Sora 2 Video API

Generate AI videos with OpenAI’s Sora 2 models via the Runcrate API — text-to-video with realistic physics simulation.

Generate videos using OpenAI’s Sora 2 models through the Runcrate API. Sora 2 excels at realistic physics simulation, consistent object permanence, and cinematic camera work.

Sora API is being discontinued in September 2026. Plan migrations to Veo 3, Seedance, or Kling v3 before the deadline.

Available models

Model	Durations	Best for
Sora 2	4, 8, 12s	General-purpose video generation
Sora 2 Pro	4, 8, 12s	Production-quality, complex scenes

---

Basic text-to-video

Python

from runcrate import Runcrate

client = Runcrate(api_key="rc_live_YOUR_API_KEY")

job = client.models.generate_video_and_save(
    "sora-demo.mp4",
    model="openai/sora-2",
    prompt="A paper airplane folds itself from a blank sheet, launches off a wooden desk, "
           "and flies through an open window into a bright blue sky. Tracking shot.",
    duration=8,
    on_status=lambda j: print(f"  {j.status}..."),
)

---

Sora 2 Pro — production quality

from runcrate import Runcrate

client = Runcrate(api_key="rc_live_YOUR_API_KEY")

job = client.models.generate_video_and_save(
    "hero-video.mp4",
    model="openai/sora-2-pro",
    prompt="Slow-motion close-up of water droplets falling onto a leaf, each creating "
           "concentric ripples. Macro lens, natural diffused lighting, shallow depth of field.",
    duration=12,
    on_status=lambda j: print(f"  {j.status}..."),
)

---

Sora 2 vs. alternatives

Feature	Sora 2	Veo 3.0	Seedance 1.0
Max duration	12s	8s	12s
Physics realism	Excellent	Good	Good
Resolution	HD	4K	HD
Audio	No	Yes (Veo 3 Audio)	No
Status	Sunsetting Sep 2026	Active	Active

Migration plan

The API shape is identical across all video models — changing the model parameter is the only code change needed:

Current	Migrate to	Why
Sora 2	Veo 3.0	Higher resolution, audio support
Sora 2 Pro	Veo 3.0	Comparable quality, actively maintained

---

Tips

• Sora 2 excels at physical interactions — dominos, water, fabric, collisions.
• Camera language works well — “tracking shot,” “dolly zoom,” “slow motion” are understood.
• 12s is the maximum — for longer videos, generate multiple clips and stitch them.
• Start with Sora 2 for drafts, use Sora 2 Pro for finals.

Next steps

• Video generation reference
• Seedance Video API — ByteDance’s alternative
• AI Video Pipeline — batch generation with multiple models