OpenAI's video generation model supporting text-to-video and image-to-video at 720p resolution with durations up to 20 seconds
Details
sora-2
Providers & Pricing (3)
Sora 2 is available from 3 providers, with per-video pricing starting at $0.1 through Replicate.
replicate/sora-2
openai/sora-2
fal/sora-2
Sora 2 API Async video generation
Integrate Sora 2 via Lumenfall’s OpenAI-compatible API to generate high-fidelity videos up to 20 seconds long with 720p resolution.
https://api.lumenfall.ai/v1
sora-2
Text to Video Generate
Submit a prompt, poll for the result, and download the video
# Step 1: Submit video generation requestVIDEO_ID=$(curl -s -X POST \ https://api.lumenfall.ai/v1/videos \ -H "Authorization: Bearer $LUMENFALL_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "sora-2", "prompt": "A serene mountain landscape at sunset", "size": "1024x1024" }' | jq -r '.id')echo "Video ID: $VIDEO_ID"# Step 2: Poll for completionwhile true; do RESULT=$(curl -s \ https://api.lumenfall.ai/v1/videos/$VIDEO_ID \ -H "Authorization: Bearer $LUMENFALL_API_KEY") STATUS=$(echo $RESULT | jq -r '.status') echo "Status: $STATUS" if [ "$STATUS" = "completed" ]; then echo $RESULT | jq -r '.output.url' break elif [ "$STATUS" = "failed" ]; then echo $RESULT | jq -r '.error.message' break fi sleep 5doneconst BASE_URL = 'https://api.lumenfall.ai/v1';const API_KEY = 'YOUR_API_KEY';// Step 1: Submit video generation requestconst submitRes = await fetch(`${BASE_URL}/videos`, { method: 'POST', headers: { 'Authorization': `Bearer ${API_KEY}`, 'Content-Type': 'application/json' }, body: JSON.stringify({ model: 'sora-2', prompt: 'A serene mountain landscape at sunset', size: '1024x1024' })});const { id: videoId } = await submitRes.json();console.log('Video ID:', videoId);// Step 2: Poll for completionwhile (true) { const pollRes = await fetch(`${BASE_URL}/videos/${videoId}`, { headers: { 'Authorization': `Bearer ${API_KEY}` } }); const result = await pollRes.json(); if (result.status === 'completed') { console.log('Video URL:', result.output.url); break; } else if (result.status === 'failed') { console.error('Error:', result.error.message); break; } await new Promise(r => setTimeout(r, 5000));}import requestsimport timeBASE_URL = "https://api.lumenfall.ai/v1"API_KEY = "YOUR_API_KEY"HEADERS = { "Authorization": f"Bearer {API_KEY}", "Content-Type": "application/json"}# Step 1: Submit video generation requestresponse = requests.post( f"{BASE_URL}/videos", headers=HEADERS, json={ "model": "sora-2", "prompt": "A serene mountain landscape at sunset", "size": "1024x1024" })video_id = response.json()["id"]print(f"Video ID: {video_id}")# Step 2: Poll for completionwhile True: result = requests.get( f"{BASE_URL}/videos/{video_id}", headers=HEADERS ).json() if result["status"] == "completed": print(f"Video URL: {result['output']['url']}") break elif result["status"] == "failed": print(f"Error: {result['error']['message']}") break time.sleep(5)Size, Aspect Ratio & Resolution Reference
Three optional parameters for controlling output dimensions
size
Exact pixel dimensions
"1920x1080"
aspect_ratio
Shape only, default scale
"16:9"
resolution
Scale tier, preserves shape
"1K"
Priority when combined
size is most specific and always wins. aspect_ratio and resolution control shape and scale independently.
How matching works
7:1 on a model with
4:1 and 8:1,
you get 8:1.
0.5K 1K 2K 4K)
or megapixel tiers (0.25 1).
If the exact tier isn't available, you get the nearest one.
Parameters & Unified Output Reference
These work the same regardless of which provider runs your request
response_format
url or
b64_json.
If you ask for a URL but the provider returns base64, we store it temporarily and hand you a link valid for 60 minutes.
output_format
Pick from png
jpeg
gif
webp
avif.
We convert if the provider generates a different format.
output_compression
Quality level from 1 to
100 for lossy formats (jpeg, webp, avif).
Higher means better quality, larger file.
n
Request multiple images in one call. If the provider caps at 1, we run parallel requests behind the scenes and return them together.
Parameter support
Not every provider speaks the same language. When you send a parameter, we handle it in one of four ways depending on what the model supports:
| Mode | What happens | Example |
|---|---|---|
passthrough |
Sent as-is to the provider | style, quality |
renamed |
Same value, mapped to the field name the provider expects | prompt |
converted |
Transformed to the provider's native format | size |
emulated |
Works even if the provider has no concept of it | n, response_format |
Parameters we don't recognize pass straight through to the upstream API, so provider-specific options still work.
Gallery
View all 2 imagesSora 2 FAQ
How much does Sora 2 cost?
Sora 2 starts at $0.1 per video through Lumenfall. Pricing varies by provider. Lumenfall does not add any markup to provider pricing.
How do I use Sora 2 via API?
You can use Sora 2 through Lumenfall's OpenAI-compatible API. Send requests to the unified endpoint with model ID "sora-2". Code examples are available in Python, JavaScript, and cURL.
Which providers offer Sora 2?
Sora 2 is available through Replicate, OpenAI, and fal.ai on Lumenfall. Lumenfall automatically routes requests to the best available provider.
Overview
Sora 2 is a high-fidelity video generation model developed by OpenAI that transforms text prompts or static images into cinematic video sequences. It represents an evolution in the Sora family, capable of producing content at 720p resolution with extended durations reaching up to 20 seconds. This model is distinguished by its ability to maintain temporal consistency and complex motion dynamics over longer spans than many first-generation video models.
Strengths
- Temporal Consistency: Maintains the identity of characters, objects, and environmental details across the entire 20-second duration, minimizing the “morphing” or warping effects common in shorter-clip models.
- Physical Simulation: Demonstrates a sophisticated understanding of physical interactions, such as fluid dynamics, lighting reflections, and gravity, leading to more realistic movement.
- Multimodal Input Flexibility: Supports both text-to-video for purely generative tasks and image-to-video for animating existing assets or extending still photography into motion.
- Enhanced Resolution: Outputs native 720p video, providing sufficient clarity for social media content, prototyping, and digital backgrounds without immediate need for upscaling.
Limitations
- Causal Reasoning: While physically grounded, the model may still struggle with complex “cause and effect” sequences, such as a character taking a bite out of a cookie and the cookie not showing a bite mark.
- Spatial Confusion: In high-action scenes involving multiple moving parts (e.g., a crowded street), the model can occasionally mix up left/right orientations or produce impossible limb movements.
- Resolution Ceiling: At 720p, it lacks the native 4K or 1080p detail required for professional film production pipelines without significant post-processing.
Technical Background
Sora 2 utilizes a diffusion transformer (DiT) architecture, which combines the scaling properties of transformers with the generative capabilities of diffusion models. It operates on spacetime patches, treating video data as a three-dimensional collection of patches that allow the model to train on diverse aspect ratios and resolutions. This architecture enables the model to look ahead and behind in a sequence to ensure global coherence rather than generating frames in a strictly linear, autoregressive fashion.
Best For
Sora 2 is best suited for rapid prototyping in creative agencies, generating b-roll for digital marketing, and creating environmental backgrounds for web design. It excels at animating conceptual art where maintaining character consistency is a priority. This model is available for testing and integration through Lumenfall’s unified API and playground, allowing developers to compare its outputs directly against other video generation frameworks.
Try Sora 2 in Playground
Generate images with custom prompts — no API key needed.