Nano Banana 2 in ComfyUI: The Complete Guide

Nano Banana 2 is an API-based image generation model powered by Google Gemini. Unlike SDXL or Flux, it does not run locally.

This means:

• No GPU required
• Very fast generation (3–5 seconds)
• High-quality outputs
• Advanced reasoning capabilities

Step 1 - Update ComfyUI

Make sure your ComfyUI installation is up to date.

cd ComfyUIgit pull

If you’re using the portable version, update it through your usual method.

Step 2 - Install Required Nodes

Nano Banana 2 is available through ComfyUI partner/API nodes.

If needed, install missing dependencies:

cd ComfyUI/custom_nodesgit clone https://github.com/comfyanonymous/ComfyUI-Manager

Restart ComfyUI and install API-related nodes from the manager.

Step 3 - Set Your API Key

Nano Banana requires an API key.

On Windows (PowerShell):

setx GEMINI_API_KEY "your_api_key_here"

On Linux / Mac:

exportGEMINI_API_KEY="your_api_key_here"

Restart ComfyUI after setting the key.

Step 4 - Basic Text-to-Image Workflow

Create this simple pipeline:

[Nano Banana 2] → [Save Image]

Example Prompt

A futuristic cyberpunk city at night, neon lights, rain reflections, ultra detailed, cinematic lighting, 4k

Recommended Settings

model: nano-banana-2aspect_ratio: 16:9resolution: 2048thinking_level: minimal

Run the workflow and you’ll get your image in a few seconds.

Minimal JSON Workflow Example

You can import this directly into ComfyUI:

{  "nodes": [    {      "id": 1,      "type": "NanoBanana2",      "params": {        "prompt": "A fantasy warrior standing on a mountain, golden sunset, ultra detailed",        "aspect_ratio": "1:1",        "resolution": 2048,        "thinking_level": "minimal"      }    },    {      "id": 2,      "type": "SaveImage",      "inputs": {        "images": [1]      }    }  ]}

Step 5 - Image-to-Image Editing

This is where Nano Banana becomes extremely powerful.

Workflow

[Load Image] → [Nano Banana 2] → [Save Image]

Example Prompt

Transform this image into a cyberpunk version with neon lighting, rain, and futuristic buildings

Use Cases

• Change art style
• Modify lighting
• Replace objects
• Translate text inside the image

Step 6 - Multi-Image Composition

You can combine multiple images into one.

Workflow

[Load Image] x2 → [Batch Images] → [Nano Banana 2] → [Save Image]

Example Prompt

Combine these characters into a single epic battle scene, cinematic lighting, highly detailed, fantasy style

This is useful for:

• Character consistency
• Scene building
• Style mixing

Understanding Key Parameters

Here are the most important settings you should know.

Prompt

A medieval castle on a cliff, dramatic sky, volumetric lighting, ultra realistic, 8k

Always be specific.

Thinking Level

minimal → fast, rough resultshigh → slower, more detailed and accurate

Use minimal for testing, high for final renders.

Aspect Ratio

1:1 → Instagram16:9 → YouTube / wallpapers9:16 → TikTok

Resolution

1024 → fast preview2048 → standard quality4096 → final render

Advanced Usage Tips

Prompt Structuring

Instead of writing short prompts:

cool futuristic city

Write structured prompts:

A futuristic cyberpunk city, neon lights, rain reflections, cinematic lighting, ultra detailed, wide angle lens, 4k

Iteration Strategy

Use this workflow:

minimal thinking + 1024 resolution → test ideas  high thinking + 2048/4096 → final image  

Combine with Other Nodes

You can extend your workflow:

[Nano Banana 2] → [Upscale] → [Color Correction] → [Save Image]

Performance and Cost

Typical stats:

Generation time: 3–5 secondsCost: ~$0.03 per imageResolution: up to 4K

Compared to local models:

• Faster
• More consistent
• No VRAM limitations

Common Issues

Node not appearing

→ Update ComfyUI→ Check API key→ Install missing nodes

No output image

→ Verify API key is set→ Restart ComfyUI→ Check console logs

Slow generation

→ Check internet connection→ Lower resolution→ Use minimal thinking level

Conclusion

Nano Banana 2 is one of the most practical tools you can use inside ComfyUI today.

It removes the usual constraints of local generation while providing:

• Fast results
• High-quality outputs
• Advanced editing capabilities

If you’re building modern AI workflows, this is a model you should definitely integrate.