Gemini Omni Flash

Gemini Omni Flash
Create the scene. Keep the conversation going.

Gemini Omni Flash is Google DeepMind’s multimodal video model for generation and conversational editing. Combine a written idea with visual or audio references, then refine the scene through follow-up instructions about the action, setting, or camera.

The current stable API model is Gemini Omni 1.1 Flash. It replaces the earlier Omni Flash Preview.

A florist in a brick-red apron arranging orange flowers in a sunlit studio

Mixed inputs

Text, image, audio, and video

Conversational edits

Refine a scene through follow-up instructions

Picture + sound

Video generation with native audio

Core capabilities

A violin maker gently polishing a wooden violin at a sunlit workbench — **Scene brief**
The violin maker gently polishes the instrument, turns it toward the window, and pauses to inspect the wood. Hold a medium shot in soft daylight, with quiet workshop ambience.

01 / Multimodal direction

Bring different references into one scene

Give every input a clear creative role.

Gemini Omni Flash can work with text, images, audio, and video. Use the image to establish the violin maker and workbench, then describe the action and pacing in words. When adding another reference, explain whether it should guide the appearance, movement, or sound. A product video benefits from a defined hero shot and a short list of supporting details.

Specific roles for references make the intended subject, setting, and motion easier to follow.

Two matching compositions of a ribbed glass vase, with orange ranunculus on the left and blue hydrangeas on the right — **Edit direction / orange → blue**
Replace the orange flowers with pale blue hydrangeas. Keep the glass vase, camera position, stone table, and soft side light.

02 / A focused revision

Refine a detail through conversation

Say what changes and what stays.

Conversational editing lets you revise a video through natural-language instructions. Keep the request local: change the flowers while retaining the vase, table, lighting, and framing. A focused revision makes it easier to judge whether the new detail fits the scene.

After a revision, compare the changed object and the surrounding scene before moving to the next edit.

A ceramic artist showing a speckled bowl to an apprentice in a daylight pottery studio — **Dialogue + action**
Artist: “Feel how thin the rim is.” He turns the bowl slightly toward the apprentice. She leans closer. Soft workshop ambience and a gentle ceramic tap.

03 / Speech and atmosphere

Write sound into the performance

Connect each sound to something the viewer can see.

The model generates video with audio. Identify the speaker, keep the line short, and give the listener time to react. In a pottery lesson, the bowl, hand gesture, and spoken instruction can all support the same moment without filling the scene with competing actions. For a short film, let each exchange change what the viewer understands.

Write the words, delivery, and relevant background sounds as part of the same scene brief.

Gemini Omni 1.1 Flash formats

Choose the duration, orientation, and delivery size for the scene.

Duration

3–10 seconds

Short video generation.

Frame rate

24 fps

A consistent output frame cadence.

Orientation

16:9 / 9:16

Landscape or portrait. Portrait framing suits YouTube Shorts and other vertical edits.

Resolution

360p to 4K

720p is the default; 1080p and 4K are upscaled.

Generate, edit, or extend

Choose the task according to what you already have and what needs to change.

Task	Starting point	Creative direction
Generate	A prompt, with optional references	Establish a new scene and its action
Edit	An existing video or previous generation	Describe the specific revision
Extend	The end of an existing clip	Describe what happens next
Interpolate	A first image and a last image	Describe the transition between them
Upscale	A video to deliver at a larger size	Select 1080p or 4K output

Use cases

Explore Pexo Create workflows for product moments, character scenes, and short stories.

A rust-orange rain jacket hanging in a softly lit studio

Product Ad Video

Build a product moment around a clear action and a consistent visual direction.

A daughter handing her mother a brass house key at a townhouse door

Short Film

Plan a brief exchange with performance, pacing, and sound.

A baker carrying a pastry box out of a neighborhood bakery at dawn

Story Video

Give a character a task, a developing action, and a closing response.

Make the revision easy to describe

Keep the original direction, the requested change, and the finishing check separate. For a story video, make each revision support the same character goal.

Matching dark teal and burnt-orange suitcases in the same sunlit apartment entryway

Establish the original

A dark teal suitcase stands on an oak floor in a sunlit apartment entryway. The handle and wheels are clearly visible.

Request one change

Change the suitcase shell to burnt orange. Retain its shape, hardware, camera position, lighting, and background.

Check the complete shot

Review the case edges, handle, wheels, and shadows. Make the next request about one remaining detail.

FAQ

What is Gemini Omni Flash?

Gemini Omni Flash is Google DeepMind’s multimodal video generation and editing model. It supports mixed references and conversational revisions. The current stable API identifier is gemini-omni-1.1-flash.

Is Omni Flash the same as Omni Flash Preview?

The names refer to the same model family, but different versions. Google lists Gemini Omni 1.1 Flash as the stable replacement for gemini-omni-flash-preview, which shut down on October 22, 2026.

Does Gemini Omni Flash generate audio?

Yes. It can generate video with native audio, including speech and sounds that belong to the scene. Describe the speaker, words, and ambience clearly, then review them with the picture.

Is 4K output native?

The generation guide describes 1080p and 4K delivery as upscaled output. The default is 720p, and 360p is also available.

Can it connect a first and last frame?

Yes. Provide the starting and ending images and describe the transition. Keep the subject and the intended change clear so the two frames form a coherent visual sequence.

Can it extend an uploaded video?

Yes. Uploaded extension sources must be 10 seconds or shorter. Extension adds footage at the end. For uploaded clips with someone speaking, adding new dialogue is currently restricted; generated-video continuation supports speech through multi-turn interactions. Uploaded-video extension is unavailable in the EEA, Switzerland, and the UK.

Describe your idea and start your next creation.

Start creating

Frequently Asked Questions (FAQ)