Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.apiyi.com/llms.txt

Use this file to discover all available pages before exploring further.

​
Plan Overview

To better serve our customers, API易 is launching the Nano Banana Pro Failed-Generation Credit Reimbursement Plan.
The only covered model is the per-request Nano Banana Pro. For Nano Banana 2, you can choose a token-based token, where failed calls are billed at a negligible amount and can be ignored.
When a request returns status code 200 but the image generation fails, this is feedback from Google’s side — API易 is a transparent proxy that simply forwards the result, and we want your images to succeed just as much as you do. This plan reimburses credits for failures caused by these non-subjective reasons.

​
When Does Nano Banana Pro Fail to Generate?

Google’s content moderation policies keep tightening. Common situations that trigger a rejection include:
  • Content safety: NSFW or minor-related content
  • Watermark removal: a particularly special category
  • Well-known IP (added January 23, 2026): such as Disney — Google appears to have introduced a new moderation policy
  • Stricter safety mechanisms (February 27, after Nano Banana 2 launched): well-known public figures, financial/order information edits, person outfit swaps/face swaps, implicit sexual content, etc., all return a text error message similar to “I cannot complete the requested edit to xxx”

​
Symptoms of a Failed Generation

Log output Tokens under 1000

Google returns a line of text, for example: “I’m unable to help with this task” / I'm just a language model and can't help with that.

Log output Tokens empty

The image generation is outright rejected and the error is empty. The key metric in the API response data is "candidatesTokenCount": 0
Log list where the completion Tokens column for the gemini-3-pro-image-preview model shows tiny values such as 173 and 176, marked as failed generations

​
Why Is a Failure Still Charged?

  • Google deducts quota: repeatedly sending the same request without modifying the prompt wastes Google’s RPD quota
  • The system cannot yet support “no charge when output is empty”
  • Requests for prohibited content make it more likely for our official KEY to be banned — an irreversible loss

​
How to Join?

Eligible customers:
  1. Monthly spend starting at USD 1000 (any on-site model) — the threshold is intentionally low; small-credit testing and personal use have a low probability of failures anyway
  2. Aimed at tool service providers: because it is hard to control end users’ input content on the customer side
  3. Non-subjective reasons only: maliciously requesting the same/similar prohibited content will not be reimbursed
  4. Time range: starting May 1 (UTC+8)
  5. We will reach out gradually to register you on our internal list

​
How Is the Reimbursement Done?

Go to the Logs section, click Export in the top-right corner, select a time range (for example, last month), and choose Async Export. After submitting, check the progress under Async Tasks in the top navigation menu, and download the final Excel data result.
Logs section toolbar with the Export button in the top-right corner highlighted
Reimbursed credits: We tally the exact number of failed requests and reimburse based on failed count × model price / discount factor (for example, with a 15% top-up bonus, divide by 1.15). Reimbursement timing: At the end of each month we tally the previous month’s data, and reimbursement is typically credited within the first 5 business days of the following month.