Gemini 4 Argon announced by Google on Sept 30, 2026

Gemini 4 writes the shot. Nano Banana films it.

Use Gemini 4's reasoning and video understanding to plan, write and critique your AI videos. Then render them with MiniMax H3 and Seedance 2.5 in one studio.

1 Your idea

Format

Style

Shots

Render with

2 Paste this into Gemini 4

Open Gemini Then render it ↓

01 / What is Gemini 4

Google's new flagship is a thinker, not a camera.

Gemini 4 is the next generation of Google's Gemini models. The first one, Gemini 4 Argon, was announced on September 30, 2026. It's built for deep reasoning, long multi-step tasks and visual understanding, which makes it a great creative director for AI video, while a video model does the filming.

Status as of October 2026. Sources: TechCrunch, Engadget.

AnnouncedSept 30, 2026

First modelGemini 4 Argon

StrengthsDeep reasoning, long-horizon workflows, coding

VisionCharts, documents and long-form video understanding

AvailabilityRolling out to Google's Fairwind Program first, then paid API customers and Google AI Ultra subscribers

Video generationNot part of the announcement. Pair it with a video model.

02 / The workflow

A creative loop that gets better every take.

Gemini 4 handles the thinking and the feedback. Nano Banana handles the pixels. Most teams go round this loop two or three times before they ship.

01 Gemini 4
Plan

Turn a one-line brief into a shot list: hook, beats, camera moves, lighting, pacing and format.
02 Gemini 4
Write

Get one production-ready prompt per shot, written for the model you'll render with.
03 Nano Banana
Render

Generate each shot with MiniMax H3 or Seedance 2.5, side by side, in 9:16, 1:1 or 16:9.
04 Gemini 4
Review

Upload the clips back to Gemini 4. Its video understanding spots what's off and rewrites the prompt.

03 / What to use it for

Five ways to pair Gemini 4 with AI video.

Storyboard

Storyboard a multi-shot ad

Ask Gemini 4 for a 3 to 5 shot sequence with a hook in the first second, then render every shot and cut them together.

Reference

Reverse-engineer a reference video

Give Gemini 4 a clip you love. It describes the framing, camera motion, grade and rhythm, and turns them into a reusable style prompt.

Variations

Product variations at scale

One packshot, ten angles, moods and backgrounds. Gemini writes the variations, Nano Banana renders them.

UGC

UGC hooks and scripts

Generate hooks, creator lines and on-camera actions for AI avatars and talking photos.

Long-form

Long video to shorts

Let Gemini 4 watch a long video, pick the moments that matter and write b-roll prompts for each short.

04 / Copy-paste templates

Prompts to give Gemini 4.

Replace the parts in [brackets], paste into Gemini 4, and send the result to Nano Banana.

Plan

Shot list

You are a commercial director. Turn this brief into a [3]-shot [9:16] video for [TikTok].
Brief: [your product and message]
For each shot give: duration (5s max), subject, action, camera move, lighting, mood.
Then write one single-paragraph video prompt per shot, under 60 words, present tense, no dialogue, ready for an AI video model.

Analyze

Reference to style

Watch this video. Describe its visual style so I can recreate it with an AI video model: framing, lens feel, camera motion, color grade, lighting, pacing, texture.
Then write one reusable style prompt (max 40 words) I can append to any scene.

Improve

Prompt doctor

Here is my AI video prompt: "[prompt]"
Rewrite it for [MiniMax H3]: one clear subject, one camera move, specific lighting, a style, and the [16:9] format. Keep it under 60 words. Give me 3 variations with different moods.

Review

Clip review

Here is a clip I generated from this prompt: "[prompt]"
Watch it and list what doesn't match the intent (subject, motion, lighting, continuity, artifacts).
Then give me a revised prompt that fixes those issues.

Write

Hooks for ads

Give me 10 scroll-stopping opening shots for a [9:16] ad about [product], each one a single visual moment that works without sound. Format each as a ready-to-render video prompt under 40 words.

Analyze

Long video to b-roll

Watch this video and pick the [5] strongest moments for short-form clips. For each, give a timestamp, a title, and a b-roll video prompt (under 50 words) that illustrates it.

05 / Who does what

Gemini 4, Gemini Omni and Nano Banana.

Two Google names, two different jobs. Here's how they fit in a video workflow.

Gemini 4

Gemini Omni Video

Nano Banana

What it is

Reasoning & understanding model

Video generation model

Multi-model video studio

Best at

Planning, writing, reviewing

Prompt-faithful scenes

Rendering and comparing models

Watches video

Yes, long-form

Outputs MP4

Yes

Yes, with MiniMax H3, Seedance 2.5 and more

Learn more

Above

Gemini Omni Video

Nano Banana Studio

06 / For developers

Automate it: brief in, video out.

Chain the Gemini API with the Nano Banana REST API: Gemini writes the prompt, Nano Banana returns the video URL. Set the Gemini 4 model ID once it's available on your Google account.

{ }

Nano Banana REST API

€59.99/mo flat, 15 jobs / 30 min. Full docs in /docs/.

MCP server NEW

Let any MCP-compatible agent render videos on your studio credits.

brief-to-video.js

import { GoogleGenAI } from '@google/genai';

// 1. Gemini writes the shot prompt
const ai = new GoogleGenAI({ apiKey: process.env.GEMINI_API_KEY });
const plan = await ai.models.generateContent({
  model: process.env.GEMINI_MODEL, // your Gemini 4 model ID
  contents: 'Write one 9:16 video prompt, under 60 words, for: ' + brief
});

// 2. Nano Banana renders it
const res = await fetch('https://nanobananavideo.com/api/v1/text-to-video.php', {
  method: 'POST',
  headers: {
    Authorization: `Bearer ${process.env.NB_API_KEY}`,
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({ prompt: plan.text, video_model: 'veo3', duration: 5 })
});
const { video_url } = await res.json();

07 / FAQ

Gemini 4 and AI video, answered

Something else? Contact us.

Can Gemini 4 generate videos?

Google's announcement of Gemini 4 Argon (September 30, 2026) focuses on reasoning, coding and visual understanding, not video generation. To get an MP4, pair Gemini 4 with a video model: it writes and reviews the prompts, and a model like MiniMax H3 or Seedance 2.5 renders them in Nano Banana.

Is Gemini 4 available inside Nano Banana?

No. Nano Banana is a video studio, Gemini 4 is a Google model you use in Gemini or through the Gemini API. The workflow on this page uses both side by side: copy the prompt Gemini 4 writes, paste it into Nano Banana, render.

When can I use Gemini 4?

According to Google, Gemini 4 Argon is rolling out first to its Fairwind Program (governments and trusted partners), then to developers, enterprises and general users, starting with paid API customers and Google AI Ultra subscribers. Until you have access, the templates on this page also work with other Gemini models.

How does Gemini 4 help with video prompts?

It turns a short brief into a shot list, writes one precise prompt per shot (subject, camera, lighting, style, format), and can watch the clips you generate to tell you what to fix. That loop usually gets you to a usable take faster than editing prompts by hand.

What is the difference between Gemini 4 and Gemini Omni Video?

Gemini 4 is a reasoning and understanding model: it reads, watches and writes. Gemini Omni Video is a video generation model: it outputs clips. See the Gemini Omni Video page.

Which video model should I render with?

MiniMax H3 is fast and strong on realistic people and physics. Seedance 2.5 shines on fluid motion and multi-shot storytelling. In Nano Banana you can render the same Gemini prompt with both and keep the best take.

Got a prompt from Gemini 4? Film it now.

Paste it below and render it with MiniMax H3 or Seedance 2.5. No camera, no crew, nothing to install.

Building a pipeline? Explore the API