No charge for failed generations

Makernice does not charge for failed AI video or image generations. Creating a generation holds the quoted price from your prepaid balance. The hold is converted to a charge when the run completes. If a run fails or times out, the hold is released in full. Input-moderation rejections before generation and queued cancellations also release the hold in full. The one exception: if a finished output is blocked by output moderation for a policy violation, the run is still charged; you can contact support for a manual review.

Applies to all 16 models on one API key. Source: Refund Policy §2 · API docs: Pricing & billing. Last checked: October 21, 2026.

Prices as of October 22, 2026 (price list v5).

How it works: quote → hold → charge or release

1. Quote. POST /v1/generations/quote returns the exact amount for your model, mode, resolution and duration (video is priced per second, images per image). The price is locked for 12 minutes (expires_at). Quoting is free and holds nothing.
2. Hold. POST /v1/generations with the quote_id puts that amount on hold; only now is any balance held, and it is settled when the generation reaches a final state. The generation starts with status: "queued" and billing: "held". The money stays in your balance; it is only reserved.
3. Charge on completion. When the generation reaches succeeded, the held amount is charged (billing: "charged"); once the output passes moderation, the file is ready to download.
4. Release on failure. If it ends as failed or canceled, the hold is released in full (billing: "released"). There is nothing to claim and nothing to request: it happens automatically.

A released hold is not a refund. The money never left your balance, so there is no refund request, no waiting period and no payment-provider round trip.

What counts as a failed generation?

From our Refund Policy (§2) and the API docs:

Quality is not a failure condition: a delivered video or image is charged, whatever you think of it. That's why every model page shows the cost of a short, low-resolution test run first.

Examples (price list v5, October 22, 2026)

Amounts in the API are strings in micro-USD (1,000,000 = $1.00).

{
  "id": "gen_…",
  "model": "seedance-2.5",
  "status": "failed",
  "amount": "2592000",
  "billing": "released",
  "outputs": [],
  "error": { "code": "timeout", "message": "…" }
}

Handle it in code (Python)

import time, requests

def wait_for(gen_id, headers, api="https://api.makernice.com/v1"):
    while True:
        gen = requests.get(f"{api}/generations/{gen_id}", headers=headers).json()
        if gen["status"] == "succeeded":
            return gen                       # billing == "charged"
        if gen["status"] in ("failed", "canceled"):
            # billing == "released": the full hold is already back in your balance
            raise RuntimeError(f"{gen['status']}: {gen.get('error')}")
        time.sleep(5)                        # queued / generating / moderating / checking

Prefer webhooks? generation.failed and generation.canceled are sent after the held amount was released; generation.succeeded after it was charged. See Webhooks.

Two more safeguards that protect your balance

• No double charges on retries. POST /v1/generations requires an Idempotency-Key. Repeating the request with the same key, quote and input returns the original generation and does not charge again. (Idempotency)
• Per-key spend limits. Each API key can have a spend limit and an expiry date. A request that would push a key over its limit is rejected with 402 key_spend_limit_reached before anything is held. Failed and canceled generations don't count toward a key's spend. (Authentication)

What to check with any AI video API before you build

Questions to ask any AI video API before you build on it:

1. Is a failed or timed-out run charged, partly charged, or free?
2. Is money charged at submit time or when the run completes?
3. Does a content-moderation rejection cost anything, before and after generation?
4. Can you cancel a queued job, and is the money returned?
5. Is the price locked before you run, or calculated afterwards?
6. Are retries idempotent, so a network error can't charge you twice?

On Makernice the answers are: free · when the run completes · not before generation (a blocked finished output is charged) · yes, in full · locked by a quote for 12 minutes · yes.

Frequently asked questions

Related

• Pricing — live unit prices, top up from $15 (to $1,000), no subscription
• How AI video API pricing works — price = unit price × seconds, quote locks for 12 minutes
• API docs: Pricing & billing · Check status · Cancel · Idempotency
• Video API best practices: idempotency, 429s and polling
• Refund Policy
• All 16 models