Comment by ryanworl

You would use two buckets in this case. Input bucket gets consumed by worker processes to do the transcoding (and validation) and then they upload into the output bucket. The output bucket is what you serve to clients (hopefully with a CDN in front).