Gemini Omni

Positioning Overview

gemini-omni-flash/reference-to-video is the reference-image-to-video route in the Gemini Omni Flash family, supporting two task forms: reference-image-to-video and first-frame-image-to-video. Developers submit one to seven reference images and a prompt to describe how the subject, scene, and style are constrained by the reference images, and produce controllable videos of 4 to 10 seconds, 720p to 4k, with selectable landscape or portrait orientation. It is suitable for developers who need to lock in character consistency and visual style in advertising, e-commerce, and short video assets.

Core Highlights

• Multi-reference images constrain subject and style——Up to seven reference images can be submitted at once, and the prompt can specify that each image controls the subject, scene, or style separately.
• Duration, resolution, and aspect ratio are all selectable——Output duration of 4/6/8/10 seconds, resolution of 720p/1080p/4k, and aspect ratio of 16:9 or 9:16 can all be specified directly in the request.
• Reproducible seed parameter——An integer seed parameter is provided, making it easy to keep results stable and comparable during batch generation under the same configuration.

Capability Description

Reference-image-driven consistent generation: Using 1 to 7 reference images as input, combined with prompts to constrain the subject's appearance, scene, and visual style, reducing identity drift across multi-shot footage.

Specifications specified on demand: Directly set duration, resolution, and aspect ratio within a single request, so the output can directly match the specification requirements of vertical short videos or landscape placements.

Seed control and batch generation: Fix random input via the seed parameter, making it easy to reproduce the same version during batch production or to make small parameter comparisons.

Long prompt descriptions: The prompt supports up to 20,000 characters, allowing complete description of the division of labor among shots, actions, and reference images.

Use Cases

• Advertising and e-commerce teams producing character-consistent assets——Upload multiple product and model reference images to produce style-consistent 4 to 10 second vertical short films for feed advertising and detail page video slots.
• Short video creators batch-producing landscape and portrait versions——Generate separately in 16:9 and 9:16 using the same set of reference images, obtaining two sets of final cuts suitable for the main site and vertical channels.
• Generation workflows requiring traceable versions——Rerun with fixed seed and prompt to obtain reproducible shot versions for client review and subsequent iteration comparison.

Limitations and Selection Advice

Only image-to-video and reference-image-to-video are supported; pure text or audio input is not accepted. Output duration is limited to 4 to 10 seconds, so longer continuous shots cannot be generated at once. The reference image limit is seven, the parameter set is fixed, and additional editing or continuation instructions are not supported. If pure text-driven or longer content is needed, choose other routes in the same family or generate in segments and stitch them together.