How often should an n8n workflow check whether a render is finished?

Use the completion mechanism the provider supports, such as a callback or a bounded polling loop. If polling is required, record the job ID, use a reasonable backoff and stop on a terminal state or a defined timeout. The appropriate cadence depends on the provider and job behavior.

Avoid restarting the render merely because a status check failed. A status request and a new generation request are different operations. Keep their retries separate so an observability problem does not become duplicated media production.