Gemini Omni Flash 1.2: Conversational Video Generation

Gemini Omni Flash 1.2 is Google’s conversational video model, available in ComfyUI through the Google Gemini Omni (Video) partner node. Describe the shot in plain language, attach reference images or a clip, and the node returns video with synchronized audio. Version 1.2 renders clips from 3 to 15 seconds at 360p, 720p, 1080p or 4K, in 16:9 or 9:16. It takes up to 14 reference images and 3 reference videos, and extends a clip to 30 seconds. Omni Flash 1.1 and the older preview model stay selectable in the node.

​

What Gemini Omni Flash 1.2 is good at

• Conversational editing: change a character, relight a scene, alter the camera angle, or add and remove objects while the rest of the shot keeps its framing and audio
• Multimodal input: combine text, up to 14 images, and up to 3 reference videos of 15 seconds each
• Keyframe interpolation: with the image_to_video task, attach a starting frame and an optional ending frame, and the model generates the footage between them
• Reference to video: tag each reference image as <IMAGE_REF_0> and name it in the prompt, so a character or product appears in a new scene instead of being used as a frame
• Scene extension: the extend task continues a clip from its last frames, up to a 30-second result, without a visible seam
• Resolution control: draft at 360p, then render the final clip at up to 4K

​

Available workflows

​

Text to Video (Omni Flash 1.2)

Write the scene, the length (3 to 15 seconds), and the motion into one prompt. The node returns the clip with a generated audio track.

​

Image to Video (Omni Flash 1.2)

Animate a still with the image_to_video task. The first image is the opening frame, an optional second image is the closing frame, and the prompt carries the motion and the sound.

​

Reference to Video (Omni Flash 1.2)

Keep a subject consistent across a new scene. Tag each reference image as <IMAGE_REF_0> and name the tag in the prompt: the character or product appears in the clip, while the image itself is never used as a frame.

​

Video Edit (Omni Flash 1.2)

Change one thing in a finished clip with the edit task: swap a background, restyle a scene, add or remove an element, while the cut, timing and audio stay. Attach exactly one video of up to 15 seconds, keep the prompt to the change you want, and add “keep everything else the same” for maximum consistency.

​

Video Extend (Omni Flash 1.2)

Continue a finished clip instead of re-shooting it. The extend task reads the last frames and keeps the subject, framing and audio running for the next stretch: a 5-second input becomes a 10-second output. Input videos can be up to 25 seconds, and the result can reach 30 seconds.

​

Get started

1. Update ComfyUI to 0.44.1 or later
2. Double-click the canvas and search for “Google Gemini Omni (Video)”, or open a workflow from the Template Library
3. Select Omni Flash 1.2 in the model dropdown
4. Pick the task that matches your input, write the prompt, and run