Video generation

/v1/videos/generations turns a still image into a short video clip (image-to-video) with Wan2.2. Because video generation takes minutes, this is an async, two-call API: you submit a job, then poll it until it’s done.

Video is image-to-video — every request needs a source image (image as base64/data URL, or image_url). The result is a base64 MP4.

Flow

Submit a job

POST /v1/videos/generations → returns a job id with status queued (HTTP 202). This places a credit hold but doesn’t charge yet.

from openai import OpenAI
import base64, time, httpx
 
BASE = "https://api.crustoff.app/v1"
KEY = "sk-crustoff-..."
headers = {"Authorization": f"Bearer {KEY}"}
 
with open("start.png", "rb") as f:
    image_b64 = "data:image/png;base64," + base64.b64encode(f.read()).decode()
 
job = httpx.post(f"{BASE}/videos/generations", headers=headers, json={
    "model": "wan2.2-i2v",
    "prompt": "the cat turns its head and blinks, cinematic",
    "image": image_b64,
    "seconds": 5,
    "size": "480x832",
}).json()
job_id = job["id"]

Poll until complete

GET /v1/videos/generations/{id} → processing while it runs, then completed with the video.

while True:
    res = httpx.get(f"{BASE}/videos/generations/{job_id}", headers=headers).json()
    if res["status"] == "completed":
        break
    if res["status"] == "failed":
        raise RuntimeError("generation failed")
    time.sleep(5)
 
with open("out.mp4", "wb") as f:
    f.write(base64.b64decode(res["data"][0]["b64_json"]))

Submit parameters

Field	Type	Default	Notes
`model`	string	—	A `video` model id (e.g. `wan2.2-i2v`)
`prompt`	string	—	Required — describes the motion
`image`	string	—	Source frame as base64 / data URL (required unless `image_url`)
`image_url`	string	—	Source frame by URL (alternative to `image`)
`seconds`	int	`5`	Output length, `1`–`10` (the billed unit)
`size`	string	`"480x832"`	One of `480x832`, `832x480`, `624x624`
`negative_prompt`	string	—	What to avoid
`seed`	int	random	Reproducibility
`steps`	int	model default	Denoising steps

Responses

Submit (202):

{ "id": "req_abc123", "object": "video.generation", "status": "queued", "model": "wan2.2-i2v", "seconds": 5 }

Poll while running:

{ "id": "req_abc123", "object": "video.generation", "status": "processing", "model": "wan2.2-i2v" }

Poll when done:

{
  "id": "req_abc123",
  "object": "video.generation",
  "status": "completed",
  "model": "wan2.2-i2v",
  "seconds": 5,
  "data": [{ "b64_json": "<base64 mp4>", "format": "mp4" }]
}

⚠️

Generation typically takes 1–3 minutes. Poll every few seconds. The finished video is retained for a limited window (~30 minutes) — download it promptly. If a job fails on our side you are not charged; the credit hold is released.

Billing

Video is billed per second of output, charged once the job completes. A 5-second clip on wan2.2-i2v ($0.030/sec) costs $0.15. See Pricing & billing.

Image generation Pricing & billing