Google-Native Codex Video Pipeline: Nano Banana 2 Lite + Gemini Omni Flash (2026)

If your Codex workflow runs entirely on Google models, you don't need to mix ecosystems. AnyCap gives you a complete Google-native pipeline inside Codex — one API key, two models, end-to-end from static image to final video.

The pipeline: Nano Banana 2 Lite for image generation → Gemini Omni Flash for video generation and conversational refinement → Drive for storage and sharing.

All three steps run from the same anycap CLI inside your Codex terminal session.

The Pipeline in Three Commands

# Step 1: Generate a keyframe image with Nano Banana 2 Lite (Gemini 3.1 Flash-Lite Image)
anycap image generate \
  --model nano-banana-2-lite \
  --prompt "Minimal product hero: dark card UI, floating interface elements, soft blue glow, professional" \
  -o keyframe.jpg

# Step 2: Animate with Gemini Omni Flash
anycap video generate \
  --model gemini-omni-flash-preview \
  --mode image-to-video \
  --param images=./keyframe.jpg \
  --prompt "Subtle parallax motion, interface elements drift gently, ambient light shift" \
  -o hero-v1.mp4

# Step 3: Upload to Drive
anycap drive upload hero-v1.mp4 --name "product-hero" --share

Three commands. One ecosystem. No separate accounts.

Why Google-Native Matters in Codex

Visual coherence across steps. Nano Banana 2 Lite and Gemini Omni Flash share the same Google model family. The visual style, color language, and rendering character carry across the image-to-video transition without prompt engineering to align aesthetics between different vendors.

Conversational refinement after generation. Once Gemini Omni Flash animates your keyframe, you can refine the clip in plain language — "slow the motion," "add warmth," "hold at the end" — without regenerating from scratch. No other model in the catalog offers this on top of an image-to-video pipeline.

Single auth, unified billing. One anycap API key covers both Nano Banana 2 Lite and Gemini Omni Flash. No separate Google AI Studio account, no Vertex AI credentials, no second billing dashboard.

Model Reference

The Full Pipeline with Conversational Refinement

The image-to-video step is just the start. After generating the first video pass, Gemini Omni Flash lets you refine through natural language:

# Generate keyframe
anycap image generate \
  --model nano-banana-2-lite \
  --prompt "SaaS feature launch: dashboard UI, highlighted new feature panel, professional dark mode" \
  -o feature-hero.jpg

# First video pass
anycap video generate \
  --model gemini-omni-flash-preview \
  --mode image-to-video \
  --param images=./feature-hero.jpg \
  --prompt "Smooth reveal of the feature panel, professional product demo pacing" \
  -o feature-v1.mp4

# Conversational refinement pass 1
anycap video edit \
  --input feature-v1.mp4 \
  --instruction "Slow the reveal slightly and add a brief pause at peak exposure of the panel" \
  --model gemini-omni-flash-preview \
  -o feature-v2.mp4

# Conversational refinement pass 2
anycap video edit \
  --input feature-v2.mp4 \
  --instruction "Warm the color grade and add a subtle vignette at the edges" \
  --model gemini-omni-flash-preview \
  -o feature-v3.mp4

# Upload final version
anycap drive upload feature-v3.mp4 --name "feature-launch-hero" --share

The creative direction discovered and refined through conversational editing — then the final clip delivered to Drive with a shareable link.

A Complete Codex Automation Script

Here's a full automation pipeline a Codex agent can run as a post-deployment step:

#!/bin/bash
# Google-native Codex video pipeline
# Usage: bash google-pipeline.sh "feature name" "visual description"

FEATURE="$1"
VISUAL="${2:-clean SaaS dashboard, dark mode, minimal UI}"

echo "🖼️  Generating keyframe with Nano Banana 2 Lite..."
anycap image generate \
  --model nano-banana-2-lite \
  --prompt "${VISUAL}, professional product hero composition" \
  -o /tmp/keyframe.jpg

echo "🎬 Animating with Gemini Omni Flash..."
anycap video generate \
  --model gemini-omni-flash-preview \
  --mode image-to-video \
  --param images=/tmp/keyframe.jpg \
  --prompt "${FEATURE} feature reveal, smooth animation, professional product demo" \
  -o /tmp/v1.mp4

echo "✏️  Refining: slowing motion and warming tone..."
anycap video edit \
  --input /tmp/v1.mp4 \
  --instruction "Slow the reveal motion slightly and warm the color grade" \
  --model gemini-omni-flash-preview \
  -o /tmp/final.mp4

echo "☁️  Uploading to Drive..."
LINK=$(anycap drive upload /tmp/final.mp4 --name "${FEATURE}-launch-video" --share)
echo "✅ Done: ${LINK}"

Run it: bash google-pipeline.sh "analytics-v2" "SaaS analytics dashboard, data visualization cards"

Codex triggers this after any feature deployment — no human in the loop, one shareable video link returned.

When to Use the Google-Native Pipeline vs Mixed Models

Use the Google-native pipeline when:

• You want visual coherence between image and video without cross-vendor prompt alignment
• You're in an iterative creative workflow where conversational editing saves time
• You prefer to minimize the number of model vendors in your stack
• You're already using Nano Banana 2 Lite for other image generation tasks in Codex

Use a mixed pipeline instead when:

• You need Veo 3.1's higher quality ceiling for final deliverables (use Nano Banana 2 Lite for the keyframe, then Veo 3.1 for the video)
• The clip needs to be longer than 8 seconds (switch to Kling 3 for the video step)
• You need maximum photorealism for human subjects (Sora 2 Pro for the video step)

The Google-native pipeline is not the only answer — but it's the cleanest when Google models are already your default.

FAQ

What is Nano Banana 2 Lite? Nano Banana 2 Lite is AnyCap's model ID for Gemini 3.1 Flash-Lite Image — Google's fast, cost-efficient image generation model. It shares the same model family as Gemini Omni Flash, making it the natural image generation partner for Google-native Codex pipelines.

Do I need a Google account to use this pipeline? No. AnyCap handles authentication for both Nano Banana 2 Lite and Gemini Omni Flash. Your existing anycap credentials cover both models with no separate Google AI Studio or Vertex AI setup.

What's the image-to-video quality like compared to text-to-video? Image-to-video with a Nano Banana 2 Lite keyframe typically produces more consistent and intentional output than text-to-video alone, because the keyframe gives Gemini Omni Flash an exact starting composition to animate from rather than interpreting the full scene from text.

Can I use a different image model for the keyframe? Yes. While Nano Banana 2 Lite is the Google-native pairing, Gemini Omni Flash accepts any image as input — GPT Image 2, Seedream 5, or any PNG/JPG file. The Google-native pipeline is a recommendation, not a requirement.

How does conversational editing work after image-to-video? Exactly the same as after text-to-video. Use anycap video edit --input <previous-output.mp4> --instruction "...". The model applies the described change to the animated clip while maintaining the original scene context from the keyframe.

What's Next

• How to Use Gemini Omni Flash in Codex — the full Gemini Omni Flash model guide
• Conversational Video Editing AI — how the edit loop works in depth
• Gemini Omni Flash vs Veo 3.1 in Codex — when to upgrade to Veo 3.1's quality ceiling for the final render
• How to Generate Video with Codex (2026) — complete Codex video setup