> ## Documentation Index
> Fetch the complete documentation index at: https://docs.dafty.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Rate limits

> Per-key limits and how to back off.

Rate limits are applied **per API key**, per minute.

| Endpoints                                                             | Limit     |
| --------------------------------------------------------------------- | --------- |
| Image creates - `generate`, `edit`, `resize`, `upscale`               | 20 / min  |
| Video creates - `image-to-video`, `frames`                            | 5 / min   |
| Style extraction - `POST /styles/extract`                             | 3 / min   |
| Uploads - `POST /uploads` (presign)                                   | 30 / min  |
| Everything else - `account`, job polling/listing, style reads, cancel | 120 / min |

## Reading the limit

Responses include standard rate-limit headers so you can pace bursts off the server's signal:

* `X-RateLimit-Limit` - allowed in the current window
* `X-RateLimit-Remaining` - remaining in the current window
* `X-RateLimit-Reset` - seconds until the window resets

When you exceed a limit you get `429` with a `Retry-After` header (seconds):

```json theme={null}
{ "error": { "code": "RATE_LIMIT_ERROR", "message": "Rate limit exceeded" } }
```

## Backoff

* Wait the number of seconds in `Retry-After`, then retry.
* For bursts, watch `X-RateLimit-Remaining` and slow down before you hit zero.
* Concurrency is also bounded - too many in-flight generations returns `429`
  `CONCURRENT_LIMIT_REACHED`; poll your existing jobs before starting more. Video has a
  lower concurrency ceiling than images and runs on a small dedicated pool, so clips may
  queue (stay `pending`) before they start `processing` - keep polling.

<Note>
  Long-polling with `?wait=true` counts as a single request against these limits while it's held - see
  [Polling & wait](/polling-and-wait).
</Note>