Generate videos

Video generation is asynchronous: submit returns a jobId immediately, then poll /v1/jobs/{jobId} until the job completes.

1. Submit

curl https://api.rlay.ai/v1/videos \
  -H "Authorization: Bearer $RLAY_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "veo-3.1",
    "prompt": "a cat surfing on neon waves",
    "duration": 5,
    "resolution": "720p"
  }'

Response:

{ "jobId": "job_a3f7...", "status": "pending" }

2. Poll

curl https://api.rlay.ai/v1/jobs/job_a3f7... \
  -H "Authorization: Bearer $RLAY_KEY"

Response while running:

{ "jobId": "job_a3f7...", "status": "processing" }

Response when done:

{
  "jobId": "job_a3f7...",
  "status": "completed",
  "resultUrl": "https://cdn.rlay.ai/...",
  "creditsCost": 1.15,
  "completedAt": "2026-05-05T12:42:10Z"
}

Polling cadence

First poll after 5 seconds, then exponential backoff capped at 30s.
Most jobs settle within 20-90s depending on duration and model.
A 4xx response means the job failed — check errorMsg.

Available models

Model	Best for	Price
`veo-3.1`	Cinematic, 720p, 5-12s	$0.23/s
`sora-2`	Long-form coherent, 1080p, 20s+	$0.20/s
`kling-3`	Camera control, realistic, 10s	$0.12/s
`seedance-2`	Image-to-video, fast, 5s	$0.08/s

Credits and refunds

Credits are only deducted on successful completion. Failed jobs cost nothing.

Error codes→

Handle failures gracefully

Migrate from fal.ai→

Map fal endpoints to rlay