API Get key

Limits

What the API enforces, read from the running code each time this page loads.

Rate limits

Counted over a sliding window of 60 seconds, per API key (per account for a signed-in browser).

LimitCounted perApplies to
30 a minuteAPI keyCalls that create or upload:POST /api/v1/characters POST /api/v1/characters/ready DELETE /api/v1/characters/{character_id} PATCH /api/v1/characters/{character_id} POST /api/v1/characters/{character_id}/sheet POST /api/v1/formats/{format_id}/series DELETE /api/v1/formats/{format_id}/series/{series_id} PATCH /api/v1/formats/{format_id}/series/{series_id} POST /api/v1/formats/{format_id}/series/{series_id}/comments POST /api/v1/generate/image POST /api/v1/generate/video POST /api/v1/jobs POST /api/v1/media/{media_id}/touchup POST /api/v1/products POST /api/v1/runs POST /api/v1/runs/from-analysis POST /api/v1/runs/propose POST /api/v1/runs/{run_id}/approve POST /api/v1/runs/{run_id}/redo POST /api/v1/runs/{run_id}/scenes POST /api/v1/runs/{run_id}/scenes/quote POST /api/v1/scripts POST /api/v1/scripts/{job_id}/make POST /api/v1/studio/glowup/captions PUT /api/v1/studio/library POST /api/v1/studio/pack/runs POST /api/v1/studio/{studio_id}/helpers/{name} POST /api/v1/studio/{studio_id}/helpers/{name}/cancel POST /api/v1/studio/{studio_id}/runs POST /api/v1/studio/{studio_id}/series DELETE /api/v1/studio/{studio_id}/series/{series_id} PATCH /api/v1/studio/{studio_id}/series/{series_id} POST /api/v1/studio/{studio_id}/series/{series_id}/comments DELETE /api/v1/studio/{studio_id}/user-cards/{card_id} POST /api/v1/uploads POST /api/v1/uploads/from-library POST /api/v1/uploads/from-url
120 a minuteAPI keyEvery other call with a key (58 endpoints)
60 a minuteIP addressCalls without a valid key
30 a minuteAPI keyMCP tools that spend credits
120 a minuteAPI keyFree MCP tools
30 a minuteAccountRuns from a model page's playground

The REST API and the MCP server share one count per key: a key's MCP calls and REST calls add up.

What a 429 looks like

Over a limit you get this answer. Wait the seconds in Retry-After (also retry_after in the body), then try again.

HTTP 429
Retry-After: 12
Content-Type: application/json
{
  "error": {
    "code": "rate_limited",
    "message": "Too many requests. Try again in 12 s.",
    "retry_after": 12
  }
}
  • Nothing is charged by a refused call.
  • More than 2 uploads from a link at once from one account, or more than 8 across the whole server: the same 429, with 10 seconds to wait.

Daily credit caps per key

  • Optional, set per key when you make it. It counts credits charged minus credits that came back with that key over the last 24 hours, not a calendar day.
  • Any whole number from 1 up to 10,000,000 cr. No cap: the balance is the only limit.
  • A job, run or studio click that would go over is refused with 402 daily_limit_reached, and nothing is charged.
  • Chat apps connected through sign-in have no daily cap unless you set one in Settings > Connections.
  • Up to 50 active keys per account, by plan.
Set caps on your keys

Uploads and files

Applies toLimit
One image (JPG, PNG, WEBP, GIF, AVIF, HEIC)20 MB
One video or audio file1024 MB
Download time for an upload from a link120 s
Uploads from a link at once, per account2
Uploads from a link at once, whole server8
Files are kept (longer on Starter and up)30 days

Waiting and repeats

  • A long poll (wait on GET) holds the request up to 50 seconds; MCP tools wait up to 45 seconds.
  • Up to 8 long polls at once per account, and 40 across the whole server. One more answers after a pause of up to 2 seconds with the current state, without waiting for the end.
  • The API sets no per-account limit on how many jobs run at once: your balance and a key's daily cap decide what starts.
  • Idempotency-Key: a repeat with the same key and body within 24 hours returns the first answer and is not charged again.

Job lengths and time limits

Your plan runs a set number of takes at once; more wait their turn and start by themselves. A job that has not started 30 minutes after its turn comes fails and its credits come back automatically. Once started, each model has a time limit below: past it, the job fails and its credits come back automatically.

ModelModel idLongest clipTime limit
Ad analyzeranalysis.adNot set20 min
Add a voiceaudio.song-vocalsNot set15 min
Change the singeraudio.song-voiceNot set15 min
Dialogueaudio.dialogueNot set10 min
Make it longeraudio.song-extendNot set15 min
Musicaudio.music360 s15 min
New backing musicaudio.song-backingNot set15 min
Replace a partaudio.song-replaceNot set15 min
Sound effectsaudio.sfxNot set10 min
Voice and music apartaudio.song-splitNot set15 min
Voiceoveraudio.ttsNot set10 min
Captionsedit.captionsNot set200 min
Merge clipsedit.mergeNot set20 min
Metadata Removeredit.cleanNot set10 min
Resizeedit.resizeNot set15 min
Flux 2 Proimage.flux-2-proNot set15 min
GPT Image 2image.gpt-image-2Not set20 min
GPT Image 2.5image.gpt-image-2-5Not set20 min
Image upscaleimage.upscaleNot set10 min
Nano Banana 2image.nano-banana-2Not set20 min
Nano Banana Proimage.nano-banana-proNot set20 min
Remove backgroundimage.remove-bgNot set10 min
Seedream 5 Flashimage.seedream-5-flashNot set15 min
AI video editvideo.video-editNot set45 min
Hailuo 2.3video.hailuo-2-310 s30 min
InfiniTalkvideo.infinitalkNot set40 min
Kling 3.0video.kling-315 s40 min
Kling 3.0 Motion Controlvideo.kling-3-motionNot set40 min
Kling 4.0video.kling-430 s40 min
Kling Avatarvideo.kling-avatarNot set60 min
Lip sync re-dubvideo.lipsyncNot set40 min
Seedance 2.0video.seedance-215 s45 min
Seedance 2.0 Minivideo.seedance-2-mini15 s30 min
Seedance 2.5video.seedance-2-530 s45 min
Seedance 2.5 final in 1080pvideo.seedance-2-5-finalNot set45 min
Veo 3.1video.veo-3-18 s45 min
Video upscalevideo.upscaleNot set60 min
Wan 2.2 Animatevideo.wan-animateNot set45 min
Wan 3.0video.wan-330 s40 min
Wan 3.0 Primevideo.wan-3-prime30 s40 min

Longest clip is the largest duration the model's fields accept, where it has one.

Webhooks

  • Each try has 10 seconds, the name lookup included, to get an answer from your endpoint.
  • 6 tries in all. After a failed try we wait 1 min, 5 min, 30 min, 2 h, 6 h before the next one.
  • Any 2xx answer counts as delivered. Redirects are not followed. Only https links to public addresses, on port 443 / 8443.
  • At most 1 try in flight at a time per account and 1 per receiving host. A try ends after 10 seconds, the name lookup included, so a slow endpoint holds one of the 4 senders that every customer shares for at most that long per try.
  • Up to 100 webhooks waiting per account (not finished yet, or between tries). Past that, a call with callback_url is refused with 429 webhook_queue_full before anything is charged.
  • A job, run or studio click that is still not finished 14 days after you sent callback_url is given up: the delivery shows failed (not_finished).
  • Bodies over 256 KB are sent with object set to null; GET the url in the body for the whole object.
  • Your last 50 deliveries are listed in the workspace; finished ones are kept 30 days.
  • After a roll, each old secret also signs for 24 hours from its own roll, up to 5 old secrets at once.
How webhooks work