Why was my account charged but I got no output?

A few rare conditions produce a charge without a visible output. The most common is a successful generation on the provider side that fails to upload to Martini's storage — a transient network error during the asset transfer, a storage quota issue, a webhook timeout that misroutes the result. The provider already ran the compute and charged Martini for the inference, so the credits were deducted; the asset never appeared on your canvas because the post-generation step failed.

Another cause is a generation that completes but is mis-attributed to a different node or project due to a webhook delivery glitch. The output exists on the platform but is not visible in the place you expect to see it. A third, less common cause is a partial output on multi-shot or batch generations where some items succeed and some never return — credits are charged for what the provider produced, even if the upload step did not finish.

Martini's job-status reconciler runs on a recurring schedule and detects orphaned charges (credits deducted with no associated output) within 24 hours. When it finds one, it refunds the credits to the same balance they were spent from — subscription quota first, wallet second — and the failed run is excluded from the billable usage total. Most refunds in this category post within a few hours, not the full 24.

Check Settings, Billing for the refund entry. The recent activity row will show the original deduction, the failure detection, and the refund posting. If the refund is there, no further action is needed. If 24 hours pass with no refund and the asset is still missing, escalate via support — that is outside the normal reconciliation window.

Contact support from the help icon at the bottom corner of the app. Include: the project ID (visible in the URL or project settings), the approximate timestamp of the run, the model used, and a brief description of what you expected to see. Screenshots of the canvas and the Billing usage row help. Support can query the upstream provider for the job ID and confirm whether the inference completed, and can either recover the asset (if it exists in the provider's storage) or restore the credits manually if recovery is not possible.

Most cases are resolved within one business day. If the provider confirms the inference never returned, that itself is a provider-side failure and the refund is posted to your account. If the provider confirms the inference completed and the asset is recoverable, support pulls it into your project so you do not have to re-run.

This category of failure is rare but not zero. Two practical mitigations: stay signed in to a stable network connection during long-running video jobs (a network blip during webhook return is one of the more common triggers), and check the generation history panel in the project for runs you remember triggering but cannot find on the canvas — sometimes the asset is there, just not where you expect. The history panel is the source of truth for what was actually generated against your account.

For high-stakes runs, consider running the same generation twice on different models — the second run is a backup if the first encounters this category of failure. The redundancy costs credits but is worth it for client deliverables on a deadline.