Recover from generation failures

A failed generation is easier to recover when you first identify where it stopped.

Read the visible state

Confirm whether the run is queued, processing, failed, or complete. An interrupted browser request does not prove that the provider stopped, so check the node again before starting a duplicate run.

Match the cause to the correction

Common causes include an unsupported input or setting, unavailable source media, temporary provider failure, moderation rejection, insufficient credits, or a broken connection to a required asset.

Change the input related to the error. Repeating the same expensive request rarely produces useful diagnostic information.

Recover before you retry

After a disconnect or reload, return to the same workspace and let the existing run state recover. Check the node or progress state before resubmitting. Recovery follows the original run; a duplicate prompt can create competing work and additional cost.

Retry safely

Recheck the model capability, prompt, references, destination format, balance, and previous status. When possible, reduce duration, resolution, or batch size for the next test.

Escalate with useful context

If the problem continues—or credits changed without a usable result—record the workspace, approximate time and timezone, media type, model if visible, final status, and expected versus observed outcome. Never include passwords, tokens, verification codes, or full payment details.

You have now completed the creative-workspace course: from intent and model choice through generation, curation, and recovery.