Gemini Omni Flash

• 🆕 First model in Google's new "Omni" family.
• 🎬 Image-to-video variant: animates a still image into clips.
• 💬 Conversational editing carries context across turns.
• 👁️ Accepts text, multiple reference images, and existing clips.
• 🧠 Combines physics understanding with Gemini world knowledge.
• 🔒 Outputs carry SynthID watermarking, verifiable via Google surfaces.
• 🔧 Preview model, trained on Google TPUs.

Google

Google is an American multinational technology corporation and one of the world's most valuable brands. A subsidiary of parent company Alphabet Inc., Google operates across search, cloud computing, consumer electronics, and artificial intelligence. Its…

Gemini Omni Flash is the first model in Google's new "Omni" family. This entry is the image-to-video surface: you supply a starting still image plus instructions, and the model animates it into a short clip. Google frames Omni as a step toward models that create and edit media from varied inputs, beginning with video and blending its generative stack with Gemini's reasoning and world knowledge.

Rather than a pure text-to-video generator, Omni is a multimodal model that can understand, analyze, and edit existing content through plain-language conversation. Because editing is conversational, context carries across turns, so each instruction refines a take instead of restarting from a blank prompt. Alongside a prompt it can ingest multiple reference images and existing video clips, drawing on physics understanding to place and animate subjects in a scene.

The model shares its release date with its text-to-video and reference-to-video siblings, Gemini Omni Flash and Gemini Omni Flash R2V, and sits apart from Google's specialized Veo line, so Veo 3.1 Full Quality and earlier Veo 3 Full Quality remain distinct offerings. Generated videos include SynthID watermarking that can be verified through Google surfaces such as the Gemini app. The model remains in preview and was trained on Google TPUs.