Gemini Omni Flash's feature set is defined by one shift: video editing as a conversation. Previous models treat each generation as isolated. Omni Flash chains interactions into a persistent editing session. Every other feature (timecode syntax, image role tags, multi-shot defaults) follows from this conversational foundation.

Stateful Multi-Turn Editing

Generate a video, then send follow-up instructions that modify the existing result. The model remembers the full video context from the previous turn. "Make the phone invisible. Keep everything else the same." The phone disappears. The person's hand, the background, the lighting, the audio all persist. This works across multiple turns: generate, edit, edit again, edit again. Each turn produces a new video that builds on the previous one. No re-uploading, no re-describing the scene.

Multi-Shot Narrative by Default

By default, Omni Flash creates videos with multiple shots, scene transitions, and narrative pacing. It interprets your prompt as a story, not a single frame. For a prompt about "a woman playing violin outdoors," the model might open with an establishing shot of the park, cut to a medium shot of the violinist, then close on her hands. If you need a single continuous take, add "in a single unbroken scene" to your prompt.

Timecode Event Syntax

Control exactly when events happen using natural language or timecode brackets. "[0-3s] A person is walking. [3-6s] They stop and turn around. [6-10s] They start running." Or: "At 5s the chorus starts in the background." "Every 2s cut to a new frame." "Every half a second change the scene to a new location." This level of per-second control is unique among video generation models.

Image Role Tagging

Upload multiple images and assign each one a specific role. <FIRST_FRAME> uses an image as the starting frame. <IMAGE_REF_0> through <IMAGE_REF_N> use images as subject references. A prompt like "a woman <IMAGE_REF_0> is walking while holding <IMAGE_REF_1>" composites two separate reference subjects into a single scene. Up to 6+ reference images demonstrated in the API docs.

Promptable Audio Generation

Audio is generated alongside video by default. Direct the audio through your prompt: "Include calm background music." "The audio is a low tinny radio broadcast." "High energy techno beat." You can also suppress unwanted audio elements: "No dialogue." "No extra sound effects." The model matches audio mood and timing to the visual content.

Edit Uploaded Videos

Upload your own video files and edit them with text instructions. "When the person touches the mirror, make the mirror ripple like liquid." The model applies VFX-style edits to real footage. This turns Omni Flash from a generation tool into an editing tool. Upload footage from your phone, describe the edit, and get modified video back. Available in most regions (not EEA/Switzerland/UK for uploaded video editing).

Text Rendering in Video

Prompt for readable text in your video. "One word on the screen at a time: did, you, know, that, Omni, can, do, awesome, text? Each word appears for 1s with a different animated style." Street signs, storefront names, license plates, and on-screen titles all render legibly. Specify exact text content in your prompt to control what appears.