Sora 2 Prompting Guide Released: OpenAI Official Best Practices

OpenAI has officially published the “Sora 2 Prompting Guide” in its Cookbook—the first official prompting guide. This article distills the core points to help you quickly get started with Sora 2 video generation.

Core Philosophy: Prompts Are Shot Briefs

Writing video prompts is like briefing a cinematographer—you provide direction, the model provides creative variations. The key is finding the balance between control and creative space.

Two Strategies

Detailed Prompts ⚙️

• Provide stronger control and consistency
• Suitable for scenes requiring precise reproduction

Concise Prompts ✨

• Give the model more creative freedom
• May bring unexpected surprises

Essential API Parameters

The following parameters cannot be changed through text description and must be explicitly set in the API call:

{
  "model": "sora-2-pro",        // sora-2 or sora-2-pro
  "size": "1280x720",           // video resolution
  "seconds": "8"                // 4, 8, or 12 seconds
}

Supported Resolutions

Five Essential Elements of Effective Prompts

A clear prompt should describe a shot like a storyboard:

1. Camera Framing - Wide? Close-up? What angle?
2. Depth of Field - Shallow (blurred background) or deep (all sharp)?
3. Action Beats - Describe movement in executable steps
4. Lighting Scheme - Describe light source, direction, and quality
5. Color Anchors - Specify 3-5 main colors

From Weak to Strong: Example Comparisons

Scene Description

Camera Setup

❌ Weak Example:

Camera shot: cinematic look

✅ Strong Example:

Camera shot: wide shot, low angle
Depth of field: shallow (sharp on subject, blurred background)
Lighting + palette: warm backlight with soft rim

Action Description

❌ Weak Example:

Actor walks across the room.

✅ Strong Example:

Actor takes four steps to the window, pauses, and pulls the curtain in the final second.

Short Prompt Example

Sometimes, concise works well:

In a 90s documentary-style interview, an old Swedish man sits in a study 
and says, "I still remember when I was young."

Why This Works:

• 90s documentary sets the overall aesthetic
• old Swedish man sits in a study provides basic scene
• Dialogue is short and clear, easy to sync

Note: Many details unspecified (time, weather, wardrobe, age, etc.)—Sora will fill these in.

Copy-and-Use Standard Template

[Describe scene in plain language: characters, costumes, setting, weather, etc.]

Cinematography:
Camera shot: [wide establishing shot, eye level]
Depth of field: [shallow / deep]
Lens/style: [anamorphic lens / handheld / locked-off]
Mood: [cinematic and tense / playful and suspenseful / warm and cozy]

Actions:
- [Action 1: clear, specific beat or gesture]
- [Action 2: another distinct beat]
- [Action 3: closing action or line]

Dialogue:
[If dialogue present, 1-3 short lines]

Advanced Techniques Quick Reference

1. Use Image Input to Lock Style

Pass an image as the input_reference parameter to lock:

• Character design
• Wardrobe style
• Set dressing
• Overall aesthetic

2. Dialogue Writing Essentials

• Place in separate block, distinct from visual description
• Keep concise and natural
• Label speakers (multi-character scenes)
• Match duration: 4-second clip = 1-2 exchanges

Example:

Dialogue:
- Detective: "You're lying. I can hear it in your silence."
- Suspect: "Or maybe I'm just tired of talking."

3. Use Remix for Precise Iteration

Remix is a fine-tuning tool, not for regeneration:

• ✅ “same shot, switch to 85mm”
• ✅ “same lighting, new palette: teal, sand, rust”
• ❌ Don’t change too many things at once

Strategy: When close to target, pin as reference and only describe “delta changes.”

Quick Problem Solving

Why This Approach Works

It aligns with Sora 2’s text-to-visual conversion logic:

1. Clear camera intent - Reduces model guessing
2. Constrained motion - Do one thing at a time
3. Stable lighting logic - Maintains visual coherence
4. Visible anchors - Use specific nouns and verbs

Practical Recommendations

Recommended Workflow

1. Clarify Goals → What should this shot achieve?
2. Set Parameters → Choose model, size, seconds
3. Write Prompt → Start with template or keep it concise
4. Generate Multiple Versions → Embrace variety, select closest
5. Remix Optimize → Targeted adjustments, not restart
6. Edit and Integrate → Combine satisfying clips into complete work

Duration Selection Tips

• ⏱️ 4-second clips - Most reliable instruction following
• ⏱️ 8-second clips - Require more precise prompts
• 💡 Tip: Editing two 4-second clips may work better than generating a single 8-second one

Complete Reference Resources

Want to dive into all the details? Check out the full guide:

📖 Read Complete Documentation →

The complete version includes:

• Cinematic ultra-detailed prompt examples
• Lighting and color theory explained
• Two complete scene cases (robot workshop, rooftop dance)
• More troubleshooting techniques

About This Guide

Official Source: OpenAI Cookbook - Sora 2 Prompting Guide
Authors: Robin Koenig (OpenAI), Joanne Shin (OpenAI)
This Article: Organized by SoratoAI Community