Video model guide

Plan a Chinese video model API workflow for Kling or Seedance.

Video models are not just longer text completions. Teams need to plan async jobs, media storage, prompt inputs, duration, resolution, callbacks and cost controls before exposing them in a product.

Status: Kling, Seedance and related Chinese video model routes are handled as early-access workflows in Black Eagle AI. Use this guide to prepare a request; do not assume instant self-serve video API access yet.

1. Choose the workflow before the model

Most video API requests fall into one of these patterns:

Text to video

Generate a clip from prompt, style, duration and aspect ratio.

Image to video

Animate a product image, character, scene or ad creative.

Batch creative pipeline

Generate many variants for ads, social posts or workflow automation.

2. Use an async job pattern

Video generation can take much longer than text generation. Product integrations should treat video calls as jobs, not synchronous chat completions.

POST /video/jobs
{
  "model": "kling-or-seedance-route",
  "input": {
    "prompt": "A product demo video for a new tea drink",
    "image_url": "https://example.com/product.png",
    "duration_seconds": 5,
    "aspect_ratio": "9:16"
  },
  "callback_url": "https://your-app.com/webhooks/video"
}

GET /video/jobs/{job_id}

3. Prepare the fields providers usually need

Field	Why it matters	Example
Model family	Kling, Seedance and other providers expose different controls and output styles.	Kling for cinematic motion; Seedance for short-form creative tests.
Input type	Text-to-video and image-to-video often use different endpoints or required fields.	Prompt only, product image URL, reference frame.
Duration and aspect ratio	Cost and queue time usually change with length and output format.	5s, 10s, 9:16, 16:9, 1:1.
Callback or polling	Your app needs a way to know when the video is ready or failed.	Webhook callback, polling every 10-30 seconds.
Storage and retention	Generated media may need to be copied into your own storage after completion.	S3, Cloudflare R2, Vercel Blob, GCS.

4. Estimate cost by job, not token

Text model pricing is usually token-based. Video model pricing is usually job-based and depends on provider, duration, resolution, concurrency and retry policy.

How many videos per day or month?
What is the average duration?
Do users need HD or draft-quality previews?
How many failed generations or retries are acceptable?
Do you need a queue, webhook and dashboard before launch?

5. What to send when requesting access

A useful request should include model family, use case, monthly volume, input type, desired duration, integration target and first-test budget.

Use the Black Eagle AI request template when you are ready to describe the workflow: request model access.

Request Kling or Seedance access

Start with a clear workload. If text access is enough for now, use the OpenAI-compatible integration recipes first.

Open request template