Skip to main content
Two separate ceilings guard the API, and both answer with a 429. A rate limit counts how many requests you send per minute. A concurrency limit counts how many long operations you keep running at the same time. Read error.type to tell them apart. A rate limit means slow down. A concurrency limit means wait for work already in flight to finish.

​
Rate limits

The sustained rate is 1,000 requests per minute, and the window holds a little headroom above that so a short burst is not punished. Every response carries the current state of the window.
429, rate limit exceeded

​
Concurrency limits

Renders and uploads hold separate concurrency budgets, and the upload budget is never the larger of the two. Only the endpoints that hold a slot report these headers: a PSD upload, a synchronous render, and a background removal.
429, concurrency limit exceeded
A concurrency 429 carries Retry-After as well, and five seconds is usually enough: the slot frees as soon as the request holding it returns.

​
Queue instead of waiting

A queued render never holds a concurrency slot while it waits. Send is_async as true on POST /api/v1/renders and the call returns a 202 with a job you poll through GET /api/v1/jobs/{job_id}, or receive over a webhook. Video mockups are always queued and behave the same way. That is the right shape for a large batch: submit the whole set, then follow the jobs.

​
Read your own ceilings

Do not hardcode the concurrency numbers. GET /api/v1/packages/plans returns max_concurrent_requests and max_concurrent_uploads for every active plan, and an account on a paid plan is held to that plan’s pair. Without one, an account is held to 1 and 1 in trial, and to 25 and 10 while it holds a prepaid balance or has a top-up from the last 90 days that was not fully refunded. A response that holds a slot reports your own ceiling in X-Concurrent-Limit. Your synchronous uploads are processed one at a time, and queued jobs, uploads included, run up to your render ceiling at a time. Keep concurrent calls within those numbers and handle a 429 with Retry-After.

Errors

Every status code, every error_code, and the retry loop.