When a generation does not succeed, you now see a clear, specific explanation instead of a generic error or a tile stuck on "Generating…" indefinitely. Messages are tailored to the cause — for example a temporary provider issue, a capacity limit, a content-policy block, or a missing input — so you know whether to simply wait, adjust your inputs, or try again.
Temporary problems (brief provider outages, rate limits, capacity shortages) are now retried automatically, with sensible spacing between attempts, so many generations that previously failed now complete on their own without any action from you. Credits are only consumed for work that is actually delivered.