Magnific adds Gemini Omni Flash: 10-second turn-by-turn scene edits

• Google shipped two creator-facing media models at once: OfficialLoganK's launch post priced Nano Banana 2 Lite at $0.034 per 1K image with sub-4 second generation, while GoogleDeepMind's release thread said Gemini Omni Flash is now in the Gemini API, AI Studio, and Google Enterprise Agent Platform.
• Magnific turned Omni Flash into a workflow story, not just a model drop: magnific's rollout post positioned it as one model for generate, animate, and edit, and magnific's memory claim said edits keep the whole scene in memory across turns.
• The most useful creator detail is the editing grammar. In magnific's rollout post, Magnific says prompts can target character, camera angle, or action, while magnific's prompt thread shows multi-shot cinematic prompts instead of one-line style requests.
• Google framed Omni Flash around conversational editing and multimodal reference handling in GoogleDeepMind's capability list, and chrisfirst's voice-command demo showed a clip reacting to spoken instructions inside the video.
• Figma's bundle is already spreading the rollout beyond Google's own tools: figmaweave's offer post added Nano Banana 2 Lite and Gemini Omni Flash to Weave, while Ash_uxi's Weave thread described 20-plus Weave tools that operate directly on design assets inside the canvas.

You can jump from Google's own launch link to Magnific's integration post, and the examples get specific fast. magnific's prompt pack is already publishing reusable shot-by-shot prompts, magnific's plugin post puts AI tools inside Premiere, After Effects, DaVinci Resolve, and Final Cut Pro, and figmaweave's offer post shows the same Google models landing in a Figma-native workflow.

The base release is a paired launch. OfficialLoganK's launch post introduced Nano Banana 2 Lite as the fast, cheap image model, while GoogleDeepMind's release thread used the same announcement to put Gemini Omni Flash into the Gemini API and AI Studio.

Google's own capability list is concise:

• Conversational video editing, per GoogleDeepMind's capability list
• Multimodal referencing and combined inputs, per GoogleDeepMind's capability list
• Real-world knowledge, per GoogleDeepMind's capability list
• Direct links between text, graphics, and video actions, per GoogleDeepMind's capability list

The pricing and speed numbers on the image side help explain the bundle. OfficialLoganK's launch post gave Nano Banana 2 Lite a sub-4 second latency claim and a $0.034 per 1K image price, and the attached chart in OfficialLoganK's benchmark card compares it against Flux, Grok Imagine, and Seedream on image generation, image editing, latency, and price.

Magnific's pitch is iterative editing. In magnific's rollout post, the company says Omni Flash can generate, animate, and edit with one model, then breaks control into three editable levers: character, camera angle, and action.

The more interesting claim sits in magnific's memory claim, where Magnific says the model keeps the whole scene in memory so each edit builds on the last. That is the detail creative teams care about, because most demo-friendly video systems still fall apart once a sequence needs multiple changes instead of a fresh render.

Magnific also says the rollout is available through Magnific MCP and Spaces in magnific's first announcement, which turns Google's API launch into a tool creators can reach from existing Magnific surfaces instead of only from AI Studio.

The early usage pattern is not terse prompting. magnific's prompt pack breaks examples into cinematic beats with explicit shot changes, camera movement, and action timing, including astronaut corridor chases, fast-rope action scenes, animation coverage, and stop-motion style clips.

That thread suggests a simple rule about what Omni Flash wants from users: not more adjectives, more blocking. The prompts are built like miniature shot lists.

A second pattern is live instruction following. In chrisfirst's voice-command demo, a speaker tells the system to match the background color to whatever color a subject says, and the video updates as the spoken command changes.

Figma Weave quietly turned this into a broader creator-tools rollout. figmaweave's offer post added both Nano Banana 2 Lite and Gemini Omni Flash to Weave's Google-model bundle, packaged with a 50 percent first-month offer.

That matters because Weave is already shifting toward asset-aware tooling inside the canvas. In Ash_uxi's Weave thread, Ash_uxi says Config brought 20-plus Weave tools directly into the design canvas and says they work by pointing at design assets instead of writing long prompts.

The screenshot in

shows the shape of that workflow: select an existing object, describe the material, generate a finished variant. Omni Flash arriving in that same bundle extends Google's rollout into design systems where iteration starts from live assets, not blank prompts.