Sora hints and real‑time video demo
Clues suggest ChatGPT may soon integrate OpenAI’s Sora to create videos, while Runway demoed real‑time, high‑def AI video generation using Nvidia GPUs — signaling rapid progress in generative video tooling. On‑device video synthesis is still constrained, but these moves foreshadow new creative APIs and heavier inference demands. (kalingatv.com) (petapixel.com)
APK teardown of ChatGPT for Android build 1.2026.076 exposes new UI strings such as “Video in ChatGPT is here” and “Transform text and image into video with dialogue, soundtrack, and style,” suggesting work-in-progress generative‑video UI elements. (androidauthority.com) OpenAI’s Sora 2 remains a two‑variant API offering—sora-2 for speed and sora-2-pro for higher fidelity—with published per‑second pricing of $0.10 for sora-2 and $0.30 for sora-2-pro. (developers.openai.com) Sora 2’s API explicitly supports asynchronous POST /videos jobs, polling or webhook notifications for completion, and generation windows commonly set at 16–20 seconds with sora-2 for rapid iteration or sora-2-pro for 1080p production exports. (developers.openai.com) Runway’s research preview shown at NVIDIA GTC produced HD frames with time‑to‑first‑frame under 100 milliseconds using models running on NVIDIA’s Vera Rubin (NVL72) hardware, presenting the same live‑stream frame flow as a game engine rather than batch renders. (petapixel.com) Runway reported porting its Gen‑4.5 video model and new GWM‑1 world models from Hopper to Rubin NVL72 rapidly and describes Rubin as providing up to 50 petaflops of inference compute per GPU for long, high‑fidelity video sequences. (runwayml.com) Public Open‑Sora replication notes and Sora documentation indicate 720×1280 portrait (and 1280×720 landscape) as common Sora output sizes and show inference for higher‑resolution configs can require more than one 80GB‑class GPU with sequence parallelism, underscoring why real‑time HD generation is currently centered on datacenter Rubin/HPC hardware rather than mobile SoCs. (github.com, developers.openai.com)
