diff --git a/EvoScientist/EvoScientist.py b/EvoScientist/EvoScientist.py
index 643e173..dabb7f0 100644
--- a/EvoScientist/EvoScientist.py
+++ b/EvoScientist/EvoScientist.py
@@ -430,7 +430,14 @@ def _build_base_kwargs(
     base_backend, base_middleware, *, cfg=None, chat_model=None, workspace_dir=None
 ):
     """Build agent kwargs *without* MCP (fast, no subprocess spawning)."""
-    from .tools import skill_manager, tavily_search, think_tool
+    from .tools import (
+        edit_image,
+        generate_image,
+        refresh_image_tool_descriptions,
+        skill_manager,
+        tavily_search,
+        think_tool,
+    )
     from .utils import load_subagents
 
     cfg = cfg if cfg is not None else _ensure_config()
@@ -438,7 +445,8 @@ def _build_base_kwargs(
     tool_registry = {"think_tool": think_tool}
     if os.environ.get("TAVILY_API_KEY"):
         tool_registry["tavily_search"] = tavily_search
-    base_tools = [think_tool]
+    refresh_image_tool_descriptions()
+    base_tools = [think_tool, generate_image, edit_image]
     if cfg.workspace_isolation != "required":
         base_tools.append(skill_manager)
 
@@ -485,7 +493,14 @@ def load_mcp_and_build_kwargs(
         chat_model: Explicit chat model to bind instead of
             ``_ensure_chat_model()`` (which would write module globals).
     """
-    from .tools import skill_manager, tavily_search, think_tool
+    from .tools import (
+        edit_image,
+        generate_image,
+        refresh_image_tool_descriptions,
+        skill_manager,
+        tavily_search,
+        think_tool,
+    )
     from .utils import load_subagents
 
     cfg = cfg if cfg is not None else _ensure_config()
@@ -503,7 +518,8 @@ def load_mcp_and_build_kwargs(
     tool_registry = {"think_tool": think_tool}
     if os.environ.get("TAVILY_API_KEY"):
         tool_registry["tavily_search"] = tavily_search
-    base_tools = [think_tool]
+    refresh_image_tool_descriptions()
+    base_tools = [think_tool, generate_image, edit_image]
     if cfg.workspace_isolation != "required":
         base_tools.append(skill_manager)
 
diff --git a/EvoScientist/tools/__init__.py b/EvoScientist/tools/__init__.py
index 5fe21d4..ac697ed 100644
--- a/EvoScientist/tools/__init__.py
+++ b/EvoScientist/tools/__init__.py
@@ -4,12 +4,16 @@ External imports like ``from EvoScientist.tools import tavily_search`` continue
 to work unchanged thanks to these re-exports.
 """
 
+from .image import edit_image, generate_image, refresh_image_tool_descriptions
 from .search import fetch_webpage_content, tavily_search
 from .skill_manager import skill_manager
 from .think import think_tool
 
 __all__ = [
+    "edit_image",
     "fetch_webpage_content",
+    "generate_image",
+    "refresh_image_tool_descriptions",
     "skill_manager",
     "tavily_search",
     "think_tool",
diff --git a/EvoScientist/tools/image.py b/EvoScientist/tools/image.py
new file mode 100644
index 0000000..1ee889b
--- /dev/null
+++ b/EvoScientist/tools/image.py
@@ -0,0 +1,132 @@
+"""Agent tools for image generation and editing.
+
+Thin wrappers over ``image_gen.service``. They never see API keys: entries
+resolve ``${ENV_VAR}`` references server-side at call time.
+"""
+
+from __future__ import annotations
+
+import json
+from pathlib import Path
+
+from langchain_core.tools import tool
+
+
+def _workspace() -> Path:
+    from EvoScientist.paths import _active_workspace
+
+    return Path(_active_workspace).resolve()
+
+
+def _settings_path() -> Path:
+    from EvoScientist.config.settings import get_config_path
+
+    return get_config_path()
+
+
+# Late-bound indirections so tests can patch without importing the service's
+# adapter stack.
+async def _generate_for_workspace(workspace: Path, **kwargs):
+    from EvoScientist.image_gen import service
+
+    return await service.generate_for_workspace(workspace, **kwargs)
+
+
+async def _edit_for_workspace(workspace: Path, **kwargs):
+    from EvoScientist.image_gen import service
+
+    return await service.edit_for_workspace(workspace, **kwargs)
+
+
+def _json_result(payload: dict) -> str:
+    return json.dumps(payload, ensure_ascii=False)
+
+
+def _available_models_hint() -> str:
+    try:
+        from EvoScientist.image_gen.config import load_image_generation_settings
+
+        settings = load_image_generation_settings(config_path=_settings_path())
+        names = [entry.display_name() for entry in settings.models]
+        if not names:
+            return ""
+        default = settings.default_model or settings.models[0].id
+        return (
+            f" Available image models: {', '.join(names)}. Default: {default}."
+        )
+    except Exception:
+        return ""
+
+
+def refresh_image_tool_descriptions() -> None:
+    """Refresh tool descriptions with the current image model list.
+
+    Called on every agent build so config.yaml edits take effect without a
+    restart. Descriptions must never include credentials.
+    """
+    hint = _available_models_hint()
+    generate_image.description = (
+        "Generate one or more images with a dedicated image model and save "
+        f"them under artifacts/.{hint}"
+    )
+    edit_image.description = (
+        "Edit an existing workspace image with a dedicated image model and "
+        f"save the result under artifacts/.{hint}"
+    )
+
+
+@tool
+async def generate_image(
+    prompt: str,
+    model: str | None = None,
+    size: str = "1024x1024",
+    quality: str = "auto",
+    background: str = "auto",
+    output_path: str | None = None,
+    n: int = 1,
+) -> str:
+    """Generate one or more images with a dedicated image model and save them under artifacts/."""
+    try:
+        result = await _generate_for_workspace(
+            _workspace(),
+            prompt=prompt,
+            model=model,
+            size=size,
+            quality=quality,
+            background=background,
+            output_path=output_path,
+            n=n,
+        )
+        return _json_result(result)
+    except Exception as exc:
+        return _json_result({"ok": False, "error": str(exc) or exc.__class__.__name__})
+
+
+@tool
+async def edit_image(
+    image_path: str,
+    prompt: str,
+    model: str | None = None,
+    mask_path: str | None = None,
+    size: str = "1024x1024",
+    quality: str = "auto",
+    output_path: str | None = None,
+) -> str:
+    """Edit an existing workspace image with a dedicated image model and save the result under artifacts/."""
+    try:
+        result = await _edit_for_workspace(
+            _workspace(),
+            image_path=image_path,
+            prompt=prompt,
+            model=model,
+            mask_path=mask_path,
+            size=size,
+            quality=quality,
+            output_path=output_path,
+        )
+        return _json_result(result)
+    except Exception as exc:
+        return _json_result({"ok": False, "error": str(exc) or exc.__class__.__name__})
+
+
+refresh_image_tool_descriptions()
diff --git a/tests/test_image_tools.py b/tests/test_image_tools.py
new file mode 100644
index 0000000..9527557
--- /dev/null
+++ b/tests/test_image_tools.py
@@ -0,0 +1,93 @@
+"""Tests for the image generation agent tools."""
+
+import json
+
+import pytest
+
+from EvoScientist.tools import image as image_tools
+
+
+@pytest.fixture(autouse=True)
+def fake_service(monkeypatch, tmp_path):
+    """Patch the service entry points the tools call."""
+    async def fake_generate(workspace, **kwargs):
+        return {
+            "ok": True,
+            "path": "artifacts/generated_1.png",
+            "paths": ["artifacts/generated_1.png"],
+            "mime_type": "image/png",
+            "model": "gpt-image-2",
+            "size": kwargs.get("size", "1024x1024"),
+        }
+
+    async def fake_edit(workspace, **kwargs):
+        return {
+            "ok": True,
+            "path": "artifacts/edited_1.png",
+            "paths": ["artifacts/edited_1.png"],
+            "mime_type": "image/png",
+            "model": "gpt-image-2",
+            "size": kwargs.get("size", "1024x1024"),
+        }
+
+    monkeypatch.setattr(image_tools, "_generate_for_workspace", fake_generate)
+    monkeypatch.setattr(image_tools, "_edit_for_workspace", fake_edit)
+    monkeypatch.setattr(image_tools, "_workspace", lambda: tmp_path)
+
+
+async def test_generate_image_success():
+    result = json.loads(
+        await image_tools.generate_image.ainvoke({"prompt": "a cat"})
+    )
+    assert result["ok"] is True
+    assert result["paths"] == ["artifacts/generated_1.png"]
+
+
+async def test_edit_image_success():
+    result = json.loads(
+        await image_tools.edit_image.ainvoke(
+            {"image_path": "artifacts/src.png", "prompt": "make blue"}
+        )
+    )
+    assert result["ok"] is True
+    assert result["paths"] == ["artifacts/edited_1.png"]
+
+
+async def test_error_becomes_json(monkeypatch):
+    from EvoScientist.image_gen.adapters.base import ImageGenError
+
+    async def boom(workspace, **kwargs):
+        raise ImageGenError("no image models configured")
+
+    monkeypatch.setattr(image_tools, "_generate_for_workspace", boom)
+    result = json.loads(await image_tools.generate_image.ainvoke({"prompt": "x"}))
+    assert result["ok"] is False
+    assert "no image models" in result["error"]
+
+
+def test_refresh_descriptions_includes_models(monkeypatch, tmp_path):
+    config = tmp_path / "config.yaml"
+    config.write_text(
+        """
+image_generation:
+  default_model: gpt-image-2
+  models:
+    - id: gpt-image-2
+      name: GPT Image 2
+      provider: openai
+      api_key: sk-x
+""",
+        encoding="utf-8",
+    )
+    monkeypatch.setattr(image_tools, "_settings_path", lambda: config)
+    image_tools.refresh_image_tool_descriptions()
+    assert "GPT Image 2" in image_tools.generate_image.description
+    assert "Default: gpt-image-2" in image_tools.generate_image.description
+    assert "sk-x" not in image_tools.generate_image.description
+
+
+def test_tools_package_exports():
+    from EvoScientist import tools
+
+    assert tools.generate_image is image_tools.generate_image
+    assert tools.edit_image is image_tools.edit_image