image-generate now has a second generation path. Venice stays primary. A
definite auth, billing, rate-limit, server, or empty-image response starts one
bounded fal.ai Z-Image Turbo call.
Fallback responses say which provider made the image. They include
source: "fal-ai", the original requested_model, a short fallback_reason,
and the fal.ai request ID when the provider returns one. The endpoint keeps its
existing price and request shape.
A disconnected Venice request remains a Venice dependency error because that job may still finish upstream. This avoids paying two providers for an ambiguous timeout.