Account, credits, and troubleshooting
Content safety: blocks, service errors, and credits
Distinguish a content block from a temporary moderation failure, including effects on images, streamed replies, and credits.
On this page
Why both inputs and outputs are checked
Submitted content and generated results are checked within the enabled moderation scope. A prompt does not fully determine the output: an accepted input does not guarantee an accepted result. The model service may also refuse a request.
Text, references, output images, chat replies, and thinking content follow the rules applicable to the request. References, negative prompts, and switching to Cards do not exempt content from moderation.
A successful upload only means the file was saved. It does not guarantee that every later generation will pass input and output checks. A working preview is not proof of moderation approval.
Provider moderation is independent of site moderation
Model providers independently check prompts, reference images, or generated results. Passing site moderation does not guarantee provider approval; decisions can differ between providers or models.
If the provider's content check fails, review and revise your input. Without a specific reason from the provider, BANANAI cannot identify the triggering text, image, or category. Site moderation switches cannot disable provider checks.
Provider refusals and site input/output blocks may have different settlement outcomes. Check the task's actual credit record rather than assuming from the moderation message that it is charged or free.
These messages have different causes
| Message or symptom | Meaning | What to do |
|---|---|---|
| Input did not pass content safety checks | Submitted text or references did not pass this input check. This does not establish deliberate misuse. | Review the text and each reference, clarify the legitimate purpose, and remove unnecessary sensitive content. Do not repeatedly submit unchanged content. |
| Generated content did not pass safety checks | The returned content did not pass output checks. Accepted input does not guarantee accepted output. | Adjust the prompt or references. Blocked results cannot be recovered through download, Favorites, or Inspiration. |
| Content moderation service error. Try again later | The check did not complete normally; content has not necessarily been classified as prohibited. | Try later. If it persists, report the time, feature, and full message instead of submitting many paid requests to test it. |
| Requests are too frequent. Try again later | The generation API has reached a rate or capacity limit, not a content safety block. | Stop clicking repeatedly and try later. Other users may consume the same service allowance. |
Why an already visible reply can disappear
Visible chat fragments have passed checks at that stage. A later check can still block the whole turn. An explicit output moderation block clears the reply and thinking already shown for this turn and replaces them with a failure notice; refreshing does not restore them.
A model error or moderation-service error differs from an explicit block. Approved, saved partial text may remain, but the turn is marked failed and is not a complete answer. Unchecked fragments are not displayed because a check failed.
Sending again creates a new turn, not a continuation from the middle of blocked output. The previous input and failure remain for review and reporting.
Credit settlement for moderation failures
| Situation | Image generation | AI chat |
|---|---|---|
| Explicit input block | Settled at the submitted estimate; no image is delivered. | The turn's reserved credits are released. |
| Explicit output block | Settled by the actual returned image count; blocked images are not delivered. | Settled by actual usage already incurred; blocked reply content is not retained. |
| Moderation unavailable or service error | Fails and releases the relevant reserved credits and storage. | Relevant reservations are released according to the failed status; approved, saved partial text may remain. |
Suspected false positives or repeated service errors
- Use Feedback in the personal area to explain the legitimate purpose, time, feature, and complete message. A single notice cannot identify the exact sentence or reference responsible; do not guess by repeated submissions.
- Capture the failure and necessary context, hiding private images, contact details, verification codes, and payment information. Never submit passwords or browser sign-in data.
- Feedback does not bypass moderation or promise immediate restoration of blocked content or an automatic refund. The platform checks actual task and settlement records.