Google launches Gemini Omni Flash at $0.10/sec for video editing

• Google shipped two creator-facing media models at once: GoogleDeepMind's launch thread introduced Gemini Omni Flash for video generation and editing, while OfficialLoganK's launch post priced it at $0.10 per second and Nano Banana 2 Lite at $0.034 per 1,000 images.
• According to GoogleDeepMind's feature list, Omni Flash is built around conversational video edits, multimodal references, real-world knowledge, and text or graphics tied directly to video actions.
• Day-one rollout was unusually broad: Magnific's announcement, Runway's announcement, and Higgsfield's first demo all put Gemini Omni Flash in creator tools immediately.
• Early examples clustered around VFX and live instruction following, with chrisfirst's voice-command demo showing spoken color changes and Higgsfield's background-removal demo showing one-prompt compositing.
• One practical caveat surfaced fast: Higgsfield's footage note said Gemini Omni Flash works best on real footage, not AI-generated video.

Google paired a cheap image model with a cheap video editor, and the more interesting part is how tightly the workflow is stitched together. You can browse Google's launch page, see GoogleDeepMind's Interactions API post stack up to three sequential edits, and jump straight into partner surfaces from Magnific or Runway.

The launch has two pieces:

• Nano Banana 2 Lite: Google's fastest and cheapest Gemini image model, with sub-4-second text-to-image generation according to GoogleDeepMind's speed post and OfficialLoganK's launch post.
• Gemini Omni Flash: a video generation and editing model available in the Gemini API, Google AI Studio, and Gemini Enterprise Agent Platform, per GoogleDeepMind's availability post.
• Shared workflow: GoogleDeepMind's Interactions API post says developers can generate an image with Nano Banana 2 Lite, animate it with Omni Flash, keep session history, and apply up to three sequential edits.
• Price point: OfficialLoganK's launch post set Omni Flash at $0.10 per second, the same price he cited for Veo 3.1 Fast.

Google's own positioning is speed first. GoogleDeepMind's launch copy frames Nano Banana 2 Lite as an ideation model for workflows where latency and cost are the blockers, while GoogleDeepMind's feature list frames Omni Flash as the editing layer that turns those quick assets into moving scenes.

The standout demos were not text-to-video trailers. They were edit loops.

chrisfirst's demo post, cited in chrisfirst's voice-command demo, shows Omni Flash listening to spoken instructions inside the video and changing the background color to match whatever the subject says. A follow-up prompt in chrisfirst's prompt post pushes the same pattern further by swapping the speaker's likeness and the surrounding scene based on each character he names.

Google's own capability list in GoogleDeepMind's feature list breaks the product into four concrete behaviors:

• Conversational video editing
• Multimodal referencing and combining inputs
• Real-world knowledge
• Connecting text and graphics directly to video actions

Partner demos leaned hard into the last point. Higgsfield's VFX demo shows a typed "remove background" instruction turning into instant rotoscoping and a new composite, and Higgsfield's chair VFX demo pushes ordinary footage into stylized effects rather than full scene generation.

This was not an API-only ship.

• Magnific: Magnific's announcement said Gemini Omni can generate, animate, and edit video in one model, and that it is available through Magnific, Magnific MCP, and Spaces.
• Runway: Runway's announcement added Omni Flash for prompt, image, or video starting points, and said users can also ask Runway Agent to use Omni.
• Higgsfield: Higgsfield's first demo, Higgsfield's follow-up VFX clip, and Higgsfield's Claude MCP post turned the launch into a stream of creator-facing examples, including usage through Higgsfield MCP on Claude.

That distribution matters because it gives creators three different entry points on the same day: Google's native surfaces, standalone creative apps, and MCP-driven agent workflows.

The most specific usage note in the evidence came from Higgsfield. Higgsfield's footage note said Gemini Omni Flash performs best when you feed it real footage, not AI-generated video.

That lines up with the strongest examples in the thread pool:

• chrisfirst's voice-command demo starts from a person speaking on camera.
• Higgsfield's VFX demo starts from live-action footage and edits the background.
• Higgsfield's Claude MCP post spotlights realistic VFX on footage routed through Claude via MCP.

The creative angle here is closer to post-production than pure generation. Google's own feature list sells character, camera, and action control, but the partner clips suggest the first breakout use case may be fast footage transformation.

Magnific turned the launch into a mini prompt pack within hours. Magnific's thread opener framed five tested cinematic prompts as reusable building blocks, and the examples in Magnific's astronaut prompt, Magnific's soldier prompt, Magnific's animation prompt, and Magnific's stop-motion prompt all share the same structure:

• shot sequence, not single-shot description
• camera verbs like whip pan, close-up, shaky shot, aerial
• action beats chained in order
• environment changes embedded into the same prompt

That is a useful tell about how creators are steering Omni Flash already. The prompts read less like image prompts and more like compressed shot lists, which fits a model marketed around editing, reference inputs, and multi-step scene control.