[Enhancement]: Add image generation model selection for Gemini Image Gen tool

[rajsodhi-2]

What features would you like to see added?

Is your feature request related to a problem?

Currently, the gemini_image_gen tool uses a single model determined by the GEMINI_IMAGE_MODEL environment variable. Users cannot switch between different image generation models (e.g., Nano Banana Pro vs Nano Banana Flash vs Imagen) without restarting the container.

This limits flexibility when users want to:

• Use a faster model (gemini-2.5-flash-image) for quick drafts
• Use a higher-quality model (gemini-3-pro-image-preview) for final outputs
• Compare outputs between models in the same session

Describe the solution you'd like

Add a model selection option for image generation tools, either:

Option 1: Per-agent configuration
In the agent builder UI, allow selecting which image model the gemini_image_gen tool should use (similar to how chat models are selected).

Option 2: Tool kwargs support
Allow tool_kwargs to override the image model, so different agents can use different models without code changes:

tools:
  - gemini_image_gen
tool_kwargs:
  gemini_image_gen:
    model: "gemini-3-pro-image-preview"

Option 3: Multiple registered tools
Register variants like gemini_image_gen_pro, gemini_image_gen_fast, imagen_ultra as separate selectable tools.

Describe alternatives you've considered

• Manually switching GEMINI_IMAGE_MODEL env var and restarting (current workaround - disruptive)
• Creating duplicate tool files with hardcoded models (requires code modification)

Additional context

Google now offers multiple image generation models with different quality/speed/cost tradeoffs:

Model	API Name	Use Case
Nano Banana Pro	gemini-3-pro-image-preview	Highest quality, ~30s generation time
Nano Banana	gemini-2.5-flash-image	Fast drafts, ~8s generation time
Imagen 4	imagen-4	Cost-effective option
Imagen 4 Ultra	imagen-4-ultra	High quality alternative

Being able to select between these at runtime would significantly improve the image generation workflow.

Thank you for considering this feature request!

More details

Technical Context

The current implementation is in api/app/clients/tools/structured/GeminiImageGen.js, line 433:

const geminiModel = process.env.GEMINI_IMAGE_MODEL || 'gemini-2.5-flash-image';

This hardcodes the model selection to an environment variable with no runtime override capability.

Suggested Implementation Approach

The cleanest solution might be to:

1. Check for a model override in tool_kwargs first
2. Fall back to the environment variable
3. Fall back to the default

Something like:

const geminiModel = fields.model || process.env.GEMINI_IMAGE_MODEL || 'gemini-2.5-flash-image';

Then expose model as a configurable field in the agent builder UI when the gemini_image_gen tool is selected.

Reference

The OpenAI image tools (OpenAIImageTools.js) face a similar limitation - this enhancement could potentially apply to both Google and OpenAI image generation tools for consistency.

Which components are impacted by your request?

Plugins

Pictures

No response

Code of Conduct

• I agree to follow this project's Code of Conduct