Failed generations
Troubleshoot failed AI image and video generations, retry requests, and check credit refunds.
Last verified
Overview
If an image fails, its tile shows a sad face and the error message in red. In Grid the tile has room only for the face, so point at it to read the message. When the model's provider refused the prompt or the request, you see its own words explaining what to change. The wording varies by model.
If nothing in a request arrives, you get its credits back automatically. You don't need to ask.
"Something went wrong on our side"
Batch Banana could not finish your request. Try running it again.
Open the arrow beside Reuse and choose Re-run. In older results, select the failed tile first to open its run. If the request fails again, try another model.
"The input or output was flagged as sensitive"
The provider’s safety filter rejected the prompt, a reference image or the generated result. Messages may mention safety, moderation, content policy, sensitive content or copyright. Credits return automatically if the entire request fails.
Review your prompt and reference images for restricted content before trying again. If the wording is ambiguous, clarify what you want to make.
"The model rejected this request"
The model could not accept your input. If the message includes a reason, use it to adjust your prompt, reference images or settings before trying again.
"Google declined this request"
Google declined the Nano Banana 2.1 or Nano Banana 2 Lite request. This message covers several refusal reasons, so it does not identify a specific problem with your content. Try changing your prompt or reference images. Credits return automatically if the entire request fails.
"Google could not make an image from this request"
Google did not return an image. Try again, or change your prompt or reference images. This message also appears when Google cannot accept the request.
"The model is unavailable right now"
The model's provider is busy or unavailable. You also see this when an image waited so long behind other requests that too little time was left to make it. That image was never sent, so its credits come back. Wait a few minutes and run it again, or choose another model.
"The model couldn't finish this request"
The model failed to finish the request. Choose Re-run to try again, or choose another model.
"The model stopped before finishing"
The model canceled the request before returning a result. Choose Re-run to try again.
"The model returned no usable result"
The model finished, but its result could not be used. Choose Re-run, or try another model if it happens again.
"Generation timed out"
The image didn't arrive in time, so we stopped waiting. Its credits come back if the model never started on it, or if nothing else in the request arrived either. Videos take longer: a video that hasn't finished 20 minutes after the model starts on it times out. Run it again; if the same model times out twice, try another.
Did I get my credits back?
Credits return automatically if the entire request fails, and for any image the model never started. To confirm a refund, open the Credits tab in Settings > Plan and credits, then filter History by Type to Usage. Look for "Refund," followed by the model name and a positive credit amount. The app does not show a separate refund notification.
Some images arrived and others failed
The request counts as complete. You pay for every image the model started, so if you asked for four images, three arrived and the model failed on the fourth, you paid for four. An image the model never started is refunded, even when the others arrived. The missing one shows as a red tile next to the others.
This applies to one prompt on one model. With Vary prompts or a variety pack, each image is its own request, so an image that fails there is refunded on its own.
In a Batch run, the credits for every row that failed or never ran come back together once the run ends, as one refund named after the batch. When you cancel a run, rows that were not sent to the model yet stop and are refunded, and rows already running finish as usual. When a run stops on an error, rows that had not started are refunded, rows already running finish and stay charged, and a row that fails after the stop is refunded on its own.
I deleted a generation before it finished
Deleting a generation that is still running removes it right away. Images the model had not started yet are refunded. Images it is still making finish and are still charged, because the model can't be stopped once it has them, and they are deleted when they arrive. Images that already failed are refunded if none had arrived, the same as when every image of a request fails. Deleting one finished image keeps the rest of its generation.
Contact support
Email support@batchbanana.com if a request failed and no refund shows in History, if a model fails every time with a prompt that should be fine, or if the same error keeps coming back across models. Paste the exact message from the red tile and the time you ran it. We respond within 24 hours.