Video Generation

​

Overview

​

Usage

from openai import OpenAI

client = OpenAI(base_url="https://api.edenai.run/v3", api_key="YOUR_API_KEY")

video = client.videos.create_and_poll(
    model="pruna/p-video",
    prompt="A red paper boat drifting on a calm pond at sunrise",
    seconds="5",
    size="1280x720",
)
assert video.status == "completed"
client.videos.download_content(video.id).write_to_file("boat.mp4")

​

Image-to-Video

with open("keyframe.png", "rb") as f:
    video = client.videos.create(
        model="pruna/p-video", 
        prompt="the boat sails away", 
        input_reference=f
    )

curl -X POST "https://api.edenai.run/v3/videos" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "pruna/p-video", 
    "prompt": "the boat sails away",
    "input_reference": {"image_url": "https://example.com/keyframe.png"}
  }'

​

Request fields

​

Video object

{
  "id": "e5eef4b4-acd1-4eb9-a5c9-2769271e42f5",
  "object": "video",
  "status": "queued",
  "progress": 0,
  "created_at": 1757435495,
  "completed_at": null,
  "expires_at": null,
  "model": "pruna/p-video",
  "seconds": "5",
  "size": "1280x720",
  "remixed_from_video_id": null,
  "error": null,
  "provider": "pruna",
  "cost": 0
}

• status: queued on the create response, in_progress on later reads while the job runs, then completed or failed. progress goes from 0 to 100.
• provider and cost (USD) are Eden extensions. cost is 0 while the job is queued or in progress, and updates to the settled amount once the job completes or fails.
• id is the universal-ai job id: the same job is visible at GET /v3/universal-ai/async/{id}.

​

Limitations

• Response size: size is only echoed on the create response; reads return null (the job row does not store the request). seconds on reads is the duration the provider actually rendered.
• Unsupported Routes: No remix, extensions, edits or characters: these OpenAI routes answer 400.
• Keyset Pagination: after is a keyset cursor on (created_at, id): pass the previous page’s last_id to get the jobs that follow it in newest-first order, rather than an offset-style page number.
• Only the video content variant (no thumbnails / spritesheets).
• The content redirect points at a signed link valid for 7 days; after that the endpoint serves the stored bytes directly.
• Authentication errors (401 for a bad token, 403 when the header is missing), rate-limit (429) and query-parameter validation errors (e.g. ?limit=0) use the platform’s {"detail": ...} shape rather than the OpenAI error envelope.

​

Pricing