Nano Banana 2 Lite × Omni 1.1: Still-to-Video Workflow

When people search Nano Banana 2 Lite Omni 1.1 Flash image-to-video, they usually are not hunting another “Lite is cheap” news blast. They already have usable stills—but the short clip will not connect. They want first-to-last transitions, scene extension, or a 4K finish, yet their notes still cite the June preview. Nano Banana 2 Lite owns fast still drafts; Gemini Omni 1.1 Flash (gemini-omni-1.1-flash) turns those stills into conversationally editable video. This post is only that creator pipeline—and what actually changed versus the June preview.

This site is not Google’s official docs. Prices, parameters, and regional limits follow the Gemini Omni Flash docs and the API Changelog. Here we turn them into a decision table: what to draw first, what to push next, and when to upscale. If you are still on the Lite + Omni launch explainer, skim the official launch note first, then come back for the 1.1 upgrades.

Try Nano Banana 2 image tools

What creators must change for Omni 1.1 vs preview

On 2026-08-27, Gemini Omni Flash reached GA on the stable endpoint gemini-omni-1.1-flash. The old gemini-omni-flash-preview endpoint sunsets on 2026-09-30. For creators and small teams that means:

Numbers and limits (extend only appends at the end; some regions cannot extend/edit uploaded videos) stay with the official docs. Below we split by delivery scenario, not SDK dumps.

Pipeline A: Lite still → Omni image-to-video (default)

This chain was announced in June and remains the fastest path from ecommerce hero to short clip, or interior concept to walkthrough:

1. Generate 3–6 still candidates with Nano Banana 2 Lite (~4s-class per 1K still; cost follows official pricing).
2. Human-filter warped faces, garbled type, and off-brand color—video models amplify still mistakes instead of fixing them.
3. Hand the winner to Omni 1.1 with an image-to-video task; describe camera, subject motion, and environment—not just “make it move.”
4. Conversational edits for a few turns: push in, relight, hold the product longer—cheaper than full regenerations.

Lite is for direction screening. If the still needs readable Chinese sell-copy, lock a higher tier on this site before video; see in-image text localization. Covers and boards can follow the short-video cover workflow.

Still prompt skeleton (swap the subject):

Studio ecommerce still life: matte black ceramic coffee mug centered, seamless light-gray backdrop, soft side light, subtle steam, brand-color accents without garbled text, crisp product silhouette, 16:9, no watermark.

Motion prompt for image-to-video (after the still):

Slow dolly-in, natural rising steam, keep mug body and handle unchanged, subtle table-top reflection shimmer, clean backdrop, cinematic shallow depth of field, 5-second portrait or landscape clip.

First–last frame interpolation: before/after as a transition

First–last interpolation is the most tangible 1.1 deliverable for creators: supply a start frame and an end frame; the model fills the transition. Great for:

• Packaging closed → opened
• Room before → after a redesign
• Expression A → expression B (same character DNA on both frames)

Generate strongly consistent “before” and “after” stills with Nano Banana 2 Lite (or the standard tier), then hand both to Omni. Matching framing, camera height, and subject scale keeps transitions stable; huge mismatches create hard-cut hallucinations.

30-second QA:

• Does the subject morph or swap faces mid-clip?
• Do brand colors and key props persist?
• Does motion only hit what you intended (lid opens; table does not teleport)?

Video extend: ship a 5s hook, then continue

Video extend only appends at the tail—no prepend, no mid-clip insert. Official guidance: about 3–10 seconds per continuation; multi-turn stacks toward roughly 40 seconds total. A safer creator pattern:

1. Ship a 5-second hook (product close-up or room first impression).
2. After it passes QA, continue in plain language: “pull back to reveal the table set,” “character enters from the right and waves.”
3. Add reference images when you need new characters/props—do not cram the whole plot into the first generate.

Note: some regions restrict extend/edit on uploaded videos; multi-turn clips the model already generated are supported more broadly. Follow official Limitations; do not assume identical global behavior.

Resolution: draft at 720p, upscale for delivery

1.1 adds explicit resolution control. Practical defaults:

Public materials still often quote video output around $0.10 / second (same ballpark as Veo 3.1 Fast messaging). Lite stills remain the cheap place to fail fast. Always confirm current official pricing.

Preview migration checklist (deadline 2026-09-30)

If scripts, workflows, or bookmarks still say gemini-omni-flash-preview:

1. Switch the model ID to gemini-omni-1.1-flash.
2. Retest image-to-video, first–last, extend, and resolution once each.
3. Check regional limits (EEA / UK / Switzerland and similar) for uploaded-clip edit/extend.
4. Update internal SOPs so teammates stop pasting the old endpoint.

For a longer AI-video case library, see the practical handbook. This article only covers how the still→video chain changes after 1.1.

Wrap-up

The right mental model for Nano Banana 2 Lite Omni 1.1 Flash image-to-video is not “buy another motion model.” It is: Lite for fast cheap still iteration → Omni 1.1 for image-to-video, first–last interpolation, and tail extend → 1080p/4K only on the final pass. Migrating the endpoint before preview sunset beats chasing the next flashy demo.

Today: generate three product stills here, pick one for a 5-second image-to-video, then add a second beat with interpolation or extend.

Try Nano Banana 2 image tools