Oxygen Video Generation Is Live

​

Key Takeaways

• Launched jointly by APIYI and AZ8: Oxygen, AZ8’s in-house all-round video model, is live as oxygen-1.0 and works in the default group
• One model, four ways to generate: text-to-video, first-frame, first-and-last-frame, and reference generation with up to 9 images, 3 videos, and 3 audio clips; every clip comes with sound
• OpenAI Videos-compatible: submit with POST /v1/videos and poll with GET /v1/videos/{id}, so existing Sora-style integrations need only small changes
• Priced at $0.02 per second, same price at all three resolutions: $0.08 for 4 seconds, $0.30 for 15 seconds; reference media cost nothing extra, and failed tasks are refunded in full automatically
• Built for volume: batch production, A/B drafts, and short-form video assets

​

Background

​

In-Depth Analysis

​

Benchmarks

​

Key Features

​

Specifications

​

Two Things You Must Get Right

​

Practical Applications

​

Recommended Use Cases

• Batch short-form assets: animate product shots, feed ads, and social visuals for a few cents each
• Storyboard drafts and A/B tests: compare several creative directions quickly at 4 seconds and 320p, then render the final version
• First-and-last-frame transitions: supply the opening and closing frames and generate the motion in between
• Character-consistent series: lock people and props with multiple reference images, and pace the clip with reference audio

​

Code Example

import json
import time
import requests

API_KEY = "sk-your-apiyi-key"
BASE = "https://api.apiyi.com/v1"
HEADERS = {"Authorization": f"Bearer {API_KEY}", "Content-Type": "application/json"}

# First-and-last-frame video: advanced options go inside the input_reference JSON envelope
envelope = {
    "images": ["https://your-cdn.example.com/first.png"],
    "last_image": "https://your-cdn.example.com/last.png",
}
task = requests.post(f"{BASE}/videos", headers=HEADERS, timeout=60, json={
    "model": "oxygen-1.0",
    "prompt": "The glass sphere slowly dissolves into a deep blue abstract wave",
    "seconds": "5",            # integer 4–15, billed per second
    "size": "1280x720",        # always pass it; sets resolution and orientation
    "input_reference": json.dumps(envelope),  # must be a string
}).json()

while True:
    task = requests.get(f"{BASE}/videos/{task['id']}", headers=HEADERS, timeout=30).json()
    if task["status"] in ("completed", "failed"):
        break
    time.sleep(5)
print(task.get("video_url") or task.get("error"))

​

Best Practices

1. Test with 4 seconds first: billing is per second, so confirm composition and style before rendering 10–15 seconds
2. Use first and last frames with similar aspect ratios: the clip follows the first frame, and a very different last frame gets cropped
3. Host media on stable public storage: hotlink protection or expired signatures make the task fail while fetching media (failures are refunded automatically)
4. Download from video_url and store it right away: right after the status turns completed, /v1/videos/{id}/content may still return 400, and the link expires after about 24 hours
5. Resubmit occasional upstream_error failures after a few minutes: they are refunded automatically; for 400 errors, fix the parameters instead of retrying

​

Pricing and Availability

​

Pricing

• Billed by the requested seconds and pre-charged when the task is accepted; the small extra length of the clip is not charged
• Tasks that end as failed are refunded in full automatically; requests rejected with 400 at submission are not charged; polling and downloads are free
• Groups: default or svip; set the token billing mode to Pay-as-you-go Priority

​

Stack with Top-up Promotions

​

Summary and Recommendations