Tried out Sora2 a bit, and it's pretty awesome (nano-banana.pro)

OpenAI’s Sora 2 is being positioned as a breakthrough AI video generator — the “GPT‑3.5 moment” for video — that combines physics‑accurate simulation, frame‑perfect synchronized audio, and fine-grained multi‑shot controllability. It can produce photorealistic, cinematic, or anime‑style sequences where objects obey real‑world dynamics (water flows, rebounds, buoyancy, object permanence) and characters perform complex motions (gymnastics, skating) while audio (dialogue, ambience, music) is generated to match video precisely. Users can inject real-world “Cameo” clips to composite people and voices into scenes, and prompts plus reference images guide style and composition. Technically, Sora 2 outputs HD video (up to 1080p) in MP4 (H.264/H.265) across common aspect ratios and 24/30/60 FPS, with 48kHz/24‑bit stereo or spatial audio and HDR/color gamut support. Uploads support JPG/PNG/WebP (optimal ≤10 MB); typical render times are 2–5 minutes depending on complexity. The result is a tool aimed at marketing, entertainment, and education—enabling realistic product demos, storyboards, and physics demos—while representing a step change in emergent model capabilities (better object permanence and failure modeling). It’s not flawless, but its combination of physical realism, synchronized audio, and persistent world state marks a significant technical advance for AI‑driven content creation.