Sora 2 Video Generation

Complete guide to OpenAI Sora 2 / Sora 2 Pro official-relay video generation: text-to-video and image-to-video on a unified async endpoint, supporting 4 / 8 / 12 second clips at 720p / 1024p / 1080p.

Overview

Sora 2 is OpenAI’s flagship video generation series, producing 4–12 second high-fidelity clips with synchronized audio from text prompts or reference images. APIYI provides a transparent proxy (official-relay) channel that forwards requests directly to OpenAI’s /v1/videos endpoint with identical request and response semantics.

🎬 Highlights: Transparent proxy to the official OpenAI API, synchronized audio + video output, flexible 4 / 8 / 12 second durations, and three resolution tiers — Standard (720p), HD (1024p), and Full HD (1080p, Pro only). Suited for ad shorts, e-commerce assets, social-media clips, and product demos where precise instruction following and consistent quality matter.

Text-to-Video API

POST /v1/videos, generate video from text only — JSON request body, the simplest entry point.

Image-to-Video API

POST /v1/videos + multipart upload of input_reference to animate a static image into a clip.

Visual API Testing

Debug this endpoint directly in the iCover visual testing tool — no code required.

Async Task Lookup / Download

View submitted video tasks and download video links in the APIYI console — a lookup entry outside the API.

Why APIYI’s Sora 2 Official Relay?

Drop-in replacement for the OpenAI official channel, optimized for production scenarios across stability, integration friction, and cost:

Direct Official Connection · 99.99% Uptime

Transparent forward to OpenAI’s official /v1/videos — no intermediate processing, no protocol bypass risk. Request and response behavior matches the upstream exactly. No need to manage OpenAI account tiers or risk-control fluctuations.

Unlimited Concurrency · Production Scale

Linearly scale batch shoots, ad pipelines, and high-volume asset production — no per-account tier ceilings. Default capacity is production-ready; contact us for custom resource pools.

Same Per-Second Pricing + Top-Up Bonuses

Identical per-second rates to OpenAI official, stacked with top-up bonuses for further savings. Failed tasks are not billed.

Global Zero-Friction Access

No overseas server or proxy required — connect to api.apiyi.com directly from Mainland China data centers, residential networks, or overseas nodes. Skip the OpenAI cross-border setup entirely.

OpenAI-Compatible · Zero Code Changes

Endpoint path /v1/videos matches OpenAI exactly. Point the official OpenAI SDK to APIYI’s base_url and call as-is — parameter and field names align one-to-one.

Professional Support · Enterprise Onboarding

Our team has deep expertise in video generation: prompt engineering, resolution selection, batch production, and post-processing. Full PoC-to-production technical support for enterprise customers.

Key Features

Synchronized Audio + Video

Sora 2 natively outputs video with synchronized audio tracks (ambient sound, dialogue, score) — no separate audio post-production needed.

Multi-Resolution Tiers

sora-2 supports 720p (720x1280 / 1280x720); sora-2-pro adds 1024p and 1080p tiers up to 1920x1080.

Flexible 4 / 8 / 12 Second Durations

Per-second billing means you pay for exactly what you produce. 8 seconds is the most common tier — balancing visual continuity and cost.

Precise Instruction Following

Sora 2 leads its tier on camera motion, object physics, and character expression fidelity — closer to your prompt intent than competitors.

Image-to-Video (input_reference)

Upload one image as the starting frame to animate static visuals. See Image-to-Video.

Async Task Model

Submit returns a video_id immediately. Poll status independently and download the final video — ideal for batch management and resume-on-failure flows.

OpenAI SDK Drop-In

base_url=https://api.apiyi.com/v1 works as a drop-in replacement for the official OpenAI SDK.

Failures Are Free

In async mode, failed generations, content-policy rejections, and capacity errors are not billed.

Pricing

Billed by video duration in seconds, identical to OpenAI official rates. sora-2-pro has three resolution tiers, each with distinct per-second pricing.

sora-2 (Standard)

Resolution	Rate	4 sec	8 sec	12 sec
720x1280 / 1280x720	$0.10/sec	$0.40	$0.80	$1.20

sora-2-pro (Pro)

Resolution	Rate	4 sec	8 sec	12 sec
720x1280 / 1280x720	$0.30/sec	$1.20	$2.40	$3.60
1024x1792 / 1792x1024	$0.50/sec	$2.00	$4.00	$6.00
1080x1920 / 1920x1080	$0.70/sec	$2.80	$5.60	$8.40

Billing notes:

• Charged by actual generated seconds (seconds × rate), independent of prompt length or whether input_reference is provided
• In async mode, failed generations / content-policy rejections / capacity errors are all not billed
• Requests must use the usage-based billing mode in your APIYI console (switch under API Key settings); per-request billing groups cannot route to the official-relay channel
• Top-up bonus tiers are listed under Top-Up Promotions

Group Setup

Sora 2 official relay routes through the dedicated Sora2Official group (1x). Two conditions must be met for a token to reach the channel:

1. Billing mode: pick Usage-Based Priority (pay-as-you-go) — per-request billing tokens cannot route to the official relay
2. Group: must include Sora2Official

Token creation: pick Usage-Based Priority for the billing mode and select Sora2Official under groups to call sora-2 / sora-2-pro

Two recommended setups — pick whichever fits your isolation needs:

Setup	When to use	How
A. Single token, multi-purpose	Same token also calls other models (GPT / Claude / image / …)	Keep your usual Default (or other) as the primary group, set Sora2Official as fallback — non-Sora requests stay on primary, while sora-2 / sora-2-pro calls auto-route to Sora2Official
B. Dedicated token	Video workloads kept separate, need isolated billing & quota controls	Create a new token with only the Sora2Official group selected, used exclusively for Sora 2

Recommend B (dedicated token) for production video workloads: cleaner billing, easier quota and alert management. A fits solo dev or low-frequency use.

Technical Specs

Dimension	sora-2	sora-2-pro
Model ID	sora-2	sora-2-pro
Current snapshot	sora-2-2025-12-08	Synced with alias
Deprecated snapshot	sora-2-2025-10-06	—
Supported resolutions	720x1280 / 1280x720	720x1280 / 1280x720 / 1024x1792 / 1792x1024 / 1080x1920 / 1920x1080
Supported durations (seconds)	4 / 8 / 12	4 / 8 / 12
Audio track	✅ Synchronized	✅ Synchronized
Image-to-video (input_reference)	✅	✅
Typical generation time	3–5 minutes	5–10 minutes
Video retention	1 day	1 day
Response fields	id / status / progress; download via /v1/videos/{id}/content	Same

API Endpoints

Endpoint	Method	Purpose	Content-Type
/v1/videos	POST	Submit a video generation task (text-to-video and image-to-video)	application/json or multipart/form-data
/v1/videos/{video_id}	GET	Poll task status and progress	—
/v1/videos/{video_id}/content	GET	Download the generated video file	—

Domain options: api.apiyi.com is the primary endpoint. vip.apiyi.com / b.apiyi.com are equivalent backup gateways with identical behavior.

Key Parameters

seconds (Video Duration)

Only three string-typed enum values (not numbers):

Value	Meaning	Use Case
"4"	4 seconds (default)	Short demos, single shots, quick prompt iteration
"8"	8 seconds	Standard short-form video, social-media clips
"12"	12 seconds	Long shots, continuous action, narrative sequences

seconds must be passed as a string "4" / "8" / "12". Passing the integer 4 or other values like "10" / "15" returns 400.

size (Output Resolution)

Supported tiers differ between sora-2 and sora-2-pro:

Tier	Pixels	sora-2	sora-2-pro
720p Portrait	720x1280	✅	✅ ($0.30/sec)
720p Landscape	1280x720	✅	✅ ($0.30/sec)
1024p Portrait	1024x1792	❌	✅ ($0.50/sec)
1024p Landscape	1792x1024	❌	✅ ($0.50/sec)
1080p Portrait	1080x1920	❌	✅ ($0.70/sec)
1080p Landscape	1920x1080	❌	✅ ($0.70/sec)

• Passing 1024p / 1080p sizes to sora-2 returns 400
• The actual rendered vertical pixel count for sora-2 720p videos is 704 (not 720) — this is OpenAI’s actual upstream behavior and does not affect display
• For image-to-video, the input_reference image dimensions must exactly match size, otherwise you get Inpaint image must match the requested width and height

Best Practices

1. Pick the model that matches your need

   • Cost-conscious → sora-2 (720p only, $0.10/sec, $0.40 for a 4-sec clip)
   • Need 1080p Full HD / strongest instruction following → sora-2-pro (up to $0.70/sec, supports 1920x1080)
   • Internal demos / first iterations → start with sora-2 4 seconds
2. Validate at 4 seconds before scaling duration

   Run each new prompt at seconds: "4" first to verify camera direction, style, and overall composition (~3 minutes, $0.40). Only scale to 8 / 12 seconds once the look is locked.

3. Switch to usage-based billing first

   In the APIYI console, set your API Key to usage-based billing and the Sora2官转 (Sora2 Official) group. Per-request billing groups cannot route to the official-relay channel.

4. Use async polling, not synchronous waits

   The official-relay channel is async-only: POST to submit and get a video_id, poll /v1/videos/{id} every 10–30 seconds until status: "completed", then download from /v1/videos/{id}/content.

5. Set client timeout to 30+ seconds

   The POST itself just enqueues the task and does not block on generation. With multipart input_reference uploads, large images extend connection time — start with a 30-second timeout.

6. Download videos immediately

   Videos are retained on OpenAI servers for 1 day only; after that, /content returns 404. Production flows must persist to your own OSS / CDN as soon as status: "completed".

7. Match resolution before image-to-video upload

   When uploading input_reference, pre-crop your image with ffmpeg / Pillow to the exact target size (e.g. 1280x720) to avoid 400 errors.

Error Codes & Retries

Status	Meaning	Recommended Action
400	Invalid parameters (seconds not in 4/8/12, size not supported, input_reference dimensions mismatch)	Validate parameters; pre-crop reference images to the target resolution
401	Invalid token	Check your Bearer Token and group setting (must be Sora2官转)
403	Content-policy rejection / wrong billing mode	Adjust prompt; confirm API Key is on usage-based billing
429	Rate limit / insufficient balance	Exponential backoff; usable immediately after top-up
5xx	Gateway / upstream error	Retry the async task 1–2 times (no charge)
Task failed	Generation failed (mostly content policy or upstream capacity)	Adjust prompt and retry; the failed task is not billed

Recommended client config:

• POST submission timeout: 30 seconds (longer for multipart uploads)
• GET polling interval: 10–30 seconds, max wait 15 minutes (Pro 1080p 12 seconds may take 8–10 minutes)
• Exponential backoff retries on 5xx and failed tasks (recommend 2 retries)
• Log the x-request-id response header for debugging

FAQ

Official-relay vs reverse-engineered — what's the difference, and is reverse-engineered still available?

Official-relay (this page): Direct forward to OpenAI’s /v1/videos, with request/response fields matching upstream. Per-second billing, 99.99% uptime, requires usage-based billing group.

Reverse-engineered: A reverse-engineered Sora 2 interface, billed per-request and cheaper, but subject to OpenAI risk control. As of January 2026 OpenAI policy adjustment, free accounts were disabled and APIYI now offers only the official-relay channel. Contact sales for special needs.

Why must I switch to usage-based billing?

The official-relay channel settles by actual OpenAI seconds, which is a different billing dimension from per-request. In the APIYI console, switch your API Key to usage-based billing + Sora2官转 group to access this route — per-request groups will get 403.

Why is async-only? Is there a sync streaming option?

The official /v1/videos endpoint is itself async task-based — there’s no SSE or WebSocket streaming. Generating a 4-second clip typically takes 3–5 minutes; 12 seconds can take 8–10 minutes. Synchronous waits would hang HTTP connections for too long and become unreliable. Always use the POST → poll → download flow.

Which seconds values are supported? Why can't I pass 10 / 15?

OpenAI officially exposes only "4" / "8" / "12" as enum string values. The 10 / 15 second values were unofficial durations from the older reverse-engineered channel and are not supported by the official relay. If your code passes "10", change it to "8" or "12".

Is the \$0.70/sec for sora-2-pro 1080p new?

Yes. OpenAI recently extended sora-2-pro to include 1080x1920 / 1920x1080 Full HD at $0.70/sec. The prior 720p ($0.30) and 1024p ($0.50) tiers are unchanged. Our pricing table above reflects the latest official rates.

How long are videos retained?

Videos are stored on OpenAI servers for 1 day only. After expiration, /v1/videos/{id}/content returns 404 / 410. Production flows must download and persist to your own OSS / CDN immediately after status: "completed".

Are failed generations billed?

No. Tasks that end up as failed, content-policy rejections, capacity errors, and parameter errors are all not billed. Only tasks that actually complete (status: "completed") and produce a video file are billed by seconds rate.

Can I use the official OpenAI SDK directly?

Yes. OpenAI Python SDK 1.50+ supports the videos namespace. Point base_url at https://api.apiyi.com/v1:

from openai import OpenAI

client = OpenAI(api_key="sk-your-key", base_url="https://api.apiyi.com/v1")
video = client.videos.create(
    model="sora-2",
    prompt="A golden retriever running on the beach at sunset",
    seconds="8",
    size="1280x720"
)
print(video.id, video.status)

Does input_reference accept base64?

No. input_reference is a multipart/form-data file upload field (accepts image/jpeg / image/png / image/webp) and requires a multipart request. If your image is base64, decode and write to a temp file first. See Image-to-Video.

Can I disable the audio track?

Not currently. Sora 2 / Pro outputs synchronized audio (ambient sound, dialogue, score) by default, and OpenAI does not expose a parameter to disable it. For audio-free output, strip with ffmpeg -an after download.

Can I cancel a running task?

No. The official /v1/videos endpoint does not provide a cancel operation — once submitted, a task runs to completion. Validate prompts at seconds: "4" first to avoid wasting long-duration runs.

What are the rate limits?

Subject to upstream OpenAI account tier limits, but pooled through APIYI’s gateway there is no obvious bottleneck for typical use. Enterprise batch needs (>10 concurrent, >100 clips per day) should contact sales for a dedicated resource pool.

Can I run multiple tasks in parallel?

Yes. Each POST /v1/videos returns an independent video_id. Submit and poll in parallel; manage your video_id list in a task queue to avoid polling storms.

Related Docs

• Text-to-Video Playground — POST /v1/videos (JSON) interactive debugger with 5 language samples
• Image-to-Video Playground — POST /v1/videos (multipart) + input_reference walkthrough
• Top-Up Promotions — Bonus tiers and applicable channels
• API Manual — General request, timeout, and retry guidance
• OpenAI official model page: platform.openai.com/docs/models/sora-2
• OpenAI official API reference: platform.openai.com/docs/api-reference/videos/create

Sora 2 on APIYI is delivered through an authorized Plus-tier account pool for stable official-relay service. Response fields, error codes, and billing dimensions match OpenAI exactly for drop-in compatibility with existing code. Open a ticket from your console for any feedback.