Image generation sometimes fails because the service producing the images has more work than it can take on. Until now that outcome was described the same way as any brief interruption — a tile headed Service is busy, closing with Please try again in a moment.
That wording was too optimistic for this particular cause. A shortage of capacity is not measured in seconds — it can take minutes, occasionally much longer — so "in a moment" set an expectation the situation could not meet. The same failure is now headed Provider at capacity and closes with Please try again in a few minutes.
The banner shown while a generation is being retried follows the same distinction, so a shoot that is waiting on a busy provider reads differently from one waiting out an ordinary interruption.
Nothing about the retries themselves has changed. The platform still retries automatically, the same number of times as before, and most generations still complete without you doing anything.
Failures with any other cause — a blocked image, a missing input, a problem on our side — are described exactly as they were.