Gemini Omni 1.2 Flash
Gemini Omni 1.2 Flash is Google's newest fast video model. Write a prompt and get a clip of up to 15 seconds with dialogue, ambience and sound effects built in. Attach reference images and short clips to lock a face, a product or a style, or switch to frame mode and animate between a first and last frame. Compared with the previous version it holds characters and scenes steadier across camera moves, follows references more closely, and runs 30% cheaper.
Gemini Omni 1.2 Flash is Google’s newest fast video model, and you can use it on Mitte right now. Describe a scene and get up to 15 seconds of video with the sound already in it: people talking with their lips in sync, footsteps on gravel, rain on a window. Attach a photo and it comes to life. Attach a few references and the people and products in them stay recognizable from the first frame to the last.
The earlier version made good eight-second clips. This one goes to 15 seconds in a single pass, keeps faces, outfits and products the same across camera moves, and follows your reference images and clips far more closely. Camera work is steadier too: FPV flythroughs, orbits and long zoom-outs with no unwanted jump cuts.
What can you do with it?
- Generate from a prompt. Write the shot the way a director would: what happens, in what order, how the camera moves, what the light looks like. Add the dialogue in quotes and it is spoken on screen.
- Bring a still to life. Upload one image and describe the motion. The video starts exactly on your picture.
- Set a first and a last frame. Switch to frame mode, give it where the shot begins and where it ends, and it fills in the motion between them. Great for match cuts and product reveals.
- Keep people and products consistent. Up to ten reference images and three short clips: a model’s face, three outfits front and back, a bottle with its label. Faces, garments, packaging type and brand colors stay put across camera angles and 360-degree turns.
- Borrow motion and setting from a clip. A two-second reference video tells it how something moves or where the scene takes place.
- Vertical or widescreen. 16:9 for YouTube and the web, 9:16 for Reels, Shorts and TikTok.
How to use it
- Write the prompt. Actions in order, then the camera, then the light and mood. Put dialogue and on-screen text in quotes.
- Attach references (optional). Images of the people, products or style, and short clips for motion. Tag them in the prompt as @Image1 or @Video1 and say what each one is for.
- Or switch to frame mode. Upload a first frame and, if you want, a last frame, and describe what happens between them.
- Set length, ratio and resolution. 3 to 15 seconds, landscape or portrait, from 360p drafts up to 4K.
- Generate. Download the result, share it, or send it into an edit, an extension or an upscale right on Mitte.
What you get
One MP4 clip per run, 3 to 15 seconds long, with native audio, in 16:9 or 9:16 at 360p, 720p, 1080p or 4K.
Tips
- Draft small, finish big. Explore at 360p or 720p, then upscale the shot you love to 1080p or 4K instead of generating everything at full size.
- Give it a reference for anything that must not drift. A product label, a logo, a face. One good photo does more than a paragraph of description.
- Keep reference clips short. A two or three second clip of the motion or the setting is all it needs.
- Write the sound. “She whispers ‘we’re late’, rain on the awning, a bus hissing past.” The audio is generated with the picture, so describe it like you describe the shot.
- Use a last frame for reveals. A closed box as the first frame and the opened product as the last gives you a clean, controlled reveal. Keep those clips to 10 seconds.
- Build long scenes in steps. Generate 15 seconds here, then extend it in 10-second steps with the extend tool.
- Steady motion over fast action. The model is at its best with natural movement and smooth camera moves. Fast sports and complex physics are still rough.