Skip to content
  • Sora
  • Prompt Engineering
  • AI Video Generation
  • Generative AI
  • Best Practices

Codex CLI Complete Guide

Sora 2 Prompt Engineering Guide: Success Patterns and Failure Avoidance

This article is a follow-up to the morning article

Morning article: Sora 2 Complete Guide: Getting Started and Usage

Goals

  • Optimize Sora 2's 6-element prompt structure (Subject/Action/Setting/Style/Audio/Length) for production use
  • Avoid failure patterns proactively to reduce credit consumption
  • Quantitatively evaluate generation efficiency across resolution/length/quality combinations

Practical Prompt Design Workflow

Step 1: Create Baseline Prompt

Start with a minimal configuration to verify quality.

【Minimal Example】
Dog running in park. Sunny weather, wide shot. 10 seconds.

Verification checklist at this stage:

  • Is the subject (dog) correctly recognized?
  • Are the settings (park, sunny) reflected?
  • Is the length as specified?

Step 2: Incremental Element Addition

Once baseline succeeds, add elements in the following order.

Priority 1: Action Specification

【Improved v1】
Golden Retriever sprinting at full speed across grassy park.
Sunny weather, wide shot. 10 seconds.

Effect: Breed specification improves fur texture; "full speed" clarifies motion dynamism

Priority 2: Style Addition

【Improved v2】
Golden Retriever sprinting at full speed across grassy park.
Sunny afternoon, wide shot, slow motion, 4K quality. 10 seconds.

Effect: Slow motion specification enables detailed rendering of fur and muscle movement

Priority 3: Audio Design

【Final Version】
Golden Retriever sprinting at full speed across grassy park.
Sunny afternoon, wide shot, slow motion, 4K quality.
Dog's breathing and footsteps, children laughing in distance. 10 seconds.

Effect: Environmental sounds dramatically enhance immersion and scene realism

Step 3: Generate Variations and Compare

Create multiple versions of the same concept to identify optimal solution.

VersionChangesCreditsQuality ScoreNotes
v1Minimal config506/10Monotonous motion
v2Action specified807.5/10Improved dynamism
v3Style added1508.5/10Significant texture improvement
v4Audio added2009/10Maximum immersion, recommended

Cost Efficiency: v3 to v4 quality difference is 0.5, but credit difference is 50 (+33%). Choose based on use case.

6-Element Optimization Techniques

Subject Design

Recommendation: Focus on 3 or fewer specific characteristics

❌ Bad: "person"
⭕ Good: "woman in her 30s, long black hair, wearing red coat"

Rationale: Vague specifications produce inconsistent results across generations.

Action Design

Recommendation: Use verb + adverb to specify motion quality

❌ Bad: "walking"
⭕ Good: "walking powerfully" / "walking slowly" / "skipping while walking"

Benchmark Results:

SpecificationMotion NaturalnessPhysics Compliance
"walking" only6/107/10
"walking powerfully"8/109/10
"skipping while walking"9/108/10

Setting Design

Recommendation: 3-element set of location + time + lighting

❌ Bad: "beach"
⭕ Good: "beach at sunset, warm orange light, calm waves"

Effect: Lighting specification stabilizes color tone and improves consistency across multiple generations.

Style Design

Cinematic Terminology Usage:

TermEffectRecommended Use
Wide ShotCaptures broad areaLandscapes, multiple subjects
MacroEmphasizes detailsProduct demos, food
HandheldImmersion, documentary feelEvents, street walks
Slow MotionDetailed motionSports, action
4K QualityResolution boostPro version only, commercial use

Audio Design

Recommendation: Combine environmental sounds + dialogue

【Environment Only】
"Sound of waves, seagull calls"

【Dialogue Only】
"Voice cheerfully shouting 'Hello!'"

【Combination (Recommended)】
"Sound of waves and seagull calls, distant voice shouting 'Hello!'"

Benchmark: Combined version shows +25% improvement in immersion score vs. standalone.

Length Design

Credit Efficiency Analysis (1080p baseline):

LengthCreditsCost per SecondRecommended Use
5 sec20040Short-form social posts
10 sec35035Standard clips
15 sec48032Most cost-efficient
20 sec65032.5Pro version only, long-form needs

Recommendation: Plus version users should baseline at 15 seconds.

Failure Patterns and Avoidance Strategies

Pattern 1: Physics Law Violations

SymptomCauseAvoidance
Objects floating in airVague action specificationUse specific verbs: "falling", "rolling"
Water flowing upwardNo gravity direction specifiedSpecify direction: "flowing down", "dripping"
Unnatural limbsComplex pose specificationStart with basic poses: "standing", "sitting"

Pattern 2: Audio-Visual Mismatch

SymptomCauseAvoidance
Dialogue doesn't match lip movementOverly detailed audio specificationKeep to "speaking", avoid specific dialogue
Environmental sounds too loudMultiple sound sources specified simultaneouslyLimit to 1 primary + 1 background sound

Pattern 3: Style Conflicts

SymptomCauseAvoidance
Blurry image"Macro" + "Wide Shot" simultaneouslyUnify to one shot type
Choppy motion"Slow Motion" + "Fast" contradictionSpecify only one speed

Credit Consumption Optimization Strategies

Strategy 1: Gradual Resolution Upscaling

【Procedure】
1. Verify composition/motion at 720p (Credits: 100)
2. If satisfied, generate same prompt at 1080p (Credits: 200)

Effect: Reduces credit loss on failures by 50%.

Strategy 2: Batch Generation (Pro Version)

Leverage Pro version's "5 simultaneous generations" feature:

【Example: Product Demo Video】
1. Generate 5 angle variations simultaneously with same prompt
2. Select best result
3. Total 1000 credits for 5 variations = 200 per video

Effect: Increased options improve quality vs. single generation.

Strategy 3: Relaxed Mode Usage (Pro Version)

Prioritize unlimited Relaxed mode for non-urgent production:

【Recommended Use Cases】
- Test generations
- Archive footage creation
- Multi-variation comparisons

Caution: Generation time is 2-3x longer, unsuitable for deadline-critical projects.

Automation & Extension Ideas

  1. Prompt Template Management: Store frequently-used structures in Notion/Obsidian for reusability
  2. Quality Score Logging: Rate each generation (1-10) to accumulate success patterns
  3. Credit Consumption Tracker: Record daily usage in Excel to avoid month-end shortage
  4. API Integration (Future): Automate prompt A/B testing after API release

Next Steps