request_id, status, and review result. A clear contract makes it easier to replace a spreadsheet, storage system, or model without rebuilding the entire workflow.

A practical node sequence

Trigger on a new asset or spreadsheet row.
Confirm that the file is reachable, reasonably sized, and supported.
Assemble the prompt from a brand template while retaining original input.
Submit an image-to-video job to ModelRush and save request_id.
Wait for a webhook or reconcile status with backoff polling.
Copy the result into owned storage and create a lightweight preview.
Send it to human review; publish accepted work and record rejection reasons.

Separate three error classes

Transient limits and network failures are retryable. Unsupported formats and missing assets need input correction. Creative dissatisfaction is not a technical failure. Route these classes into different branches to prevent wasteful automatic retries.

ModelRush recommends a maximum attempt count and budget for each job. When either threshold is reached, pause and notify an owner instead of continuing to spend.

Make the workflow operable

Write every job to a simple table with current state, attempts, selected model, cost, duration, and reviewer. At minimum, a dashboard should show waiting jobs, failure reasons, average delivery time, and cost per usable output.

Finally, verify replay behavior with fixed test assets. Resuming at any node should neither rebill nor republish, and it must not overwrite a human-approved result.