When credits come back, when they don't, and why.
Our AI provider accepted your job and started generating it on their GPUs. Partway through (or right at the end), their safety classifier flagged the output and refused to return it.
The provider already charged us in real dollars for the GPU time the run consumed — even though it produced no usable output. If we refunded every safety block, every blocked prompt would cost us money out of pocket and the service wouldn't survive.
This is not a bug or a glitch. Every major AI image/video provider works this way.
Stay clear of the categories above and your prompts will go through — credits only get spent on real outputs.
If you believe a generation failed for an infrastructure reason (not a safety block) and wasn't auto-refunded, reach out from the report page with the time of the failed generation and we'll take a look.