Sora settings guide

Sora 2 generates videos up to 25 seconds at 1080p with synchronised audio — the longest and highest-quality output of any mainstream AI video model.

Use image-to-video mode — start with a high-quality Flux or Midjourney image and animate it; this gives far more control over the starting frame than pure text-to-video.

Sora follows film language precisely — 'dolly zoom on subject', 'rack focus from foreground to background', 'handheld shaky cam' produce technically accurate camera work.

Generate short 5-second clips and stitch in a video editor — quality degrades in longer generations; multiple precise short clips edited together produce better results than one long generation.

Sora's character consistency within a single generation is its strongest feature — characters maintain appearance, clothing, and proportions across the full clip without morphing.

Include specific lighting descriptions — 'golden hour backlight with lens flare', 'neon-lit rain-slicked street at night', 'overcast diffuse natural light' dramatically change the visual quality.

Use Sora for concept visualisation and pre-vis — it is currently too expensive and inconsistent for final production, but unbeatable for rapidly communicating a visual concept to stakeholders.