How to Generate Video with Sora 3: First Project Setup
Sora 3 is OpenAI’s video model that generates clips up to 20 seconds long with synchronized audio. You type a scene, choose resolution and length, and the model returns a clip that includes ambient sound and, if you ask for it, spoken lines. A 10-second 720p generation typically takes around two minutes.
This is a setup guide, not a film-school lecture. It covers where to start, what the settings actually change, and the two mistakes that waste your first hour.
What is Sora 3?
Sora 3 is a text-to-video and image-to-video model from OpenAI. It supports clip lengths from 5 to 20 seconds, resolutions up to 1080p, and native audio generation. It lives inside the ChatGPT interface for Plus and Pro subscribers, with Pro getting longer clips and higher priority in the queue.
The audio part is the real difference from earlier versions. You get footsteps, room tone, and dialogue in the same pass, which saves a separate sound-design step for short-form content.
How do you start your first project?
Open the video section, then create a new project. You get a single prompt box and a settings panel. That panel is where most beginners go wrong, so read it carefully.
- Pick resolution first. 720p for drafts, 1080p only for the final take.
- Set clip length. Start at 5 seconds. Longer clips cost more and drift more.
- Choose aspect ratio. 16:9 for landscape, 9:16 for vertical social.
- Write the prompt with subject, action, setting, and camera move.
- Generate and review. Regenerate the same prompt gives a different result every time.
How should you write a Sora 3 prompt?
Describe it like a shot list. Sora 3 rewards specificity about sequence and camera. “A barista slides a cup across a counter, close-up on the cup, then a wide shot of the cafe, morning light through the window” produces two distinct beats in one clip.
Things that help:
- Name the camera angle: “overhead,” “handheld,” “locked-off wide.”
- Keep the number of characters low. Three or more people in one shot tends to produce messy faces.
- Say “no on-screen text” if you keep seeing fake captions appear.
For audio, ask for it explicitly. “Include the sound of rain on a metal roof” works. Left alone, Sora 3 often adds generic ambience that does not match the scene.
Is Sora 3 free?
No, not on its own. Access comes through ChatGPT Plus at $20/month or Pro at $200/month. Plus gives you a monthly generation allowance; Pro gives longer clips and faster queue priority. The free ChatGPT tier does not include video generation.
That pricing shapes how you should work. Draft at 720p and 5 seconds, review, then spend a Pro-tier generation on the one take that looks right.
What should you avoid?
Do not expect clean text in frame — signage and labels still smear. Do not expect precise physics in complex interactions; objects passed between hands can merge. And do not generate at 20 seconds on your first try. You will pay for it and probably throw it away.
Quick Summary
- Sora 3 makes 5 to 20 second clips up to 1080p with native audio.
- Available with ChatGPT Plus ($20/mo) and Pro ($200/mo), not the free tier.
- Draft at 720p/5s, save 1080p for the final take.
- Write a shot list; specify camera, characters, and audio.
- On-screen text and complex object physics are still unreliable.
