'Sora 2 is still the only video model that generates indistinguishable real video' — and it's gone
gandamu_ml · x · 2026-11-09
A practitioner argues Sora 2 remains the only video model that can generate indistinguishable real video: other models feel like GPT-4o text did at first — one output wows you, ten outputs reveal mode-collapsed slop.
The poster says they're still struck by memories of incredible Sora 2 output about once a week, finds it odd that it was widely accessible and now isn't, and worries its particular taste may never come back.
More from Multimodal
- Designer builds 2027 series and short film concepts with Claude Opus 5.5 — threepointone · 2026-11-09
- Microsoft Patents AI That Rewrites Live Speech Per Viewer and Syncs Lip Movements — ZeroStateReflex · 2026-11-09
- VideoRebirth hyperflow-h3-flash, a MiniMax H3-based video model, trends on Hugging Face — VideoRebirthAI · 2026-11-09
- "We mistook fatigue for transcendence": retro-futurist AI art strikes a chord — piotrbinkowski · 2026-11-09
- User generates cinematic human-history-to-AGI 3D animation with a single GPT prompt — imjustnewatai · 2026-11-09
- AI-generated 80s VHS-style ad for fictional "Z-AWAY Zombie Repellent" spray — brythehuman · 2026-11-09