Google's AI Remembers Every Video Edit | Gemini Omni Honest Review

I stress-tested Google's new Gemini Omni across 3 completely different scenarios — and found exactly where it holds up and where it breaks. CASE 1 — THE GUITARIST (16:9 Cinematic) → Background swap with full re-lighting → Outfit + guitar swap via reference images → Style transfer to Makoto Shinkai anime → Particle effects FAILED (fireflies = flat overlay) CASE 2 — THE WALKING WOMAN (9:16 Vertical) → Wardrobe swap (crop top → leather jacket) → Environment transform (farmers market → neon rain street) → Flower dress via reference image input → Camera angle change BROKE (walked backwards) CASE 3 — THE ADRENALINE DRONE (16:9 Cinematic) → Timestamp-driven FPV canyon generation → Night magma environment swap → Dragon added via reference image → Smooth transitions still a weak spot KEY LESSON: Don't prompt intent. Direct physics. 🛠️ Tools used: Gemini Omni Flash Nano Banana Pro (reference images) 🔗 AI video ads for your brand: https://www.fiverr.com/users/deepak_ms_ 🔗 Instagram: @cine.studio.ai #geminiomni #aifilmmaking #aivideo #googleai #geminiomniflash #texttovideo #aicommercial #promptengineering #cinestudioai #tutorial #videoediting #googleio2026 #aitools