How to Configure Image Generation in OpenClaw After nano-banana-pro Removal

Mar 24, 2026

The Problem: Image Generation Stopped Working

I upgraded OpenClaw to version 2026.3.22-beta.1 and suddenly my image generation workflows broke. My agents couldn’t create or edit images anymore, and I couldn’t figure out why.

When I tried to generate images, I got errors like this:

Error: image_generate tool not configured
Skill wrapper 'nano-banana-pro' not found

I checked my configuration, and everything looked fine. I had the nano-banana-pro skill wrapper set up exactly as the documentation showed. What went wrong?

Discovery: The Breaking Change

After digging through the changelog, I found the culprit. The 2026.3.22-beta.1 release included this breaking change:

Tools/image generation: standardize the stock image create/edit path on the core image_generate tool. The old nano-banana-pro docs/examples are gone.
Skills/image generation: remove the bundled nano-banana-pro skill wrapper. Use agents.defaults.imageGenerationModel.primary: 'google/gemini-3-pro-image-preview' for the native path.

The entire nano-banana-pro skill wrapper was removed. This means my old configuration was completely broken.

My Old Configuration (Now Broken)

Here’s what my configuration looked like before the update:

agents:
  skills:
    - name: image-generation
      type: nano-banana-pro
      config:
        model: gemini
        temperature: 0.7
        max_tokens: 2048

This configuration worked perfectly before the update. After the upgrade, it just stopped working. The skill wrapper was gone, and there was no fallback.

The Trial-and-Error Process

Attempt 1: Keeping the Old Config

First, I tried keeping my old configuration and hoping it would work. Bad idea. I got immediate errors about the missing skill wrapper.

Error: Skill 'nano-banana-pro' not found
Available skills: (empty list)

Attempt 2: Looking for Migration Guide

I searched for migration documentation but found nothing specific about this change. The old docs were gone, and the new docs only showed the final configuration without explaining the migration path.

Attempt 3: Reddit Community Help

I found a Reddit thread discussing this exact issue. The community confirmed what the changelog suggested:

Users who copied the nano-banana-pro sample-skill config will find it broken after upgrade. The new path is cleaner: set agents.defaults.imageGenerationModel.primary and built-in image generation just works.

The Solution: Single Line Configuration

After understanding the change, the fix was surprisingly simple. I removed the entire skill wrapper block and replaced it with a single configuration line:

agents:
  defaults:
    imageGenerationModel:
      primary: 'google/gemini-3-pro-image-preview'

That’s it. One line replaces what used to be a complex skill wrapper configuration.

Complete Working Example

Here’s my full agent configuration now:

agents:
  defaults:
    imageGenerationModel:
      primary: 'google/gemini-3-pro-image-preview'

  myAgent:
    model: claude-3-5-sonnet
    tools:
      - image_generate  # Now uses the configured model
      - web_search
      - code_execute

The image_generate tool now automatically uses the configured Gemini model. No skill wrapper needed.

Environment Configuration

Before testing, I made sure my API credentials were set:

# Option 1: Direct Gemini API access
export GOOGLE_API_KEY="your-google-api-key"

# Option 2: Use OpenRouter for Gemini access
export OPENROUTER_API_KEY="your-openrouter-key"

Testing the Configuration

I tested with a simple prompt to verify everything worked:

Generate an image of a sunset over mountains

The image generated successfully. The new configuration worked perfectly.

Why This Change Matters

The removal of nano-banana-pro represents a significant simplification in OpenClaw’s architecture:

Simpler Setup: One config line replaces a complex skill wrapper
Native Integration: Image generation is now a first-class core feature
Better Maintainability: Fewer moving parts means fewer things to break
Standardized Tool: Consistent image_generate tool for all operations

The old approach required maintaining a separate skill wrapper, configuration templates, and documentation. The new approach integrates image generation directly into the core toolchain.

Common Mistakes to Avoid

Mistake 1: Mixing Old and New Configs

agents:
  defaults:
    imageGenerationModel:
      primary: 'google/gemini-3-pro-image-preview'
  skills:
    - name: image-generation
      type: nano-banana-pro  # This will cause conflicts

Don’t mix the old skill wrapper with the new configuration. Remove all nano-banana-pro references completely.

Mistake 2: Incorrect Model Identifier

agents:
  defaults:
    imageGenerationModel:
      primary: 'gemini-3-pro-image-preview'  # WRONG - missing prefix

Use the exact model identifier: 'google/gemini-3-pro-image-preview'. Typos will cause silent failures.

Mistake 3: Missing API Credentials

The configuration won’t work without proper API access. Make sure you have:

# Verify environment variables are set
echo $GOOGLE_API_KEY
echo $OPENROUTER_API_KEY

# Test API access
curl -H "Authorization: Bearer $GOOGLE_API_KEY" \
  https://generativelanguage.googleapis.com/v1/models

Mistake 4: Ignoring the Change

If your image generation silently stopped working after an upgrade, this is likely the cause. Don’t assume the old skill will continue working.

Migration Checklist

Here’s the checklist I followed to ensure a smooth migration:

Identify all agent configs using nano-banana-pro
Remove nano-banana-pro skill wrapper configurations
Add agents.defaults.imageGenerationModel.primary setting
Verify API credentials for Gemini access
Test image generation with simple prompts
Test image edit operations if applicable
Update any documentation referencing old config
Archive or delete old skill wrapper files

Troubleshooting Tips

If your image generation still doesn’t work after the migration:

Check API keys: Ensure GOOGLE_API_KEY or OPENROUTER_API_KEY is set
Verify model name: Use exact string 'google/gemini-3-pro-image-preview'
Restart OpenClaw: Some configuration changes require a restart
Check logs: Look for specific error messages about image generation
Test API directly: Verify Gemini API access works outside OpenClaw

What About Custom Skill Wrappers?

If you had custom configurations in your nano-banana-pro skill wrapper, you might need to adjust. The new approach doesn’t support the same level of customization through the skill wrapper. However, most common configurations are now handled through environment variables or the core tool configuration.

The Bottom Line

The removal of nano-banana-pro simplifies OpenClaw image generation significantly. What previously required a complex skill wrapper now needs only a single configuration line. If you’re upgrading to OpenClaw 2026.3.22-beta.1 or later, make sure to:

Remove all nano-banana-pro references from your configs
Add agents.defaults.imageGenerationModel.primary: 'google/gemini-3-pro-image-preview'
Verify your API credentials are set
Test image generation to confirm everything works

This change aligns with OpenClaw’s broader push toward cleaner, more maintainable configurations. The native path is simpler, more reliable, and easier to debug.

Final Words + More Resources

My intention with this article was to help others share my knowledge and experience. If you want to contact me, you can contact by email: Email me

Here are also the most important links from this article along with some further resources that will help you in this scope:

👨‍💻 OpenClaw Documentation
👨‍💻 Reddit Discussion: OpenClaw Release Notes

Oh, and if you found these resources useful, don’t forget to support me by starring the repo on GitHub!