Image Generation

Wunderland agents can generate images from text prompts using OpenAI, OpenRouter, Stability AI, or Replicate. The agent gets a single generate_image tool that routes to whichever provider you configure.

Providers

Provider	Example Models	Strengths
OpenAI	dall-e-3, gpt-image-1	Strong prompt adherence, text rendering, polished defaults
OpenRouter	Routed image-capable models	Provider routing and unified account management
Stability AI	stable-image-core, stable-image-ultra, sd3-large	Granular diffusion controls and seedable generations
Replicate	black-forest-labs/flux-schnell, flux-dev, community models	Broad model catalog and provider-native low-level inputs

API Keys

Set the key for your preferred provider:

# OpenAI
export OPENAI_API_KEY=sk-...

# OpenRouter
export OPENROUTER_API_KEY=sk-or-...

# Stability AI
export STABILITY_API_KEY=sk-...

# Replicate
export REPLICATE_API_TOKEN=r8_...

Any combination can be set simultaneously. The extension uses its configured default provider and falls back to the first available credentialed provider.

Auto-Detection

The image generation extension loads automatically when one of these is present:

• OPENAI_API_KEY
• OPENROUTER_API_KEY
• STABILITY_API_KEY
• REPLICATE_API_TOKEN

No need to run wunderland extensions enable in the common case. If you want to force the extension on with only non-default provider credentials present, you can still enable it explicitly:

wunderland extensions enable image-generation

The generate_image Tool

The agent exposes a single tool with these options:

Parameter	Required	Default	Description
prompt	Yes	—	Text description of the image to generate
size	No	1024x1024	Output dimensions
aspectRatio	No	Provider default	Explicit aspect ratio hint such as 1:1, 16:9, or 9:16
quality	No	standard	Image quality level
style	No	vivid	Visual style
provider	No	Configured default	Override the provider for this request
model	No	Provider default	Use a provider-native model id for this request
seed	No	Random	Reproducible generation for providers that support seeds
negativePrompt	No	—	Exclude traits or artifacts for providers that support it

Sizes

Size	Aspect Ratio	Use Case
1024x1024	1:1 (square)	Social media posts, profile images
1792x1024	16:9 (landscape)	Blog headers, presentations, desktop wallpapers
1024x1792	9:16 (portrait)	Mobile wallpapers, stories, vertical content

Quality

Quality	Description	Cost Impact
standard	Good for most use cases — fast and affordable	1x
hd	Higher detail and consistency, especially for complex scenes	~2x

Style

Style	Description
vivid	Hyper-real, dramatic lighting, saturated colors
natural	Photographic, muted, realistic tones

Deep provider-native controls still live in the SDK/runtime layer. If you need raw Stability presets or Replicate model inputs from code, use AgentOS generateImage() with namespaced providerOptions.

Configuration

Default Provider

Set your preferred provider globally:

wunderland extensions configure

The CLI provider-default picker supports:

• openai
• openrouter
• stability
• replicate

Or in ~/.wunderland/config.json:

{
  "providerDefaults": {
    "imageGeneration": "openai"
  }
}

Per-Agent Override

Override the global default in agent.config.json:

{
  "extensionOverrides": {
    "image-generation": {
      "options": { "defaultProvider": "stability" }
    }
  }
}

Example Usage

In a chat session:

You: Generate an image of a rabbit wearing VR goggles,
     floating in a neon-lit cyberpunk city

Agent: [calls generate_image with prompt, size: 1024x1024,
        quality: standard, style: vivid]

       Here's your image: [generated_image.png]
       A rabbit in VR goggles hovering above a cyberpunk cityscape
       with neon signs and rain-slicked streets.

Requesting specific options:

You: Create a landscape HD image of a minimalist Japanese garden,
     natural style

Agent: [calls generate_image with prompt, size: 1792x1024,
        quality: hd, style: natural]

       Here's your image: [generated_image.png]

Pricing

Pricing varies materially by provider, model, and quality setting. Check the provider’s current pricing page before relying on cost assumptions in production.

Need Lower-Level Provider Controls?

Wunderland exposes a single generate_image tool for agents. If you need direct provider-native knobs from code, use AgentOS generateImage() with namespaced providerOptions and keep Wunderland for orchestration:

import { generateImage } from '@framers/agentos';

const result = await generateImage({
  model: 'stability:stable-image-core',
  prompt: 'A brutalist cabin in alpine fog',
  providerOptions: {
    stability: {
      stylePreset: 'photographic',
      seed: 42,
      cfgScale: 8,
    },
  },
});

Troubleshooting

"Image generation not available" — Check that one of OPENAI_API_KEY, OPENROUTER_API_KEY, STABILITY_API_KEY, or REPLICATE_API_TOKEN is set. Run wunderland extensions info image-generation to see key status.

"Content policy violation" — Both providers enforce content safety policies. Rephrase the prompt to avoid restricted content categories.

Slow generation — Larger sizes, HD quality, and routed/community models can take noticeably longer. Start with smaller sizes and standard quality until you know your provider’s latency profile.