diff --git a/paperbanana/core/config.py b/paperbanana/core/config.py
index b23c843b..c6d902bc 100644
--- a/paperbanana/core/config.py
+++ b/paperbanana/core/config.py
@@ -73,6 +73,7 @@ class Settings(BaseSettings):
 
     # API Keys (loaded from environment)
     google_api_key: Optional[str] = Field(default=None, alias="GOOGLE_API_KEY")
+    openrouter_api_key: Optional[str] = Field(default=None, alias="OPENROUTER_API_KEY")
 
     # SSL
     skip_ssl_verification: bool = Field(default=False, alias="SKIP_SSL_VERIFICATION")
diff --git a/paperbanana/providers/image_gen/openrouter_imagen.py b/paperbanana/providers/image_gen/openrouter_imagen.py
new file mode 100644
index 00000000..c9bc157b
--- /dev/null
+++ b/paperbanana/providers/image_gen/openrouter_imagen.py
@@ -0,0 +1,137 @@
+"""OpenRouter image generation provider — uses any image model via the OpenAI-compatible API."""
+
+from __future__ import annotations
+
+import base64
+import re
+from io import BytesIO
+from typing import Optional
+
+import structlog
+from PIL import Image
+from tenacity import retry, stop_after_attempt, wait_exponential
+
+from paperbanana.providers.base import ImageGenProvider
+
+logger = structlog.get_logger()
+
+
+class OpenRouterImageGen(ImageGenProvider):
+    """Image generation routed through OpenRouter.
+
+    Talks to models that support ``modalities: ["image", "text"]``
+    (e.g. google/gemini-3-pro-image-preview) and returns a PIL Image
+    decoded from the base64 response.
+
+    Get an API key at https://openrouter.ai/keys
+    """
+
+    def __init__(
+        self,
+        api_key: Optional[str] = None,
+        model: str = "google/gemini-3-pro-image-preview",
+    ):
+        self._api_key = api_key
+        self._model = model
+        self._client = None
+
+    @property
+    def name(self) -> str:
+        return "openrouter_imagen"
+
+    @property
+    def model_name(self) -> str:
+        return self._model
+
+    def _get_client(self):
+        """Lazy-init an async httpx client pointed at the OpenRouter API."""
+        if self._client is None:
+            import httpx
+
+            self._client = httpx.AsyncClient(
+                base_url="https://openrouter.ai/api/v1",
+                headers={
+                    "Authorization": f"Bearer {self._api_key}",
+                    "HTTP-Referer": "https://github.com/llmsresearch/paperbanana",
+                    "X-Title": "PaperBanana",
+                },
+                # Image generation can take a while
+                timeout=180.0,
+            )
+        return self._client
+
+    def is_available(self) -> bool:
+        return self._api_key is not None
+
+    def _aspect_ratio_hint(self, width: int, height: int) -> str:
+        """Turn pixel dimensions into a human-readable aspect ratio hint for the prompt."""
+        ratio = width / height
+        if ratio > 1.5:
+            return "wide landscape format (16:9)"
+        if ratio > 1.2:
+            return "landscape format (3:2)"
+        if ratio < 0.67:
+            return "tall portrait format (9:16)"
+        if ratio < 0.83:
+            return "portrait format (2:3)"
+        return "square format (1:1)"
+
+    @retry(stop=stop_after_attempt(3), wait=wait_exponential(min=2, max=30))
+    async def generate(
+        self,
+        prompt: str,
+        negative_prompt: Optional[str] = None,
+        width: int = 1024,
+        height: int = 1024,
+        seed: Optional[int] = None,
+    ) -> Image.Image:
+        client = self._get_client()
+
+        # OpenRouter doesn't have native aspect-ratio params like the Google SDK,
+        # so we bake the desired format into the prompt itself.
+        aspect_hint = self._aspect_ratio_hint(width, height)
+        full_prompt = f"{prompt}\n\nGenerate this as a {aspect_hint} image."
+        if negative_prompt:
+            full_prompt += f"\n\nAvoid: {negative_prompt}"
+
+        payload = {
+            "model": self._model,
+            "messages": [
+                {"role": "user", "content": full_prompt},
+            ],
+            # This tells OpenRouter we want an image back, not just text
+            "modalities": ["image", "text"],
+        }
+
+        if seed is not None:
+            payload["seed"] = seed
+
+        response = await client.post("/chat/completions", json=payload)
+        response.raise_for_status()
+        data = response.json()
+
+        message = data["choices"][0]["message"]
+
+        # Primary path: images come as base64 data-URLs in the "images" array
+        images = message.get("images", [])
+        if images:
+            for img_entry in images:
+                url = img_entry.get("image_url", {}).get("url", "")
+                if url.startswith("data:image/"):
+                    b64_data = url.split(",", 1)[1]
+                    image_bytes = base64.b64decode(b64_data)
+                    return Image.open(BytesIO(image_bytes))
+
+        # Fallback: some models inline the base64 data directly in the text content
+        content = message.get("content", "")
+        if "data:image/" in content:
+            match = re.search(r"data:image/[^;]+;base64,([A-Za-z0-9+/=]+)", content)
+            if match:
+                image_bytes = base64.b64decode(match.group(1))
+                return Image.open(BytesIO(image_bytes))
+
+        logger.error("No image data in OpenRouter response", model=self._model)
+        raise ValueError(
+            f"OpenRouter response for {self._model} did not contain image data. "
+            f"Content preview: {content[:200]}"
+        )
diff --git a/paperbanana/providers/registry.py b/paperbanana/providers/registry.py
index de57766c..f8f3247b 100644
--- a/paperbanana/providers/registry.py
+++ b/paperbanana/providers/registry.py
@@ -26,8 +26,15 @@ def create_vlm(settings: Settings) -> VLMProvider:
                 api_key=settings.google_api_key,
                 model=settings.vlm_model,
             )
+        elif provider == "openrouter":
+            from paperbanana.providers.vlm.openrouter import OpenRouterVLM
+
+            return OpenRouterVLM(
+                api_key=settings.openrouter_api_key,
+                model=settings.vlm_model,
+            )
         else:
-            raise ValueError(f"Unknown VLM provider: {provider}. Available: gemini")
+            raise ValueError(f"Unknown VLM provider: {provider}. Available: gemini, openrouter")
 
     @staticmethod
     def create_image_gen(settings: Settings) -> ImageGenProvider:
@@ -42,5 +49,14 @@ def create_image_gen(settings: Settings) -> ImageGenProvider:
                 api_key=settings.google_api_key,
                 model=settings.image_model,
             )
+        elif provider == "openrouter_imagen":
+            from paperbanana.providers.image_gen.openrouter_imagen import OpenRouterImageGen
+
+            return OpenRouterImageGen(
+                api_key=settings.openrouter_api_key,
+                model=settings.image_model,
+            )
         else:
-            raise ValueError(f"Unknown image provider: {provider}. Available: google_imagen")
+            raise ValueError(
+                f"Unknown image provider: {provider}. Available: google_imagen, openrouter_imagen"
+            )
diff --git a/paperbanana/providers/vlm/openrouter.py b/paperbanana/providers/vlm/openrouter.py
new file mode 100644
index 00000000..4605857c
--- /dev/null
+++ b/paperbanana/providers/vlm/openrouter.py
@@ -0,0 +1,111 @@
+"""OpenRouter VLM provider — OpenAI-compatible API for any model."""
+
+from __future__ import annotations
+
+from typing import Optional
+
+import structlog
+from PIL import Image
+from tenacity import retry, stop_after_attempt, wait_exponential
+
+from paperbanana.core.utils import image_to_base64
+from paperbanana.providers.base import VLMProvider
+
+logger = structlog.get_logger()
+
+
+class OpenRouterVLM(VLMProvider):
+    """VLM provider that routes through OpenRouter's OpenAI-compatible API.
+
+    Works with any model on OpenRouter (Gemini, Claude, GPT, Llama, etc.).
+    Get an API key at https://openrouter.ai/keys
+    """
+
+    def __init__(
+        self,
+        api_key: Optional[str] = None,
+        model: str = "google/gemini-3-flash-preview",
+    ):
+        self._api_key = api_key
+        self._model = model
+        self._client = None
+
+    @property
+    def name(self) -> str:
+        return "openrouter"
+
+    @property
+    def model_name(self) -> str:
+        return self._model
+
+    def _get_client(self):
+        """Lazy-init an async httpx client pointed at the OpenRouter API."""
+        if self._client is None:
+            import httpx
+
+            self._client = httpx.AsyncClient(
+                base_url="https://openrouter.ai/api/v1",
+                headers={
+                    "Authorization": f"Bearer {self._api_key}",
+                    "HTTP-Referer": "https://github.com/llmsresearch/paperbanana",
+                    "X-Title": "PaperBanana",
+                },
+                timeout=120.0,
+            )
+        return self._client
+
+    def is_available(self) -> bool:
+        return self._api_key is not None
+
+    @retry(stop=stop_after_attempt(3), wait=wait_exponential(min=2, max=30))
+    async def generate(
+        self,
+        prompt: str,
+        images: Optional[list[Image.Image]] = None,
+        system_prompt: Optional[str] = None,
+        temperature: float = 1.0,
+        max_tokens: int = 4096,
+        response_format: Optional[str] = None,
+    ) -> str:
+        client = self._get_client()
+
+        messages = []
+        if system_prompt:
+            messages.append({"role": "system", "content": system_prompt})
+
+        # Build multimodal content array (vision images + text)
+        content = []
+        if images:
+            for img in images:
+                b64 = image_to_base64(img)
+                content.append(
+                    {
+                        "type": "image_url",
+                        "image_url": {"url": f"data:image/png;base64,{b64}"},
+                    }
+                )
+        content.append({"type": "text", "text": prompt})
+        messages.append({"role": "user", "content": content})
+
+        payload = {
+            "model": self._model,
+            "messages": messages,
+            "temperature": temperature,
+            "max_tokens": max_tokens,
+        }
+
+        if response_format == "json":
+            payload["response_format"] = {"type": "json_object"}
+
+        response = await client.post("/chat/completions", json=payload)
+        response.raise_for_status()
+
+        data = response.json()
+        text = data["choices"][0]["message"]["content"]
+
+        logger.debug(
+            "OpenRouter response",
+            model=self._model,
+            usage=data.get("usage"),
+        )
+        return text