Completion: polling, not webhooks

There is no webhook_url parameter on POST /api/v1/generate today. When a video is ready, you learn it by polling GET /api/v1/status/{id} until the status is completed or failed.

Payment provider webhooks (Dodo, Mollie) exist for billing only. They do not notify you that a generation finished.

How long to wait

Model family	Typical time	Suggested polling
FLUX 3 Max	~12s	Every 3-5s
Seedance 2.0, Grok Imagine 1.5, LTX-2	~30s	Every 5-10s
Veo 3 Fast, Higgsfield, Wan 2.5	45-60s	Every 10s
MiniMax H3, Veo 2	90-110s	Every 15s
Seedance 2.5, Kling 3	150-250s	Every 15-20s

These are catalogue averages; queue load and longer durations add time. A 10-minute timeout covers every model. Status checks do not count against generation rate limits.

Pattern 1: wait inline (scripts, CLIs)

For a script or a one-off job, submit and poll in the same process. The status page has a complete Python function, and the example script does the full round trip.

Pattern 2: background worker (web apps)

In a web app, do not keep the user's HTTP request open for minutes. Store the generation_id, return immediately, and let a worker resolve pending jobs:

# 1. Request handler: submit and persist
gen_id = submit_generation(prompt)          # POST /api/v1/generate
db.insert(job_id=gen_id, status="in_progress", user_id=user.id)

# 2. Worker (cron every 15s, or a queue consumer)
for job in db.select(status="in_progress"):
    data = get_status(job.job_id)           # GET /api/v1/status/{id}
    if data["status"] in ("completed", "failed"):
        db.update(job.job_id, status=data["status"], video_url=data.get("video_url"))
        notify(job.user_id)                 # websocket, email, push...

Your own notify() step is effectively the webhook: your front end listens to your backend, and your backend polls VideoGenAPI.

Pattern 3: no-code (n8n, Make, Zapier)

Use an HTTP node to submit, a Wait node of 15-30 seconds, then a status request inside a loop that exits on completed or failed. The n8n guide shows the full workflow.

Good practice

• Persist the generation_id as soon as you receive it, so a restart does not lose a paid generation.
• Do not resubmit on a slow job. A job still in_progress after a few minutes is usually queued, not lost; resubmitting doubles your usage.
• Download the video to your own storage once completed if you need it long-term.
• Back off on errors. If the status call itself fails, wait longer before the next attempt (see error codes).