← All models

Gemini Omni Flash

Video Preview

google-deepmind/gemini-omni-flash · by Google DeepMind · Diffusion transformer (DiT)

Pricing — 1 offering(s)

Google DeepMind (Gemini API)

Standard (720p)

$0.10 / second Current 2026-06-30 → present

Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.

Capability profile

Prompt adherence moderate

temporal consistency moderate

motion quality moderate

native audio weak

camera control unknown

character consistency moderate

Resolution ceiling weak

Inference speed strong

clip duration ceiling weak

image to video quality unknown

Compositional accuracy unknown

cost efficiency strong

video editing moderate

Operator guidance

Default choice when editing capability and low latency are required and native audio is not needed. At $0.10/s (720p only), this is the cheapest option in the registry with API-level video editing. For silent T2V/I2V at lower cost, use Veo 3.1 Lite ($0.05–0.08/s); for native audio or 4K, use Veo 3.1 Standard/Fast. The Interactions API's stateful session model (previous_interaction_id chains turns) enables multi-turn refinement — note that conversational session management is an API design pattern, not a product-only chat layer. The edit task (V2V) has an active limitation with video input processing as of 2026-07-02 — verify current status before committing to production editing workflows. Geo restriction: externally-uploaded video editing is unavailable in EEA, Switzerland, and UK.

Use cases

Fast T2V/I2V for latency-sensitive production pipelines
Multi-turn conversational video editing sessions (Interactions API)
Reference-conditioned generation anchored to character or subject images
Cost-efficient video generation with editing capability at $0.10/s

Limitations

Preview status; pricing and capabilities subject to change before GA
720p only — no 1080p or 4K output
No native audio; audio input references unsupported
Max 10 seconds per clip; no video extension mode
edit task (V2V) — video references up to 3s accepted by API schema but not correctly processed (active limitation, 2026-07-02)
Editing of externally-uploaded video geo-restricted (unavailable EEA, Switzerland, UK)
Multi-video referencing unsupported
System instructions, temperature, and top_p parameters not supported
No mask-guided inpainting
Input token costs ($1.50/M) apply separately — significant for video editing inputs