When an AI Generation Fails or the Browser Closes: Charges, Records, and Recovery
Understand what happens to your source design, usage record, charge, generated asset, and draft when an AI preview fails or finishes after you leave the page.
AI image and video generation depends on several systems: your saved draft, source media, the configured model, a provider request, output storage, and a usage charge. A failure in one step should not make the other records ambiguous. Labelix keeps them separate so the outcome can be diagnosed and, where required, recovered or refunded safely.
Your source design and the generated output are different
The garment surface, uploaded artwork, placement settings, and selected craft are source inputs. A craft effect, virtual try-on, or showcase video is a generated output. A failed generation does not delete the source design.
Return to the draft and check the underlying surface first. If the base image, artwork, placement, or craft changed, fix that before retrying. Repeating the same invalid input usually produces another avoidable failure.
Validation happens before charging
labelix checks authentication, owned inputs, provider capability, model pricing, and required media before starting a charge. A request rejected at this boundary should not consume an AI generation charge because no provider work began.
After validation, the usage record and charge create an audit trail for the operation. The record can remain visible even when the provider later fails; a failed row is evidence, not a successful asset.
Provider failures follow the refund path
When a charged operation fails in a route that requires refunding, Labelix records the failure and returns the linked charge through the wallet policy. Promotional credits and wallet funds can follow different ledger rules, so review the account balance and AI usage history rather than expecting a card-style refund notification.
The refund is idempotent: reconciliation can retry safely without crediting the same failed usage twice.
Closing the browser does not always stop video work
Video generation can continue on the server after the page closes. If the provider succeeds, Labelix persists the usage and asset even though the browser did not receive the response. The matching draft can recover the newest eligible completed video when you reopen it.
Recovery uses the draft identity, mode, source provenance, and completion time. It will not intentionally resurrect an older video that you removed after saving, and it cannot attach a result whose source surface no longer exists.
A server restart has a separate recovery path
Long-running video tasks store their provider task identity after submission. A scheduled reconciliation process revisits older requested tasks: completed provider work becomes a stored asset, terminal failures are marked and refunded where required, and unresolved tasks are retried within a bounded window.
Only one finisher can settle a usage record, preventing a long browser request and the scheduled process from creating duplicate assets or conflicting refunds.
Safe troubleshooting sequence
- Reopen the same draft and wait for any eligible recovered result.
- Check the AI usage history and wallet ledger.
- Confirm the source surface is complete and every design-bearing print has a craft effect.
- Read the customer-facing failure message and correct actionable input issues.
- Retry only after the previous usage has reached a terminal outcome.
If a charged failure has no refund record after reconciliation, or a successful usage has no accessible asset, contact support with the usage time and draft context. Do not send private API details or duplicate the generation repeatedly.
Frequently asked questions
- Does a failed AI generation delete my source artwork or garment design?
- No. Source media and draft placement are separate from the generated output. Reopen the draft, check the surface and craft setup, and correct the input before retrying.
- What happens if I close the browser during video generation?
- The server can continue the provider task. If it succeeds, the usage and asset are persisted, and the matching draft can recover the newest eligible result when reopened.
- Can a failed charged generation be refunded twice?
- No. The charge and refund are linked through the usage and wallet ledger, and the refund path is idempotent so retries or scheduled reconciliation do not credit the same failure twice.