🎬 Sora 2 / Runway / Luma Video Prompt Engineer

Platform-Optimized Prompt

A solo female dancer in her 30s, mid-dark skin tone, long dark hair, wearing a flowing red silk dress, begins motionless in deep shadow with eyes closed. Slowly, a single warm tungsten light fades up from upper camera-left, illuminating only the right side of her face and the red silk falling across her shoulder. Her head tilts up by 15 degrees over 6 seconds, eyes opening with controlled stillness. She does not dance yet — this is the moment before. The camera dollies in slowly from medium-wide framing to medium-close-up over the full 12 seconds, with a subtle upward arc that elevates her presence. Theatrical chiaroscuro lighting. The shadow side of her face remains in black silhouette throughout. Background: pure black, no visible environment. Style: 16mm film grain texture, slight gate weave, rich saturated red against pure black, restrained palette, cinematic. 16:9 aspect. Duration 12 seconds. Negative: no warped face, no extra fingers, no flickering background, no anachronistic elements, no AI distortion in clothing folds.

Why This Order (Front-Load Audit)

Sora 2 weights the first ~100-150 chars heavily for character consistency + scene setup. The ordering: character → setting (motionless in shadow) → action (light fades up, head tilts) → camera (dolly-in) → style (16mm grain) → negatives.

If you front-loaded the camera move first, Sora would over-weight the technical motion and under-weight the character. Subject + state first, then action, then camera, then aesthetic.

Camera Specification Detail

• Start framing: Medium-wide. Dancer occupies center 40% of frame vertically, head at upper third.
• End framing: Medium-close-up. Head + shoulders fill upper 60% of frame.
• Dolly-in over: 12 seconds, EASE-IN-OUT (not linear) — accelerates slightly in middle then settles at end.
• Subtle upward arc: ~5-degree vertical rise during the dolly. Elevates the subject psychologically without becoming dramatic crane move.
• Sora 2 handles this kind of slow controlled move well IF the action is also slow (which it is — eyes opening + head tilt is sub-shot-scale motion).

Lighting + Mood Specification

• Source: Single tungsten practical, ~3200K, hard quality, upper-camera-left ~30 degrees above eye line.
• No fill: Deep shadow on camera-right side of face. Pure black, not soft fill.
• Practical motivation: Implied stage spotlight, but Sora doesn't need that explained — the chiaroscuro instruction handles it.
• Color temp: Warm key (~3200K) against pure black background. Red silk picks up the warm tone, becomes near-orange in highlight, deep crimson in mid-tone.
• No backlight, no rim light, no spill — keeps composition theatrical-minimal.

Iteration Strategy

Generation 1 (expected outcome): Sora produces a roughly correct shot. Likely failures: face may have slight inconsistency between start and end frames; the dolly-in may be too fast (Sora tends to compress timing); silk fabric may show 'AI sheen' (over-smooth motion).

Generation 2 tweaks: If face inconsistent, add 'consistent same dancer throughout, identical face features.' If dolly too fast, increase 'over the full 12 seconds' emphasis + add 'slow patient camera move, deliberate pacing.' If fabric issue, add 'realistic silk physics with weighted folds, natural fabric drape.'

Generation 3 refinement: Color-correct in post — Sora tends to slightly over-saturate red. Pull saturation back 5-10% in DaVinci.

If Used in a Sequence (4-shot music video opening)

Continuity anchors for next 3 shots:

• Same dancer description (lock to: 'mid-dark skin tone, long dark hair, red silk dress')
• Same lighting language (warm tungsten, no fill, theatrical chiaroscuro) — adjust for new shot's framing
• Same color palette (red against black)
• Same film-grain style (16mm)

Shot 2 suggestion: First gesture begin — arm extends up into the light. Camera locks off. 8 seconds.

Shot 3: Full body silhouette as light expands. Wider framing. 10 seconds.

Shot 4: First spin into full motion. Multi-camera-friendly. 12 seconds.

Common Failure Modes for Sora 2 + This Subject

• Face inconsistency between start (eyes closed) and middle (opening) and end (open): Sora 2 has improved but still drifts. Mitigate with explicit 'same dancer throughout, consistent face features, no morphing.'
• Hands appearing or hand-distortion: Specify 'arms relaxed at sides, hands not visible until shot 2.' Pre-empt the hand-failure mode.
• Background creep: Sora may add subtle environmental details to 'pure black.' Reinforce with 'pure black void background, no visible environment, no shadows on a floor.'
• Over-dramatic camera move: Sora 2 sometimes accelerates dolly-ins. Counter with 'patient, slow camera, deliberate pacing.'

Post-Generation Workflow

1. Generate 3 variations. Pick the best.

2. Upscale to 4K in Topaz Video AI (Sora 2 outputs at 1080p typically).

3. Color grade in DaVinci Resolve: pull saturation back 5%, lift mid-tones slightly, deepen blacks.

4. Add subtle film grain if Sora's grain looks too clean — Filmconvert or DaVinci's grain tool.

5. Stabilize subtle camera-shake if any drift. Resolve's optical-flow stabilizer.

6. Edit into sequence with the next 3 shots in DaVinci or Premiere. Maintain consistent color treatment across all 4 shots.

7. Final delivery: ProRes 422 HQ at 24fps for cinema-grade or H.264 at 60Mbps for YouTube/Vimeo.