01
In brief
Gemini Omni Flash 1.1 is available through Google's Interactions API under the stable model ID gemini-omni-1.1-flash. This guide focuses on the implementation intent: getting from an API key to generated video, then using the controls that make Omni Flash different from a basic text-to-video endpoint.
02
Basic text-to-video
Google's Python SDK uses client.interactions.create(model="gemini-omni-1.1-flash", input="..."). JavaScript uses the same model through ai.interactions.create. REST calls POST to the v1beta interactions endpoint and return generated video in the response steps. In the SDK, Google exposes a convenience output_video field.
03
Portrait video for Reels and Shorts
Set response_format to video and aspect_ratio to 9:16. The other documented aspect ratio is 16:9. This is the most immediately useful creator control for vertical social video workflows.
04
Resolution control
Set the response resolution to 360p, 720p, 1080p or 4k. 720p is the default. Google's docs explicitly describe 1080p and 4K as upscaled outputs, so do not describe them as native-resolution generation.
05
Image-to-video
Pass an image plus a motion-specific text instruction. Google recommends high-resolution source images and specific descriptions of subject movement, camera motion and environment rather than vague prompts such as 'make it move'.
06
First/last-frame interpolation
Provide two images and describe the transition. Omni Flash generates the movement between the starting and ending frames. This is useful when the creator needs controlled endpoints rather than an unconstrained generation.
07
Explicit tasks
The documented video task values are text_to_video, image_to_video, reference_to_video, edit and extend. Google recommends relying primarily on prompting and using the task field when the model needs additional mode guidance.
08
Conversational/stateful editing
After generating a video, follow-up edits can reference the previous interaction ID. This lets the model retain video state and apply natural-language changes without requiring the developer to treat every edit as an unrelated generation.
09
Video extension
Use the extend behavior to continue an existing clip. The model card supports video input up to 10 seconds for editing and extension, and Google's current guide says clips can be extended by 10 seconds at a time up to 40 seconds total. Verify those limits again before production deployment because implementation constraints can change.
10
Practical implementation checklist
Use the stable model ID, choose 9:16 or 16:9, prototype at an appropriate resolution, keep prompt instructions specific, preserve previous interaction IDs for iterative edits, and store generated video output safely rather than assuming an SDK-only convenience field exists in raw REST responses.
11
Pricing and limits
Google's paid-tier table lists $1.50 per million input tokens, $9 per million text-output tokens and $17.50 per million video-output tokens. At the documented 5,792 output tokens per second for 720p, Google estimates about $0.10 per second. The current table has no free tier for Gemini Omni Flash. Rate limits are account- and tier-dependent, so verify them in your project rather than inventing a universal quota.
For the broader release, capabilities and availability context, read the main Gemini Omni Flash 1.1 guide.
Sources
Primary and supporting sources
Facts were rechecked against the linked sources immediately before publication. Pricing, product availability and rollout status can change.