Zense AI checks your balance before expensive operations and only deducts credits after output is delivered.

This article explains every case where you keep your credits.

Quick reference

Image Studio

Before generation

1. Zense validates model, size, quality, and image count.
2. It computes credits_per_image × num_images and checks your image pool.
3. If you do not have enough → Insufficient image studio credits — nothing is deducted, no fal.ai job runs.

After generation

• Credits deduct only when images are successfully saved (totalImagesGenerated > 0).
• If fal.ai returns an error, the queue fails, or zero images come back → session status FAILED → no deduction.
• If you request 4 images but only 2 succeed → you pay for 2, not 4.

Video Studio

Same pattern as images:

• Pre-check — estimated credits from model + duration + resolution; blocked if insufficient.
• Post-success — deductVideoStudioCredits runs only inside the successful result handler after the video is uploaded.
• Failure — session → FAILED with error message → no charge.

Chat & agent tools

Insufficient chat credits

If you have no chat credits left, new messages return InsufficientCredit before the model runs → no message credit.

Successful messages

After the assistant finishes, the flat model credit is deducted. Agent tools (web search, crawl, YouTube, in-chat images) deduct when the tool succeeds — each tool has its own rate (see Chat & text model pricing).

Edge case — deduction failure after reply

In rare cases the reply may complete but the credit ledger write fails (logged server-side). The message still appears; ops can reconcile from transaction logs.

Tutor Studio (slides, infographics)

Studio items that call image generation follow the image pool rules:

• Pre-check for enough image credits.
• Deduct per successful slide image or infographic render.
• Failed render → no charge for that item.

What you still see in the UI

• A failed session may appear in Image/Video Studio history (for retry).
• Token counts may be recorded on partial chat turns that errored mid-stream — but if the assistant message did not complete, the per-message model charge should not apply.

Check Settings → Usage → Transaction history to confirm what was actually deducted.

School pools

• The same rules apply to school accounts:
• Failed runs do not increase used on the school pool.
• Insufficient balance blocks the action for that user — no overspend.
• Admins can see aggregate usage in School Management → Usage without paying for students' failed attempts.

If you believe you were charged for a failed generation, export your credit history and contact support with the session or message ID.