While testing Figma Make with large, multi-screen prompts, I noticed what appears to be a gap in credit enforcement during generation.
If a complex generation is initiated and credits are exhausted mid-process, the job does not appear to hard-stop. Instead, generation continues server-side until a valid, renderable result is produced, particularly if the client/app is still loading or reconnecting. This suggests credits may be checked only at task start rather than continuously enforced during execution.
From a systems perspective, this could result in:
-
Long-running generation jobs completing without corresponding credit coverage
-
Excess compute being consumed in scenarios where delivery integrity is prioritized over quota enforcement
-
Increased backend load for large, high-coherence prompts
I’m reporting this in good faith because it looks like an architectural edge case rather than intended behavior, and it could lead to unnecessary processing cost at scale.