Skip to content
Book a callStart free

Limits, and how to raise them

How many requests and jobs you can have at once. A request over a limit is refused with a Retry-After, never queued, never charged.

View as Markdown

Requests per minute, in fixed 60-second windows. Each is counted for the key and for every key on the same paying account together.

ClassWhat countsPer keyPer account
startA paid start (make_video, make_image, edit_image, recast, make_music, creator_video, run_template)1030
quoteA dry_run call, and GET /models with a key60180
readGET /results, GET /results/{id}, GET /templates, the callback log120360
uploadPOST /uploads3060
writeDrafting an ads launch2040
  • Jobs at once: up to 8 jobs queued or running per paying account. A key can carry a lower limit of its own. Test-key runs do not count.
  • Test keys: up to 200 test runs per person per 24 hours (TEST_LIMIT).

Over a request limit the answer is 429 RATE_LIMITED with Retry-After (seconds to the end of the window). A refused request still counts, so waiting works better than hammering. Over the jobs limit the answer is 429 CONCURRENCY_LIMIT with Retry-After: 15. Work you already started keeps running.

If our own capacity is full the answer is 503 PROVIDER_BUSY. That is not your fault and nothing is charged; retry after Retry-After.

You can cap spend per key and per account, per day and per month, on the Developers page. Over a cap the answer is 402 BUDGET_EXCEEDED with budget.resets_at. It never triggers an automatic top-up.

At most 25 active keys per person, and 20 new keys an hour (KEY_LIMIT). Too many bad keys from one address in a minute answers AUTH_RATE_LIMITED.

Poll politely: every 15 to 30 seconds for a job, or use a callback. Dry runs have their own, larger limit (quote), so check prices as often as you need.

If you need higher limits, book a call or write to hello@diffusr.ai. Say your use, how many jobs at once, and how many requests a minute.