Oxygen Video Generation

AZ8 Oxygen (oxygen-1.0) video generation guide: OpenAI Videos-compatible API covering text, first-frame, first-and-last-frame, and reference image/video/audio video, 480p / 768p, 4–15 s, billed at $0.02 per second.

Overview

Oxygen is a video generation model provided by AZ8, an AI video creation platform based in Singapore (formerly Videoinu). APIYI serves it as oxygen-1.0 through an OpenAI Videos-compatible API (POST /v1/videos to submit, GET /v1/videos/{id} to query), with 4–15 second clips at 480p / 768p, billed at $0.02 per second regardless of resolution.

Highlights: one model covers text-to-video, first-frame video, first-and-last-frame video, and reference generation with up to 9 images, 3 videos, and 3 audio clips. Output comes with an audio track. $0.02 per second: a 5-second clip costs $0.10 and a 15-second clip $0.30, and failed tasks are refunded automatically. Built for high-volume, low-cost video generation.

Video Generation API Reference

Submit, poll, and download, with Python / cURL / Node.js examples and a live Playground

Top-up Bonuses

Top-up bonuses lower the effective price further

Let an AI Agent Integrate It for You

Copy this prompt and paste it into Codex / Claude Code / Cursor

The agent first fetches the plain-text version of this page (append .md to any docs URL), then writes code for your stack. The common mistakes are already spelled out: always pass size, put advanced parameters in the input_reference JSON envelope, and set the length only with seconds.

Integrate or debug Oxygen (oxygen-1.0) video generation in this project (text-to-video / first frame / first and last frame / reference media).

Read the docs before writing code: fetch https://docs.apiyi.com/en/api-capabilities/oxygen/overview.md for the plain-text version of this page; parameters and code samples are at https://docs.apiyi. …

1. Endpoints: OpenAI Videos format. Submit with POST https://api.apiyi.com/v1/videos (JSON), query with GET https://api.apiyi.com/v1/videos/{id}. Subm …

10 requirements in total. Use the button below to copy the full prompt.

CopiedCopy prompt

What this prompt protects you from

Requirement	Mistake it prevents
Always pass size	Without it the gateway defaults to a portrait size, so a landscape request comes back portrait
Advanced parameters in the input_reference envelope	last_image, reference_images, or resolution at the top level are silently dropped with no error: the last frame is ignored, references are ignored, and the resolution is wrong
Length only via seconds	duration inside the envelope is rejected; billing and clip length both follow seconds
Download video_url	Calling /content right as the status turns completed may return 400 and look like a failure
Copy the file right away	The link expires after about 24 hours

Why APIYI’s Oxygen?

Volume pricing

$0.02 per second at any resolution; $0.08 for 4 seconds, good for batch output and A/B drafts

Automatic refunds

Failed tasks are refunded in full and rejected submissions are free, so you only pay for videos you get

OpenAI Videos compatible

Same submit-and-query pattern as /v1/videos, so existing Sora-style code needs few changes

Top-up bonuses stack

Combine with Top-up Bonuses for a lower effective cost

Full video model lineup

The same key also calls Seedance 2.0 / 2.5, MiniMax-H3, Wan2.7, and more

Global access

Connect directly to api.apiyi.com with one API key, no overseas account needed

Key Features

Four generation modes

Text, first frame, first and last frame, and reference image / video / audio, all on one endpoint

Two resolutions

480p / 768p at the same price; trade speed for detail as needed

Any whole length from 4 to 15 s

Billed by the requested seconds, so short clips cost less

Built-in audio

The output MP4 includes an audio track, no separate dubbing needed

Pricing

Item	Price
Output video (480p / 768p, same price)	$0.02 / second
First frame, last frame, reference image / video / audio	No extra charge
Example: 4-second clip	$0.08
Example: 15-second clip	$0.30

Prices may change; the table above is for reference only, and the Model Pricing tab in the top navigation is authoritative: Model Pricing.

Billing notes:

• Billed by the requested seconds, charged when the task is accepted; the actual clip runs slightly longer (about 4.5 s for a 4 s request) at no extra cost
• Resolution, aspect ratio, quality tier (quality), prompt enhancement (prompt_enhance), and reference media do not affect the price
• Failed tasks (provider failure, timeout, etc.) are refunded in full automatically
• Submissions that return 400 are not charged; queries and downloads are free

Group Setup

oxygen-1.0 works in the default group, and the svip group works too. Set the token’s billing mode to Pay-as-you-go Priority. If a call returns “no available channel in the current group”, the token’s group does not include this model or the model name is misspelled.

Item	Requirement
Group	default or svip
Billing mode	Pay-as-you-go Priority
Model name	oxygen-1.0 (the -1.0 suffix is required)

Technical Specs

Item	Spec
Model ID	oxygen-1.0
Length	Integer 4–15 seconds (top-level seconds)
Resolution	480p (default) / 768p (320p retired on 2026-10-14)
Quality tier	quality: high (default) / normal
Prompt enhancement	prompt_enhance: refine (default) / format / none
Aspect ratio	Landscape 16:9, portrait 9:16, square 1:1; image-to-video follows the first frame
Output size (measured)	480p: 864×480 / 480×864 / 480×480; 768p: 1376×768 / 768×1376 / 768×768
Audio	Output includes an audio track
Image input	Public https URL or image data URI
Reference media	Images up to 9, videos up to 3, audio up to 3 (video and audio as https URLs only)
Output	MP4 via video_url in the query response, valid for about 24 hours
Generation time	Usually 1–3 minutes; 5+ minutes when the queue is busy

API Endpoints

Purpose	Method	Path
Create task	POST	/v1/videos
Query task	GET	/v1/videos/{id}
Download (optional)	GET	/v1/videos/{id}/content

Primary domain https://api.apiyi.com, backup domain https://b.apiyi.com, same paths. For downloads, use video_url from the query response directly.

Generation Modes

Only five top-level fields take effect: model, prompt, seconds, size, and input_reference. First and last frames, reference media, 1:1, and the quality tier are advanced parameters that go into the input_reference JSON envelope (a JSON string starting with {):

Mode	How to write it	Resolution / aspect ratio
Text-to-video	Omit input_reference	Set by size
First-frame video	input_reference = image URL or data URI	Resolution from size, aspect ratio follows the first frame
First-and-last-frame video	Envelope {"images":["first"],"last_image":"last"}	Same as above
Reference media video	Envelope {"reference_images":[...],"reference_videos":[...],"reference_audios":[...]}	Set by size; aspect_ratio can go in the envelope
Resolution or 1:1	Envelope {"resolution":"768p","aspect_ratio":"1:1"} (can be combined with any mode above)	Envelope wins over size
Quality tier / prompt enhancement	Envelope {"quality":"normal","prompt_enhance":"none"} (can be combined with any mode above)	No effect on resolution or price

How size maps to resolution:

size	Resolution	Orientation
1280x720	480p	Landscape 16:9
720x1280	480p	Portrait 9:16
1792x1024	768p	Landscape 16:9
1024x1792	768p	Portrait 9:16

Envelope example (first and last frame):

{
  "model": "oxygen-1.0",
  "prompt": "The glass sphere slowly dissolves into a deep blue abstract wave",
  "seconds": "5",
  "size": "1280x720",
  "input_reference": "{\"images\":[\"https://your-cdn.example.com/first.png\"],\"last_image\":\"https://your-cdn.example.com/last.png\"}"
}

• last_image, reference_images, resolution, aspect_ratio, and similar fields are silently dropped when placed at the top level: no error and normal billing, but the last frame is ignored, references are ignored, and resolution follows size. Always put them in the input_reference envelope
• duration is not allowed in the envelope; use top-level seconds. Invalid JSON or misspelled keys return 400 (param: input_reference) with no charge
• First/last frames cannot be mixed with reference media
• input_reference must be a string: serialize the envelope first (json.dumps in Python, JSON.stringify in JS). Passing an object or array directly is rejected

Best Practices

1. Test with 4 seconds first

   Billing is per second, so confirm composition and style at 4 seconds before rendering 10–15 second finals

2. Always set size

   Landscape 1280x720, portrait 720x1280; for more detail use 1792x1024 / 1024x1792 (768p)

3. Use first and last frames with similar aspect ratios

   The output follows the first frame’s ratio; a last frame with a very different ratio gets cropped in the transition

4. Host media on stable public storage

   Use your own OSS / CDN direct links to avoid hotlink protection or expired signatures breaking the download

5. Poll every 5 seconds with a 15-minute timeout

   Most clips finish in 1–3 minutes; it can take longer at peak times

6. Copy video_url to your storage right away

   The link expires after about 24 hours; download it and serve it from your own storage

Error Codes & Retries

Stage	Symptom	Cause	Fix
Submit	400, invalid_params, param: input.duration	seconds outside 4–15	Change the length; not charged
Submit	400, invalid_params, param: input_reference	Invalid envelope JSON, unknown key, or duration in the envelope	Fix the envelope per the message; not charged
Submit	400, invalid_params, param: resolution / quality / prompt_enhance	Value outside the allowed set (e.g. 320p)	Use an allowed value; not charged
Submit	400, cannot unmarshal array ... input_reference	input_reference sent as an array or object instead of a string	Serialize it with json.dumps / JSON.stringify
Submit	500, reference file download failed	The image URL in input_reference is not reachable	Use a publicly reachable https URL
Submit	503, no available channel	Token group does not include this model, or the model name is misspelled	Check the group and oxygen-1.0
Run	failed, upstream_error	Occasional provider-side failure	Resubmit after a few minutes (already refunded)
Run	failed, upstream_timeout	Queued too long upstream and hit the time limit	Resubmit later (already refunded)

The error detail is a JSON string inside the response’s message field, e.g. {"message":"{\"error\":{\"code\":\"invalid_params\",...}}","type":"task_error"}, so parse it a second time.

FAQ

Why did my landscape video come out portrait?

size was not passed. Without it the gateway defaults to 720x1280 (portrait). For landscape, pass 1280x720 or 1792x1024 explicitly.

Why do last_image / reference_images have no effect?

They were placed at the top level of the request. Only model, prompt, seconds, size, and input_reference take effect there; everything else is silently dropped. Put them in the input_reference JSON envelope, see “Generation Modes” above.

Can I still use 320p?

No. AZ8 retired the 320p tier on 2026-10-14. Sending "resolution":"320p" in the envelope returns 400 at submission (must be one of 480p/768p) and is not charged. Use 480p instead.

Passing resolution does nothing?

A top-level resolution is dropped. Put it in the envelope: "input_reference": "{\"resolution\":\"768p\"}", or pick it with size (1792x1024 / 1024x1792 = 768p). Both resolutions cost the same.

Do the resolutions cost different amounts?

No, both are $0.02 per second. 480p renders faster with smaller files; 768p is sharper.

What are quality and prompt_enhance? Do I need them?

Usually not. quality is the quality tier (high / normal); prompt_enhance controls whether the model rewrites your prompt: refine polishes and expands it, format only tidies the formatting, none uses it as written. The defaults, high + refine, match the behavior before 2026-10-14. To have your prompt applied word for word, send "prompt_enhance":"none" in the envelope. Neither parameter affects the price.

The status says completed but /content returns 400?

Right after the status turns completed, /v1/videos/{id}/content may need a few more seconds. Use video_url from the query response instead.

How long does the video link last?

About 24 hours (see expires_at in the query response). Download and store it promptly.

Am I charged for failed tasks?

No. A task that ends failed is refunded in full automatically, and submissions that return 400 are not charged.

What should I do about an occasional upstream_error?

It is an occasional provider-side failure and has already been refunded. Resubmitting after a few minutes usually works.

Why is the clip slightly longer than seconds?

Output runs a little longer (about 4.5 s for 4 s, 5.2 s for 5 s). Billing uses the requested seconds, so there is no extra charge.

Can the first frame be Base64?

Yes. Both input_reference and images in the envelope accept image data URIs (such as data:image/jpeg;base64,...). Reference video and audio accept https URLs only.

Can I set the aspect ratio for image-to-video?

No. Image-to-video follows the first frame’s aspect ratio and ignores aspect_ratio. You can still pick the resolution via size or resolution in the envelope.

Related Docs

• Oxygen Video Generation API Reference
• MiniMax-H3 Video Generation
• Seedance 2.0 / 2.5 Video Generation
• Wan2.7 Video Generation
• Top-up Bonuses