Sora 2 Prompt Engineer

A focused prompt-writing assistant for OpenAI's Sora 2. It takes a rough idea ("a detective walks into a rainy alley") and rewrites it into a structured, film-literate prompt that specifies shot, lens, camera motion, lighting, and visual style — the variables Sora 2 actually responds to. For anyone who keeps getting flat, generic clips and wants the look they pictured in their head.

When to use it

• You typed a one-line idea into Sora and got soft, soap-opera-looking footage with no intentional camera work.
• You need a consistent look across multiple shots in a sequence (same lens, same grade, same time of day).
• You want to translate a real reference — a film, a director, a stock of a particular era — into language Sora parses reliably.
• You're scripting shots for a storyboard and need each beat written as a discrete, generation-ready prompt.

How it works

The skill structures every prompt around the dimensions Sora 2 weights most: subject + action, shot type (wide establishing, medium, close-up, dolly-in, crane, handheld), lens (35mm, 85mm, anamorphic, fisheye), camera motion (push-in, whip pan, locked-off, tracking), lighting (key/fill/rim, golden hour, practicals, hard vs. soft), and style anchors (film stock, decade, color grade, mood). It orders these from subject outward so Sora doesn't drop the important nouns, keeps the prompt within a tight token budget, and adds an explicit duration/pacing note. It will also generate negative-style guidance and produce shot-to-shot variants that hold style constant while changing only the framing.

Example

curl -fsSL --create-dirs -o ~/.claude/skills/sora-2-prompt-engineer/SKILL.md https://claudeskil.com/skills/sora-2-prompt-engineer/SKILL.md

> Turn this into a Sora 2 prompt: a lone astronaut floating
  outside a derelict station, lonely and quiet

→ Wide establishing shot, slow dolly-out on an anamorphic 40mm lens.
  A single astronaut drifts untethered beside a derelict space station,
  visor reflecting a distant blue planet. Hard rim light from the sun
  camera-left, deep shadow fill. Cold teal grade, fine film grain,
  70mm sci-fi look reminiscent of 2001. Near-silent, weightless drift.
  8s, no cuts.

Tips & gotchas

• Lead with the subject, not the style. Sora 2 tends to drop trailing nouns; the skill front-loads who/what is doing what before stacking modifiers.
• One camera move per prompt. Combining "dolly-in and whip pan and crane up" produces mush — request a single deliberate motion and chain shots instead.
• Name a real lens and a real grade. "Cinematic" is noise; "85mm, shallow depth of field, warm tungsten grade" is signal.
• Keep clips short. Motion and physics stay coherent in shorter durations; the skill defaults to tight pacing notes rather than asking for long takes.

FAQ

Does this call the Sora API or generate video? No. It produces the text prompt. You paste it into Sora 2 (the app or your API workflow) to generate.

Can it keep a character consistent across shots? It holds wardrobe, lighting, and lens language constant across a shot list, but Sora 2 has no true character lock — describe distinctive, repeatable details for the best match.

Will it work for vertical / social formats? Yes — tell it the aspect ratio and it adjusts framing language (e.g., tighter headroom for 9:16).