diff --git a/py/packages/genkit-google-genai/pyproject.toml b/py/packages/genkit-google-genai/pyproject.toml
index c2237af7c9..7315583c0a 100644
--- a/py/packages/genkit-google-genai/pyproject.toml
+++ b/py/packages/genkit-google-genai/pyproject.toml
@@ -50,11 +50,10 @@ dependencies = [
   "structlog>=25.2.0",
   "strenum>=0.4.15; python_version < '3.11'",
 ]
-description = "Gemini, Imagen, and Veo on Genkit"
+description = "Gemini and Veo on Genkit"
 keywords = [
   "genkit",
   "gemini",
-  "imagen",
   "veo",
   "vertex-ai",
   "google",
diff --git a/py/packages/genkit-google-genai/src/genkit_google_genai/__init__.py b/py/packages/genkit-google-genai/src/genkit_google_genai/__init__.py
index 3f3259c581..cdb4542a85 100644
--- a/py/packages/genkit-google-genai/src/genkit_google_genai/__init__.py
+++ b/py/packages/genkit-google-genai/src/genkit_google_genai/__init__.py
@@ -95,7 +95,6 @@
     KnownGemma,
     VertexAIGeminiVersion,
 )
-from genkit_google_genai.models.imagen import ImagenConfigSchema, ImagenVersion, KnownImagen
 from genkit_google_genai.models.interactions_lyria import LyriaConfig
 from genkit_google_genai.models.interactions_registry import LyriaVersion
 from genkit_google_genai.models.veo import KnownVeo, VeoConfig, VeoVersion
@@ -121,13 +120,10 @@ def package_name() -> str:
     'DeepResearchConfig',
     'GoogleAI',
     'GoogleAIGeminiVersion',
-    'ImagenConfigSchema',
-    'ImagenVersion',
     'KnownGemini',
     'KnownGeminiImage',
     'KnownGeminiTts',
     'KnownGemma',
-    'KnownImagen',
     'KnownVeo',
     'LyriaConfig',
     'LyriaVersion',
diff --git a/py/packages/genkit-google-genai/src/genkit_google_genai/google.py b/py/packages/genkit-google-genai/src/genkit_google_genai/google.py
index a363832683..22e078b144 100644
--- a/py/packages/genkit-google-genai/src/genkit_google_genai/google.py
+++ b/py/packages/genkit-google-genai/src/genkit_google_genai/google.py
@@ -22,7 +22,7 @@
 Google GenAI SDK to detect and register available models at runtime.
 
 Supported capabilities include text generation (Gemini/Gemma), text embeddings,
-image generation (Imagen), and video generation (Veo).
+image generation (Gemini native image), and video generation (Veo).
 
 Example:
     ```python
@@ -92,7 +92,11 @@
     family_embedder_ref,
     family_model_ref,
 )
-from genkit_google_genai.models._routing import is_unroutable_model_id, strip_ref_prefixes
+from genkit_google_genai.models._routing import (
+    is_unroutable_model_id,
+    is_unsupported_image_model_name,
+    strip_ref_prefixes,
+)
 from genkit_google_genai.models.antigravity import AntigravityConfig, create_antigravity_action
 from genkit_google_genai.models.deep_research import (
     DeepResearchConfig,
@@ -122,15 +126,6 @@
     is_tts_model,
     is_tuned_gemini_name,
 )
-from genkit_google_genai.models.imagen import (
-    SUPPORTED_MODELS as IMAGE_SUPPORTED_MODELS,
-    ImagenConfigSchema,
-    ImagenModel,
-    KnownImagen,
-    is_imagen_model_name,
-    is_unsupported_image_model_name,
-    vertexai_image_model_info,
-)
 from genkit_google_genai.models.interactions_lyria import (
     LyriaConfig as InteractionsLyriaConfig,
     create_lyria_action,
@@ -163,20 +158,17 @@ class GenaiModels:
 
     Attributes:
         gemini: List of Gemini/Gemma model names (generateContent action).
-        imagen: List of Imagen model names (predict action, Vertex AI only).
         embedders: List of embedding model names (embedContent action).
         veo: List of Veo video generation model names (generateVideos action).
     """
 
     gemini: list[str]
-    imagen: list[str]
     embedders: list[str]
     veo: list[str]
 
     def __init__(self) -> None:
         """Initialize empty model lists."""
         self.gemini = []
-        self.imagen = []
         self.embedders = []
         self.veo = []
 
@@ -192,18 +184,16 @@ def _list_genai_models(client: genai.Client, is_vertex: bool) -> GenaiModels:
     - Google AI populates each model's ``supported_actions`` field, so models
       are categorized by action:
         - 'embedContent' action → embedders
-        - 'predict' + Imagen name (``imagen-``) → imagen
         - 'generateVideos' or Veo name (``veo-``) → veo
         - 'generateContent' + 'gemini'/'gemma' in name → gemini
     - Vertex AI returns ``supported_actions = None`` for every publisher model,
       so categorizing by action would skip them all. The Vertex path instead
       categorizes by model name:
-        - Imagen name (``imagen-``) → imagen
         - Veo name (``veo-``) → veo
         - 'gemini'/'gemma' in name (and not an embedding) → gemini
-      Ids with no working generate path here (``imagegeneration@*``,
-      ``imagetext@*``, ``virtual-try-on-*``) are not categorized at all, so
-      they are never advertised or registered.
+      Ids with no working generate path here (``imagen-*``,
+      ``imagegeneration@*``, ``imagetext@*``, ``virtual-try-on-*``) are not
+      categorized at all, so they are never advertised or registered.
       Embedders are intentionally NOT discovered here. The Vertex catalog
       over-lists embedders that are published but not callable, so they
       are advertised from a curated list (``VERTEX_KNOWN_EMBEDDERS``) instead.
@@ -248,8 +238,6 @@ def _list_genai_models(client: genai.Client, is_vertex: bool) -> GenaiModels:
                 continue
             elif is_unsupported_image_model_name(name):
                 continue
-            elif is_imagen_model_name(name):
-                models.imagen.append(name)
             elif is_veo_model(name):
                 models.veo.append(name)
             elif 'gemini' in lower_name or 'gemma' in lower_name:
@@ -263,10 +251,6 @@ def _list_genai_models(client: genai.Client, is_vertex: bool) -> GenaiModels:
         if 'embedContent' in m.supported_actions:
             models.embedders.append(name)
 
-        # Imagen (imagen- prefix, not a bare "image" substring)
-        if 'predict' in m.supported_actions and is_imagen_model_name(name):
-            models.imagen.append(name)
-
         # Veo
         if 'generateVideos' in m.supported_actions or is_veo_model(name):
             models.veo.append(name)
@@ -500,21 +484,6 @@ def gemma_model(
             config=config,
         )
 
-    @classmethod
-    def imagen_model(
-        cls, name: KnownImagen | str, *, config: ImagenConfigSchema | None = None
-    ) -> ModelRef[ImagenConfigSchema]:
-        """Typed ref for an Imagen model (``imagen-…``)."""
-        return family_model_ref(
-            name,
-            namespace=cls.name,
-            plugin_class=cls.__name__,
-            family='imagen',
-            method='imagen_model',
-            config_schema=ImagenConfigSchema,
-            config=config,
-        )
-
     @classmethod
     def veo_model(cls, name: KnownVeo | str, *, config: VeoConfig | None = None) -> ModelRef[VeoConfig]:
         """Typed ref for a Veo video model (``veo-…``)."""
@@ -574,7 +543,6 @@ class GoogleAI(GoogleFamilyRefs, Plugin):
         | Type | Action Kind | Example |
         |---|---|---|
         | Gemini / Gemma | MODEL | ``googleai/gemini-flash-latest`` |
-        | Imagen | MODEL | ``googleai/imagen-3.0-generate-002`` |
         | Embedders | EMBEDDER | ``googleai/gemini-embedding-001`` |
         | Veo (Video) | BACKGROUND_MODEL | ``googleai/veo-3.1-generate-preview`` |
 
@@ -740,11 +708,6 @@ async def init(self) -> list[Action]:
             if action := self._resolve_model(googleai_name(name)):
                 actions.append(action)
 
-        # Imagen Models
-        for name in genai_models.imagen:
-            if action := self._resolve_model(googleai_name(name)):
-                actions.append(action)
-
         # Veo Models (background models)
         for name in genai_models.veo:
             actions.extend(_background_actions(self._resolve_veo_model(googleai_name(name))))
@@ -785,15 +748,12 @@ async def init(self) -> list[Action]:
         return actions
 
     def _list_known_models(self) -> list[Action]:
-        """List known Gemini and Imagen models as Action objects."""
+        """List known Gemini models as Action objects."""
         genai_models = _list_genai_models(self._runtime_client(), is_vertex=False)
         actions = []
         for name in genai_models.gemini:
             if action := self._resolve_model(googleai_name(name)):
                 actions.append(action)
-        for name in genai_models.imagen:
-            if action := self._resolve_model(googleai_name(name)):
-                actions.append(action)
         return actions
 
     def _list_known_veo_models(self) -> list[Action]:
@@ -901,15 +861,6 @@ def _resolve_model(self, name: str) -> Action | None:
             return None
         # One annotated closure per family. Action validates request.config
         # from the fn annotation; a single _run cannot switch schemas at runtime.
-        if is_imagen_model_name(clean_name):
-            model_info = vertexai_image_model_info(clean_name)
-            IMAGE_SUPPORTED_MODELS[clean_name] = model_info  # pyright: ignore[reportArgumentType]
-
-            async def _run_imagen(request: ModelRequest[ImagenConfigSchema], ctx: ActionRunContext) -> ModelResponse:
-                return await ImagenModel(clean_name, self._runtime_client()).generate(request, ctx)
-
-            return _model_action(name, _run_imagen, model_info, ImagenConfigSchema)
-
         model_info = google_model_info(clean_name)
         SUPPORTED_MODELS[clean_name] = model_info
 
@@ -976,15 +927,6 @@ async def list_actions(self) -> list[ActionMetadata]:
                 )
             )
 
-        for name in genai_models.imagen:
-            actions_list.append(
-                model_action_metadata(
-                    name=googleai_name(name),
-                    info=vertexai_image_model_info(name).model_dump(by_alias=True),
-                    config_schema=ImagenConfigSchema,
-                )
-            )
-
         for name in genai_models.veo:
             actions_list.append(_veo_background_action_metadata(googleai_name(name)))
 
@@ -1040,7 +982,7 @@ class VertexAI(GoogleFamilyRefs, Plugin):
     """Vertex AI plugin for Genkit with dynamic model discovery.
 
     This plugin provides access to Google Cloud Vertex AI models including
-    Gemini, Imagen, Veo, and embedders. Models are discovered dynamically,
+    Gemini, Veo, and embedders. Models are discovered dynamically,
     ensuring new models are available without SDK updates.
 
     Vertex AI vs Google AI:
@@ -1049,14 +991,12 @@ class VertexAI(GoogleFamilyRefs, Plugin):
         - Customer-managed encryption keys (CMEK)
         - Data residency controls
         - IAM-based access control
-        - Imagen image generation models
 
     Model Types:
         | Type | Action Kind | Example |
         |---|---|---|
         | Gemini / Gemma | MODEL | ``vertexai/gemini-flash-latest`` |
-        | Imagen | MODEL | ``vertexai/imagen-3.0-generate-002`` |
-        | Veo (Video) | BACKGROUND_MODEL | ``vertexai/veo-3.1-generate-preview`` |
+        | Veo (Video) | BACKGROUND_MODEL | ``vertexai/veo-3.1-generate-001`` |
         | Embedders | EMBEDDER | ``vertexai/text-embedding-005`` |
 
     Example:
@@ -1185,10 +1125,6 @@ async def init(self) -> list[Action]:
             if action := self._resolve_model(vertexai_name(name)):
                 actions.append(action)
 
-        for name in genai_models.imagen:
-            if action := self._resolve_model(vertexai_name(name)):
-                actions.append(action)
-
         # Veo Models (background models)
         for name in genai_models.veo:
             bg_action = self._resolve_veo_model(vertexai_name(name))
@@ -1226,9 +1162,6 @@ def _list_known_models(self) -> list[Action]:
         for name in genai_models.gemini:
             if action := self._resolve_model(vertexai_name(name)):
                 actions.append(action)
-        for name in genai_models.imagen:
-            if action := self._resolve_model(vertexai_name(name)):
-                actions.append(action)
         return actions
 
     def _list_known_veo_models(self) -> list[Action]:
@@ -1368,15 +1301,6 @@ async def _run_tuned(request: ModelRequest[GeminiConfigSchema], ctx: ActionRunCo
 
             return _model_action(name, _run_tuned, model_info, GeminiConfigSchema)
 
-        if is_imagen_model_name(clean_name):
-            model_info = vertexai_image_model_info(clean_name)
-            IMAGE_SUPPORTED_MODELS[clean_name] = model_info  # pyright: ignore[reportArgumentType]
-
-            async def _run_imagen(request: ModelRequest[ImagenConfigSchema], ctx: ActionRunContext) -> ModelResponse:
-                return await ImagenModel(clean_name, self._runtime_client()).generate(request, ctx)
-
-            return _model_action(name, _run_imagen, model_info, ImagenConfigSchema)
-
         model_info = google_model_info(clean_name)
         SUPPORTED_MODELS[clean_name] = model_info
 
@@ -1443,15 +1367,6 @@ async def list_actions(self) -> list[ActionMetadata]:
                 )
             )
 
-        for name in genai_models.imagen:
-            actions_list.append(
-                model_action_metadata(
-                    name=vertexai_name(name),
-                    info=vertexai_image_model_info(name).model_dump(by_alias=True),
-                    config_schema=ImagenConfigSchema,
-                )
-            )
-
         for name in genai_models.veo:
             actions_list.append(_veo_background_action_metadata(vertexai_name(name)))
 
diff --git a/py/packages/genkit-google-genai/src/genkit_google_genai/models/_model_refs.py b/py/packages/genkit-google-genai/src/genkit_google_genai/models/_model_refs.py
index 5679f6e7c8..12126fc933 100644
--- a/py/packages/genkit-google-genai/src/genkit_google_genai/models/_model_refs.py
+++ b/py/packages/genkit-google-genai/src/genkit_google_genai/models/_model_refs.py
@@ -17,7 +17,7 @@
 """Shared machinery for the typed family ref constructors.
 
 Each plugin class exposes one constructor per config family
-(``gemini_model``, ``imagen_model``, ...). The return type is the contract:
+(``gemini_model``, ``gemini_image_model``, ...). The return type is the contract:
 ``ModelRef[GeminiConfigSchema]`` tells generate-time code exactly which
 config is legal, so a constructor must refuse ids whose runtime action
 would validate a different schema — otherwise the ref lies and the wrong
@@ -41,7 +41,6 @@
     'tts': 'gemini_tts_model',
     'image': 'gemini_image_model',
     'gemma': 'gemma_model',
-    'imagen': 'imagen_model',
     'veo': 'veo_model',
     'embedder': 'embedding',
 }
@@ -65,7 +64,10 @@ def wrong_family_error(*, plugin_class: str, method: str, family: str, local: st
             else f"'{local}' has no ref constructor in this plugin."
         )
     elif actual == 'unsupported':
-        hint = f"'{local}' is not a supported model."
+        if local.lower().startswith('imagen-'):
+            hint = f"'{local}' is not a supported model; for image generation use {plugin_class}.gemini_image_model()."
+        else:
+            hint = f"'{local}' is not a supported model."
     elif actual in FAMILY_METHOD:
         hint = f"'{local}' is not a {family} model; use {plugin_class}.{FAMILY_METHOD[actual]}()."
     else:
diff --git a/py/packages/genkit-google-genai/src/genkit_google_genai/models/_routing.py b/py/packages/genkit-google-genai/src/genkit_google_genai/models/_routing.py
index de61e8e432..26c0d6c646 100644
--- a/py/packages/genkit-google-genai/src/genkit_google_genai/models/_routing.py
+++ b/py/packages/genkit-google-genai/src/genkit_google_genai/models/_routing.py
@@ -28,10 +28,6 @@
     is_image_model,
     is_tts_model,
 )
-from genkit_google_genai.models.imagen import (
-    is_imagen_model_name,
-    is_unsupported_image_model_name,
-)
 from genkit_google_genai.models.lyria import is_lyria_model
 from genkit_google_genai.models.veo import is_veo_model
 
@@ -49,8 +45,8 @@
 )
 
 # Families with no MODEL generate path on this plugin. Constructor
-# refusal is a different table — TTS, native image, Gemma, and Imagen
-# still resolve as MODEL.
+# refusal is a different table — TTS, native image, and Gemma still
+# resolve as MODEL.
 UNROUTABLE_FAMILIES = frozenset({
     'embedder',
     'unsupported',
@@ -74,6 +70,17 @@ def strip_ref_prefixes(name: str) -> str:
     return local
 
 
+def is_unsupported_image_model_name(name: str) -> bool:
+    """True for image ids with no generate path here.
+
+    Matches the ``imagen-``, ``imagegeneration@``, ``imagetext@`` and
+    ``virtual-try-on-`` prefixes on the last path segment. Gemini native
+    image (``gemini-…-image``) is a different family and routes normally.
+    """
+    local = name.split('/')[-1].lower()
+    return local.startswith(('imagen-', 'imagegeneration@', 'imagetext@', 'virtual-try-on-'))
+
+
 def classify_family(name: str) -> str:
     """Bucket a model id by the last path segment.
 
@@ -93,8 +100,6 @@ def classify_family(name: str) -> str:
         return 'deep-research'
     if leaf.startswith('antigravity-'):
         return 'antigravity'
-    if is_imagen_model_name(leaf):
-        return 'imagen'
     if is_tts_model(leaf):
         return 'tts'
     if is_image_model(leaf):
diff --git a/py/packages/genkit-google-genai/src/genkit_google_genai/models/gemini.py b/py/packages/genkit-google-genai/src/genkit_google_genai/models/gemini.py
index 0f17e8767d..fa6db95589 100644
--- a/py/packages/genkit-google-genai/src/genkit_google_genai/models/gemini.py
+++ b/py/packages/genkit-google-genai/src/genkit_google_genai/models/gemini.py
@@ -960,9 +960,9 @@ def is_tts_model(name: str) -> bool:
 def is_image_model(name: str) -> bool:
     """Check if the model is a Gemini native image generation model.
 
-    Native image is a ``gemini-`` name that contains ``-image``. Imagen
-    (``imagen-…``) is a different family — a bare ``image`` substring
-    would catch both.
+    Native image is a ``gemini-`` name that contains ``-image``. The
+    ``gemini-`` prefix is required: a bare ``image`` substring would also
+    match ``imagen-`` ids, which have no generate path here.
 
     Args:
         name: The model name to check.
diff --git a/py/packages/genkit-google-genai/src/genkit_google_genai/models/imagen.py b/py/packages/genkit-google-genai/src/genkit_google_genai/models/imagen.py
deleted file mode 100644
index 0b74ca0ac0..0000000000
--- a/py/packages/genkit-google-genai/src/genkit_google_genai/models/imagen.py
+++ /dev/null
@@ -1,291 +0,0 @@
-# Copyright 2025 Google LLC
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-#
-# SPDX-License-Identifier: Apache-2.0
-
-"""Imagen model implementation for Google GenAI plugin."""
-
-import base64
-import sys
-
-if sys.version_info < (3, 11):
-    from strenum import StrEnum
-else:
-    from enum import StrEnum
-
-import json
-from functools import cached_property
-from typing import Any, Literal, TypeAlias
-
-from google import genai
-from google.genai import types as genai_types
-from google.genai.errors import APIError
-from pydantic import BaseModel, ConfigDict, ValidationError
-
-from genkit import (
-    GenkitError,
-    Media,
-    MediaPart,
-    Message,
-    ModelInfo,
-    ModelRequest,
-    ModelResponse,
-    Part,
-    Role,
-    Supports,
-    TextPart,
-)
-from genkit.plugin_api import ActionRunContext, tracer, wrap_http_error
-from genkit_google_genai.models._sdk_config import (
-    attach_leftovers,
-    dump_family_config,
-    sdk_config_error,
-    split_sdk_fields,
-)
-
-
-def _to_dict(obj: Any) -> Any:  # noqa: ANN401
-    """Convert object to dict if it's a Pydantic model, otherwise return as-is."""
-    return obj.model_dump() if isinstance(obj, BaseModel) else obj
-
-
-class ImagenVersion(StrEnum):
-    """Supported text-to-image models."""
-
-    IMAGEN3 = 'imagen-3.0-generate-002'
-    IMAGEN3_FAST = 'imagen-3.0-fast-generate-001'
-
-
-# Quote autocomplete needs a Literal. The enum above is the catalog; a test
-# requires these members and the enum values to be the same set.
-KnownImagen: TypeAlias = Literal[
-    'imagen-3.0-generate-002',
-    'imagen-3.0-fast-generate-001',
-]
-
-
-SUPPORTED_MODELS = {
-    ImagenVersion.IMAGEN3: ModelInfo(
-        label='Vertex AI - Imagen3',
-        supports=Supports(
-            media=True,
-            multiturn=False,
-            tools=False,
-            system_role=True,
-            output=['media'],
-        ),
-    ),
-    ImagenVersion.IMAGEN3_FAST: ModelInfo(
-        label='Vertex AI - Imagen3 Fast',
-        supports=Supports(
-            media=False,
-            multiturn=False,
-            tools=False,
-            system_role=True,
-            output=['media'],
-        ),
-    ),
-}
-
-DEFAULT_IMAGE_SUPPORT = Supports(
-    media=True,
-    multiturn=False,
-    tools=False,
-    system_role=True,
-    output=['media'],
-)
-
-
-def is_imagen_model_name(name: str) -> bool:
-    """Return True if ``name`` is an Imagen model.
-
-    Imagen ids start with ``imagen-`` on the local name after stripping the
-    plugin / ``models/`` prefix. Gemini native image (``gemini-…-image``) is
-    not Imagen.
-    """
-    return name.split('/')[-1].lower().startswith('imagen-')
-
-
-def is_unsupported_image_model_name(name: str) -> bool:
-    """Return True for image ids that must not route anywhere.
-
-    ``imagegeneration@*`` and ``imagetext@*`` were shut down by Google in
-    June 2026, and ``virtual-try-on-*`` needs a person+product image request
-    shape this plugin does not implement. Letting any of those fall through
-    to the Gemini default would answer with the wrong model, so callers
-    treat these ids as not-a-model instead.
-    """
-    local = name.split('/')[-1].lower()
-    return local.startswith('imagegeneration@') or local.startswith('imagetext@') or local.startswith('virtual-try-on-')
-
-
-def vertexai_image_model_info(
-    version: str,
-) -> ModelInfo:
-    """Generates a ModelInfo object.
-
-    This function tries to get the best ModelInfo Supports
-    for the given version.
-
-    Args:
-        version: Version of the model.
-
-    Returns:
-        ModelInfo object.
-    """
-    return ModelInfo(
-        label=f'Vertex AI - {version}',
-        supports=DEFAULT_IMAGE_SUPPORT,
-    )
-
-
-class ImagenConfigSchema(BaseModel):
-    """Imagen Config Schema."""
-
-    model_config = ConfigDict(extra='allow')
-
-
-class ImagenModel:
-    """Imagen text-to-image model."""
-
-    def __init__(self, version: str | ImagenVersion, client: genai.Client) -> None:
-        """Initialize Imagen model.
-
-        Args:
-            version: Imagen version
-            client: Google AI client
-        """
-        self._version = version
-        self._client = client
-
-    def _build_prompt(self, request: ModelRequest) -> str:
-        """Build prompt request from Genkit request.
-
-        Args:
-            request: Genkit request.
-
-        Returns:
-            prompt for Imagen
-        """
-        prompt = []
-        for message in request.messages:
-            for part in message.content:
-                if isinstance(part.root, TextPart):
-                    prompt.append(part.root.text)
-                else:
-                    raise GenkitError(status='INVALID_ARGUMENT', message='Non-text messages are not supported')
-        return ' '.join(prompt)
-
-    async def generate(self, request: ModelRequest, _: ActionRunContext) -> ModelResponse:
-        """Handle a generation request.
-
-        Args:
-            request: The generation request containing messages and parameters.
-            _: action context
-
-        Returns:
-            The model's response to the generation request.
-        """
-        prompt = self._build_prompt(request)
-        config = self._get_config(request)
-        if request.tools:
-            raise GenkitError(status='UNIMPLEMENTED', message='Tools are not supported for this model.')
-
-        with tracer.start_as_current_span('generate_images') as span:
-            span.set_attribute(
-                'genkit:input',
-                json.dumps({
-                    'config': _to_dict(config),
-                    'contents': prompt,
-                    'model': self._version,
-                }),
-            )
-            try:
-                response = await self._client.aio.models.generate_images(
-                    model=self._version, prompt=prompt, config=config
-                )
-            except APIError as e:
-                raise wrap_http_error(e, status_code=e.code, message=e.message or str(e)) from e
-            span.set_attribute('genkit:output', json.dumps(_to_dict(response), default=str))
-
-        content = self._contents_from_response(response)
-
-        return ModelResponse(
-            message=Message(
-                content=content,
-                role=Role.MODEL,
-            )
-        )
-
-    def _get_config(self, request: ModelRequest) -> genai_types.GenerateImagesConfig | None:
-        dumped = dump_family_config(
-            config=request.config,
-            expected_type=ImagenConfigSchema,
-            action_name=self._version,
-        )
-        if not dumped:
-            return None
-
-        known, leftovers = split_sdk_fields(dumped, genai_types.GenerateImagesConfig)
-        try:
-            cfg = genai_types.GenerateImagesConfig(**known) if known else genai_types.GenerateImagesConfig()
-        except ValidationError as e:
-            raise sdk_config_error(action_name=self._version, error=e) from e
-        return attach_leftovers(cfg, leftovers, nest='parameters')
-
-    def _contents_from_response(self, response: genai_types.GenerateImagesResponse) -> list:
-        """Retrieve contents from google-genai response.
-
-        Args:
-            response: google-genai response.
-
-        Returns:
-            list of generated contents.
-        """
-        content = []
-        if response.generated_images:
-            for image in response.generated_images:
-                if image.image and image.image.image_bytes:
-                    b64_data = base64.b64encode(image.image.image_bytes).decode('utf-8')
-                    content.append(
-                        Part(
-                            root=MediaPart(
-                                media=Media(
-                                    url=f'data:{image.image.mime_type};base64,{b64_data}',
-                                    content_type=image.image.mime_type,
-                                )
-                            )
-                        )
-                    )
-
-        return content
-
-    @cached_property
-    def metadata(self) -> dict:
-        """Model metadata.
-
-        Returns:
-            model metadata.
-        """
-        supports = {}
-        if self._version in SUPPORTED_MODELS:
-            model_supports = SUPPORTED_MODELS[self._version].supports  # pyright: ignore[reportArgumentType]
-            if model_supports:
-                supports = model_supports.model_dump(by_alias=True)
-        else:
-            model_supports = vertexai_image_model_info(self._version).supports
-            if model_supports:
-                supports = model_supports.model_dump(by_alias=True)
-
-        return {'model': {'supports': supports}}
diff --git a/py/packages/genkit-google-genai/src/genkit_google_genai/models/veo.py b/py/packages/genkit-google-genai/src/genkit_google_genai/models/veo.py
index 0862ddfa6e..9329b8f04a 100644
--- a/py/packages/genkit-google-genai/src/genkit_google_genai/models/veo.py
+++ b/py/packages/genkit-google-genai/src/genkit_google_genai/models/veo.py
@@ -62,30 +62,27 @@ class VeoVersion(StrEnum):
     """Supported Veo video generation models.
 
     Note: Models are discovered dynamically. This enum provides convenience
-    constants for commonly used Veo models.
+    constants for commonly used Veo models. Google AI serves Veo 3.1 as the
+    ``-preview`` ids; Vertex AI serves it as the ``-001`` ids.
     """
 
-    VEO_2_0 = 'veo-2.0-generate-001'
-    VEO_2_0_EXP = 'veo-2.0-generate-exp'
-    VEO_3_0 = 'veo-3.0-generate-001'
-    VEO_3_0_FAST = 'veo-3.0-fast-generate-001'
     VEO_3_1_PREVIEW = 'veo-3.1-generate-preview'
     VEO_3_1_FAST_PREVIEW = 'veo-3.1-fast-generate-preview'
+    VEO_3_1_LITE_PREVIEW = 'veo-3.1-lite-generate-preview'
     VEO_3_1 = 'veo-3.1-generate-001'
     VEO_3_1_FAST = 'veo-3.1-fast-generate-001'
+    VEO_3_1_LITE = 'veo-3.1-lite-generate-001'
 
 
 # Quote autocomplete needs a Literal. The enum above is the catalog; a test
 # requires these members and the enum values to be the same set.
 KnownVeo: TypeAlias = Literal[
-    'veo-2.0-generate-001',
-    'veo-2.0-generate-exp',
-    'veo-3.0-generate-001',
-    'veo-3.0-fast-generate-001',
     'veo-3.1-generate-preview',
     'veo-3.1-fast-generate-preview',
+    'veo-3.1-lite-generate-preview',
     'veo-3.1-generate-001',
     'veo-3.1-fast-generate-001',
+    'veo-3.1-lite-generate-001',
 ]
 
 
diff --git a/py/packages/genkit-google-genai/test/google_plugin_test.py b/py/packages/genkit-google-genai/test/google_plugin_test.py
index 78bf78fd51..1c0f2ca17a 100644
--- a/py/packages/genkit-google-genai/test/google_plugin_test.py
+++ b/py/packages/genkit-google-genai/test/google_plugin_test.py
@@ -34,10 +34,6 @@
     GeminiConfigSchema,
     GeminiModel,
 )
-from genkit_google_genai.models.imagen import (
-    DEFAULT_IMAGE_SUPPORT,
-    SUPPORTED_MODELS as IMAGE_SUPPORTED_MODELS,
-)
 from google import genai
 from google.auth.credentials import Credentials
 from google.genai.types import HttpOptions
@@ -414,7 +410,7 @@ class MockModel:
         description: str = ''
 
     models_return_value = [
-        MockModel(supported_actions=['generateVideos'], name='models/veo-2.0-generate-001'),
+        MockModel(supported_actions=['generateVideos'], name='models/veo-3.1-generate-preview'),
     ]
 
     mock_client = MagicMock()
@@ -424,7 +420,7 @@ class MockModel:
     result = googleai_plugin_instance._list_known_veo_models()
 
     # Check Veo
-    action1 = next(a for a in result if a.name == googleai_name('veo-2.0-generate-001'))
+    action1 = next(a for a in result if a.name == googleai_name('veo-3.1-generate-preview'))
     assert action1 is not None
 
 
@@ -712,53 +708,42 @@ async def test_vertexai_resolve_action_embedder(
     'genkit_google_genai.models.gemini.google_model_info',
     new_callable=MagicMock,
 )
-@patch(
-    'genkit_google_genai.models.imagen.vertexai_image_model_info',
-    new_callable=MagicMock,
-)
 @pytest.mark.parametrize(
-    'model_name, expected_model_name, key, image',
+    'model_name, expected_model_name, key',
     [
         (
             'gemini-pro-deluxe-max',
             'vertexai/gemini-pro-deluxe-max',
             'gemini-pro-deluxe-max',
-            False,
         ),
         (
             'vertexai/gemini-pro-deluxe-max',
             'vertexai/gemini-pro-deluxe-max',
             'gemini-pro-deluxe-max',
-            False,
         ),
-        # A bare "image" prefix is not Imagen; only imagen- ids route there.
+        # A bare "image" substring still resolves as a Gemini text model.
         (
             'vertexai/image-gemini-pro-deluxe-max',
             'vertexai/image-gemini-pro-deluxe-max',
             'image-gemini-pro-deluxe-max',
-            False,
         ),
         (
             'image-gemini-pro-deluxe-max',
             'vertexai/image-gemini-pro-deluxe-max',
             'image-gemini-pro-deluxe-max',
-            False,
         ),
         (
             'gemini-pro-deluxe-max-image',
             'vertexai/gemini-pro-deluxe-max-image',
             'gemini-pro-deluxe-max-image',
-            False,
         ),
     ],
 )
 def test_vertexai__resolve_model(
     mock_google_model_info: MagicMock,
-    mock_vertexai_image_model_info: MagicMock,
     model_name: str,
     expected_model_name: str,
     key: str,
-    image: bool,
     vertexai_plugin_instance: VertexAI,
 ) -> None:
     """Tests for VertexAI._resolve_model method."""
@@ -770,21 +755,12 @@ def test_vertexai__resolve_model(
         supports=DEFAULT_SUPPORTS_MODEL,
     )
 
-    mock_vertexai_image_model_info.return_value = ModelInfo(
-        label=f'Vertex AI - {model_name}',
-        supports=DEFAULT_IMAGE_SUPPORT,
-    )
-
     action = plugin._resolve_model(name=expected_model_name)
 
     assert action is not None
     assert action.kind == ActionKind.MODEL
     assert action.name == expected_model_name
-
-    if image:
-        assert key in IMAGE_SUPPORTED_MODELS
-    else:
-        assert key in SUPPORTED_MODELS
+    assert key in SUPPORTED_MODELS
 
 
 @pytest.mark.parametrize(
@@ -821,19 +797,6 @@ def test_vertexai__resolve_embedder(
 @pytest.mark.asyncio
 async def test_vertexai_list_actions(vertexai_plugin_instance: VertexAI) -> None:
     """Unit test for list actions."""
-
-    @dataclass
-    class MockModel:
-        name: str
-        description: str = ''
-
-    [
-        MockModel(name='publishers/google/models/gemini-1.5-flash'),
-        MockModel(name='publishers/google/models/gemini-embedding-001'),
-        MockModel(name='publishers/google/models/imagen-3.0-generate-001'),
-        MockModel(name='publishers/google/models/veo-2.0-generate-001'),
-    ]
-
     mock_client = MagicMock()
     # Create sophisticated mocks that have supported_actions
     m1 = MagicMock()
@@ -848,11 +811,11 @@ class MockModel:
 
     m3 = MagicMock()
     m3.name = 'publishers/google/models/imagen-3.0-generate-001'
-    m3.supported_actions = ['predict']  # Imagen uses predict
+    m3.supported_actions = ['predict']
     m3.description = 'Imagen'
 
     m4 = MagicMock()
-    m4.name = 'publishers/google/models/veo-2.0-generate-001'
+    m4.name = 'publishers/google/models/veo-3.1-generate-001'
     m4.supported_actions = ['generateVideos']  # Veo uses generateVideos
     m4.description = 'Veo'
 
@@ -869,13 +832,11 @@ class MockModel:
     action2 = next(a for a in result if a.name == vertexai_name('gemini-embedding-001'))
     assert action2 is not None
 
-    # Verify Imagen
-    action3 = next(a for a in result if a.name == vertexai_name('imagen-3.0-generate-001'))
-    assert action3 is not None
-    assert action3.action_type == ActionKind.MODEL
+    # Imagen has no generate path here, so it is never advertised.
+    assert not any(a.name == vertexai_name('imagen-3.0-generate-001') for a in result)
 
     # Verify Veo
-    action4 = next(a for a in result if a.name == vertexai_name('veo-2.0-generate-001'))
+    action4 = next(a for a in result if a.name == vertexai_name('veo-3.1-generate-001'))
     assert action4 is not None
     # from genkit_google_genai.models.veo import VeoConfigSchema
     # assert action4.config_schema == VeoConfigSchema
@@ -903,7 +864,7 @@ def mock_model(name: str) -> MagicMock:
         mock_model('publishers/google/models/gemini-embedding-001'),
         mock_model('publishers/google/models/gemini-embedding-2'),
         mock_model('publishers/google/models/imagen-3.0-generate-002'),
-        mock_model('publishers/google/models/veo-2.0-generate-001'),
+        mock_model('publishers/google/models/veo-3.1-generate-001'),
     ]
     vertexai_plugin_instance._runtime_client = lambda: mock_client
 
@@ -912,9 +873,9 @@ def mock_model(name: str) -> MagicMock:
 
     # Gemini text model discovered despite supported_actions=None.
     assert vertexai_name('gemini-2.5-pro') in names
-    # Imagen and Veo discovered.
-    assert vertexai_name('imagen-3.0-generate-002') in names
-    assert vertexai_name('veo-2.0-generate-001') in names
+    # Veo discovered; Imagen is not advertised.
+    assert vertexai_name('veo-3.1-generate-001') in names
+    assert vertexai_name('imagen-3.0-generate-002') not in names
 
     # gemini-embedding-001 is registered as an embedder, not a gemini model.
     embedder = next(a for a in result if a.name == vertexai_name('gemini-embedding-001'))
@@ -928,10 +889,12 @@ async def test_googleai_resolve_background_model(googleai_plugin_instance: Googl
     """Test resolve action for background model."""
     plugin = googleai_plugin_instance
 
-    action = await plugin.resolve(action_type=ActionKind.BACKGROUND_MODEL, name=googleai_name('veo-2.0-generate-001'))
+    action = await plugin.resolve(
+        action_type=ActionKind.BACKGROUND_MODEL, name=googleai_name('veo-3.1-generate-preview')
+    )
     assert action is not None
     assert action.kind == ActionKind.BACKGROUND_MODEL
-    assert action.name == googleai_name('veo-2.0-generate-001')
+    assert action.name == googleai_name('veo-3.1-generate-preview')
 
 
 @pytest.mark.asyncio
@@ -940,29 +903,16 @@ async def test_googleai_resolve_check_operation(googleai_plugin_instance: Google
     plugin = googleai_plugin_instance
 
     action = await plugin.resolve(
-        action_type=ActionKind.CHECK_OPERATION, name=googleai_name('veo-2.0-generate-001/check')
+        action_type=ActionKind.CHECK_OPERATION, name=googleai_name('veo-3.1-generate-preview/check')
     )
     assert action is not None
     assert action.kind == ActionKind.CHECK_OPERATION
-    assert action.name == googleai_name('veo-2.0-generate-001/check')
+    assert action.name == googleai_name('veo-3.1-generate-preview/check')
 
 
 @pytest.mark.asyncio
 async def test_vertexai_list_known_models(vertexai_plugin_instance: VertexAI) -> None:
     """Unit test for list known models."""
-
-    @dataclass
-    class MockModel:
-        name: str
-        description: str = ''
-
-    [
-        MockModel(name='publishers/google/models/gemini-1.5-flash'),
-        MockModel(name='publishers/google/models/gemini-embedding-001'),
-        MockModel(name='publishers/google/models/imagen-3.0-generate-001'),
-        MockModel(name='publishers/google/models/veo-2.0-generate-001'),
-    ]
-
     mock_client = MagicMock()
     # Create sophisticated mocks that have supported_actions
     m1 = MagicMock()
@@ -981,7 +931,7 @@ class MockModel:
     m3.description = 'Imagen'
 
     m4 = MagicMock()
-    m4.name = 'publishers/google/models/veo-2.0-generate-001'
+    m4.name = 'publishers/google/models/veo-3.1-generate-001'
     m4.supported_actions = ['generateVideos']
     m4.description = 'Veo'
 
@@ -994,12 +944,11 @@ class MockModel:
     action1 = next(a for a in result if a.name == vertexai_name('gemini-1.5-flash'))
     assert action1 is not None
 
-    # Verify Imagen
-    action3 = next(a for a in result if a.name == vertexai_name('imagen-3.0-generate-001'))
-    assert action3 is not None
+    # Imagen has no generate path here, so it is not a known MODEL.
+    assert not any(a.name == vertexai_name('imagen-3.0-generate-001') for a in result)
 
     # Veo is background-only, so it is not a known generate MODEL.
-    assert not any(a.name == vertexai_name('veo-2.0-generate-001') for a in result)
+    assert not any(a.name == vertexai_name('veo-3.1-generate-001') for a in result)
 
     veo_actions = vertexai_plugin_instance._list_known_veo_models()
     assert {a.kind for a in veo_actions} == {ActionKind.BACKGROUND_MODEL, ActionKind.CHECK_OPERATION}
diff --git a/py/packages/genkit-google-genai/test/models/family_name_test.py b/py/packages/genkit-google-genai/test/models/family_name_test.py
index aa6f615722..1225e132b2 100644
--- a/py/packages/genkit-google-genai/test/models/family_name_test.py
+++ b/py/packages/genkit-google-genai/test/models/family_name_test.py
@@ -17,41 +17,21 @@
 """Family name checks for Google model routing."""
 
 import pytest
-from genkit_google_genai.models._routing import classify_family, is_unroutable_model_id
+from genkit_google_genai.models._routing import (
+    classify_family,
+    is_unroutable_model_id,
+    is_unsupported_image_model_name,
+)
 from genkit_google_genai.models.gemini import (
     is_gemini_model,
     is_gemma_model,
     is_image_model,
     is_tts_model,
 )
-from genkit_google_genai.models.imagen import (
-    is_imagen_model_name,
-    is_unsupported_image_model_name,
-)
 from genkit_google_genai.models.lyria import is_lyria_model
 from genkit_google_genai.models.veo import is_veo_model
 
 
-@pytest.mark.parametrize(
-    ('name', 'expected'),
-    [
-        ('imagen-3.0-generate-002', True),
-        ('IMAGEN-3.0-generate-002', True),
-        ('googleai/imagen-3.0-generate-002', True),
-        ('vertexai/imagen-3.0-generate-002', True),
-        ('models/imagen-3.0-generate-002', True),
-        ('gemini-2.5-flash-image', False),
-        ('gemini-2.5-flash-image-preview', False),
-        ('imagegeneration@006', False),
-        ('virtual-try-on-001', False),
-        ('veo-3.0-generate-001', False),
-    ],
-)
-def test_is_imagen_model_name(name: str, expected: bool) -> None:
-    """Imagen is the ``imagen-`` prefix only, on both plugins."""
-    assert is_imagen_model_name(name) is expected
-
-
 @pytest.mark.parametrize(
     ('name', 'expected'),
     [
@@ -62,13 +42,18 @@ def test_is_imagen_model_name(name: str, expected: bool) -> None:
         ('vertexai/imagetext@001', True),
         ('virtual-try-on-001', True),
         ('vertexai/virtual-try-on-001', True),
-        ('imagen-3.0-generate-002', False),
+        ('imagen-3.0-generate-002', True),
+        ('IMAGEN-4.0-generate-001', True),
+        ('googleai/imagen-4.0-generate-001', True),
+        ('models/imagen-4.0-ultra-generate-001', True),
         ('gemini-2.5-flash-image', False),
+        ('gemini-2.5-flash-image-preview', False),
         ('gemini-2.5-flash', False),
+        ('veo-3.1-generate-001', False),
     ],
 )
 def test_is_unsupported_image_model_name(name: str, expected: bool) -> None:
-    """Retired / unimplemented image ids fail closed instead of routing to Gemini."""
+    """Image ids with no generate path fail closed instead of routing to Gemini."""
     assert is_unsupported_image_model_name(name) is expected
 
 
@@ -137,9 +122,9 @@ def test_is_gemini_model(name: str, expected: bool) -> None:
 @pytest.mark.parametrize(
     ('name', 'expected'),
     [
-        ('veo-3.0-generate-001', True),
+        ('veo-3.1-generate-001', True),
         ('googleai/veo-3.1-generate-preview', True),
-        ('VEO-2.0-generate-001', True),
+        ('VEO-3.1-generate-001', True),
         ('gemini-2.0-flash', False),
         ('devotional-hymn', False),
     ],
@@ -174,7 +159,9 @@ def test_is_lyria_model(name: str, expected: bool) -> None:
         'imagegeneration@006',
         'imagetext@001',
         'virtual-try-on-001',
-        'veo-3.0-generate-001',
+        'imagen-3.0-generate-002',
+        'vertexai/imagen-4.0-generate-001',
+        'veo-3.1-generate-001',
         'googleai/lyria-002',
         'models/deep-research-pro-preview',
         'googleai/deep-research-pro-preview',
@@ -199,6 +186,7 @@ def test_unroutable_ids_fail_closed(name: str) -> None:
         ('publishers/google/models/antigravity-preview-05-2026', 'antigravity'),
         ('lyria-002', 'lyria'),
         ('imagetext@001', 'unsupported'),
+        ('imagen-4.0-generate-001', 'unsupported'),
     ],
 )
 def test_classify_family_uses_leaf_name(name: str, family: str) -> None:
@@ -211,7 +199,7 @@ def test_classify_family_uses_leaf_name(name: str, family: str) -> None:
     [
         'gemini-2.5-flash',
         'GEMINI-2.5-FLASH-PREVIEW-TTS',
-        'imagen-3.0-generate-002',
+        'gemini-2.5-flash-image',
         'gemma-3-12b-it',
     ],
 )
diff --git a/py/packages/genkit-google-genai/test/models/googlegenai_imagen_test.py b/py/packages/genkit-google-genai/test/models/googlegenai_imagen_test.py
deleted file mode 100644
index c6c7d096e0..0000000000
--- a/py/packages/genkit-google-genai/test/models/googlegenai_imagen_test.py
+++ /dev/null
@@ -1,136 +0,0 @@
-# Copyright 2025 Google LLC
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-#
-# SPDX-License-Identifier: Apache-2.0
-
-
-"""Tests for the Imagen model implementation."""
-
-import base64
-from unittest.mock import MagicMock
-
-import pytest
-from genkit_google_genai.models.imagen import ImagenConfigSchema, ImagenModel, ImagenVersion
-from google import genai
-from pytest_mock import MockerFixture
-
-from genkit import (
-    ActionRunContext,
-    GenkitError,
-    MediaPart,
-    Message,
-    ModelRequest,
-    ModelResponse,
-    Part,
-    Role,
-    TextPart,
-)
-
-
-@pytest.mark.asyncio
-@pytest.mark.parametrize('version', [x for x in ImagenVersion])
-async def test_generate_media_response(mocker: MockerFixture, version: ImagenVersion) -> None:
-    """Test generate method for media responses."""
-    request_text = 'response question'
-    response_byte_string = b'\x89PNG\r\n\x1a\n'
-    response_mimetype = 'image/png'
-
-    request = ModelRequest(
-        messages=[
-            Message(
-                role=Role.USER,
-                content=[
-                    Part(root=TextPart(text=request_text)),
-                ],
-            ),
-        ],
-    )
-
-    response_images = genai.types.GenerateImagesResponse(
-        generated_images=[
-            genai.types.GeneratedImage(
-                image=genai.types.Image(image_bytes=response_byte_string, mime_type=response_mimetype)
-            )
-        ]
-    )
-
-    googleai_client_mock = mocker.AsyncMock()
-    googleai_client_mock.aio.models.generate_images.return_value = response_images
-
-    imagen = ImagenModel(version, googleai_client_mock)
-
-    ctx = ActionRunContext()
-    response = await imagen.generate(request, ctx)
-
-    googleai_client_mock.assert_has_calls([
-        mocker.call.aio.models.generate_images(model=version, prompt=request_text, config=None)
-    ])
-    assert isinstance(response, ModelResponse)
-    assert response.message is not None
-    content = response.message.content[0]
-    assert isinstance(content.root, MediaPart)
-
-    assert content.root.media.content_type == response_mimetype
-
-    # Verify the data URL contains the correct base64-encoded content
-    # Data URLs have format: data:<mimetype>;base64,<data>
-    data_url = content.root.media.url
-    assert data_url.startswith(f'data:{response_mimetype};base64,')
-    encoded_data = data_url.split(',', 1)[1]
-    assert base64.b64decode(encoded_data) == response_byte_string
-
-
-def test_imagen_unknown_extra_rides_on_extra_body() -> None:
-    """Leftover keys ride on extra_body so a newly supported field still reaches the API."""
-    imagen = ImagenModel(ImagenVersion.IMAGEN3, MagicMock())
-    request = ModelRequest(
-        messages=[Message(role=Role.USER, content=[Part(root=TextPart(text='a cat'))])],
-        config=ImagenConfigSchema.model_validate({'fooBar': 1}),
-    )
-
-    cfg = imagen._get_config(request)
-
-    assert cfg is not None
-    assert cfg.http_options is not None
-    assert cfg.http_options.extra_body == {'parameters': {'fooBar': 1}}
-
-
-def test_imagen_rejects_raw_dicts() -> None:
-    """A dict at the dump leaf means Action never produced the family instance."""
-    imagen = ImagenModel(ImagenVersion.IMAGEN3, MagicMock())
-    request = ModelRequest(
-        messages=[Message(role=Role.USER, content=[Part(root=TextPart(text='a cat'))])],
-        config={'number_of_images': 1},  # type: ignore[arg-type]
-    )
-
-    with pytest.raises(GenkitError) as exc_info:
-        imagen._get_config(request)
-
-    assert exc_info.value.status == 'INVALID_ARGUMENT'
-    assert imagen._version in str(exc_info.value)
-
-
-def test_imagen_invalid_sdk_field_is_invalid_argument() -> None:
-    """SDK type errors become a named INVALID_ARGUMENT."""
-    imagen = ImagenModel(ImagenVersion.IMAGEN3, MagicMock())
-    request = ModelRequest(
-        messages=[Message(role=Role.USER, content=[Part(root=TextPart(text='a cat'))])],
-        config=ImagenConfigSchema.model_validate({'number_of_images': 'nope'}),
-    )
-
-    with pytest.raises(GenkitError) as exc_info:
-        imagen._get_config(request)
-
-    assert exc_info.value.status == 'INVALID_ARGUMENT'
-    assert 'number_of_images' in str(exc_info.value)
diff --git a/py/packages/genkit-google-genai/test/models/model_refs_test.py b/py/packages/genkit-google-genai/test/models/model_refs_test.py
index 02d9e37557..6263b736d1 100644
--- a/py/packages/genkit-google-genai/test/models/model_refs_test.py
+++ b/py/packages/genkit-google-genai/test/models/model_refs_test.py
@@ -16,10 +16,12 @@
 
 """Tests for the typed family ref constructors on GoogleAI / VertexAI."""
 
+import importlib
 from collections.abc import Callable
 from enum import Enum
 from typing import get_args
 
+import genkit_google_genai
 import pytest
 from genkit_google_genai import (
     GoogleAI,
@@ -27,7 +29,6 @@
     KnownGeminiImage,
     KnownGeminiTts,
     KnownGemma,
-    KnownImagen,
     KnownVeo,
     VertexAI,
 )
@@ -44,11 +45,6 @@
     is_image_model,
     is_tts_model,
 )
-from genkit_google_genai.models.imagen import (
-    ImagenConfigSchema,
-    ImagenVersion,
-    is_imagen_model_name,
-)
 from genkit_google_genai.models.veo import VeoConfig, VeoVersion, is_veo_model
 
 from genkit import GenkitError
@@ -72,23 +68,22 @@ def test_gemini_model_both_plugins(self) -> None:
     def test_enum_names_still_work(self) -> None:
         """The existing version enums remain valid constructor input."""
         assert GoogleAI.gemini_model(GoogleAIGeminiVersion.GEMINI_2_5_FLASH).name == 'googleai/gemini-2.5-flash'
-        assert GoogleAI.imagen_model(ImagenVersion.IMAGEN3).name == 'googleai/imagen-3.0-generate-002'
         assert GoogleAI.veo_model(VeoVersion.VEO_3_1_FAST_PREVIEW).name == 'googleai/veo-3.1-fast-generate-preview'
         assert VertexAI.veo_model(VeoVersion.VEO_3_1).name == 'vertexai/veo-3.1-generate-001'
+        assert GoogleAI.veo_model(VeoVersion.VEO_3_1_LITE_PREVIEW).name == 'googleai/veo-3.1-lite-generate-preview'
+        assert VertexAI.veo_model(VeoVersion.VEO_3_1_LITE).name == 'vertexai/veo-3.1-lite-generate-001'
 
     def test_family_constructors_type_their_config(self) -> None:
         """Each family constructor carries its own config schema."""
         tts = GoogleAI.gemini_tts_model('gemini-2.5-flash-preview-tts')
         image = VertexAI.gemini_image_model('gemini-2.5-flash-image')
         gemma = GoogleAI.gemma_model('gemma-3-12b-it')
-        imagen = VertexAI.imagen_model('imagen-3.0-generate-002')
         veo = GoogleAI.veo_model('veo-3.1-fast-generate-preview')
 
         assert tts.config_schema is GeminiTtsConfigSchema
         assert image.config_schema is GeminiImageConfigSchema
+        assert image.name == 'vertexai/gemini-2.5-flash-image'
         assert gemma.config_schema is GemmaConfigSchema
-        assert imagen.config_schema is ImagenConfigSchema
-        assert imagen.name == 'vertexai/imagen-3.0-generate-002'
         assert veo.config_schema is VeoConfig
         assert veo.name == 'googleai/veo-3.1-fast-generate-preview'
 
@@ -104,7 +99,7 @@ def test_unknown_id_allowed_on_gemini_model_and_embedding(self) -> None:
         assert GoogleAI.embedding('totally-new-embedder').name == 'googleai/totally-new-embedder'
 
         with pytest.raises(GenkitError):
-            GoogleAI.imagen_model('totally-new-model')
+            GoogleAI.gemma_model('totally-new-model')
 
 
 class TestStripThenPrefix:
@@ -161,7 +156,7 @@ class TestClosedRejectSet:
     @pytest.mark.parametrize(
         'bad_id',
         [
-            'veo-3.0-generate-001',  # wrong family: has its own constructor
+            'veo-3.1-generate-preview',  # wrong family: has its own constructor
             'lyria-002',  # no constructor in this plugin
             'googleai/lyria-002',  # prefix must not defeat the gate
             'deep-research-pro-preview',  # Interactions API family
@@ -169,7 +164,8 @@ class TestClosedRejectSet:
             'imagegeneration@006',  # retired June 2026
             'virtual-try-on-001',  # predict shape not implemented
             'gemini-embedding-001',  # embedder, not a generate model
-            'imagen-3.0-generate-002',  # wrong family: has its own constructor
+            'imagen-3.0-generate-002',  # not a supported model
+            'imagen-4.0-generate-001',  # not a supported model
             'gemini-2.5-flash-preview-tts',  # wrong family: TTS
             'gemini-2.5-flash-image',  # wrong family: native image
             'gemma-3-12b-it',  # wrong family: Gemma
@@ -186,11 +182,11 @@ def test_error_points_at_the_right_constructor(self) -> None:
         with pytest.raises(GenkitError, match=r'gemini_tts_model'):
             GoogleAI.gemini_model('gemini-2.5-flash-preview-tts')
         with pytest.raises(GenkitError, match=r'VertexAI\.gemini_model'):
-            VertexAI.imagen_model('gemini-2.5-flash')
+            VertexAI.gemma_model('gemini-2.5-flash')
         with pytest.raises(GenkitError, match=r'embedding'):
             GoogleAI.gemini_model('gemini-embedding-001')
         with pytest.raises(GenkitError, match=r'veo_model'):
-            GoogleAI.gemini_model('veo-3.0-generate-001')
+            GoogleAI.gemini_model('veo-3.1-generate-preview')
         with pytest.raises(GenkitError, match=r'lyria_model'):
             GoogleAI.gemini_model('lyria-002')
         with pytest.raises(GenkitError, match=r'deep_research_model'):
@@ -205,13 +201,15 @@ def test_error_points_at_the_right_constructor(self) -> None:
             GoogleAI.gemini_model('imagegeneration@006')
         with pytest.raises(GenkitError, match=r'is not a supported model'):
             GoogleAI.gemini_model('virtual-try-on-001')
-        with pytest.raises(GenkitError, match=r'is not a imagen model'):
-            GoogleAI.imagen_model('totally-new-model')
+        with pytest.raises(GenkitError, match=r'GoogleAI\.gemini_image_model'):
+            GoogleAI.gemini_model('imagen-4.0-generate-001')
+        with pytest.raises(GenkitError, match=r'VertexAI\.gemini_image_model'):
+            VertexAI.gemini_image_model('imagen-3.0-generate-002')
+        with pytest.raises(GenkitError, match=r'is not a gemma model'):
+            GoogleAI.gemma_model('totally-new-model')
 
     def test_family_constructors_reject_other_families(self) -> None:
         """Non-gemini constructors take only their own family ids."""
-        with pytest.raises(GenkitError):
-            GoogleAI.imagen_model('gemini-2.5-flash')
         with pytest.raises(GenkitError):
             GoogleAI.gemini_tts_model('gemini-2.5-flash')
         with pytest.raises(GenkitError):
@@ -274,7 +272,48 @@ def test_family_literals_match_their_catalog(self) -> None:
         assert set(get_args(KnownGeminiTts)) == _family_catalog(is_tts_model)
         assert set(get_args(KnownGeminiImage)) == _family_catalog(is_image_model)
         assert set(get_args(KnownGemma)) == _family_catalog(is_gemma_model)
-        assert set(get_args(KnownImagen)) == {str(member.value) for member in ImagenVersion}
-        assert all(is_imagen_model_name(value) for value in get_args(KnownImagen))
         assert set(get_args(KnownVeo)) == {str(member.value) for member in VeoVersion}
         assert all(is_veo_model(value) for value in get_args(KnownVeo))
+
+    def test_veo_catalog_is_the_3_1_family(self) -> None:
+        """The Veo catalog is the 3.1 family only: preview ids on Google AI, 001 ids on Vertex AI."""
+        assert {str(member.value) for member in VeoVersion} == {
+            'veo-3.1-generate-preview',
+            'veo-3.1-fast-generate-preview',
+            'veo-3.1-lite-generate-preview',
+            'veo-3.1-generate-001',
+            'veo-3.1-fast-generate-001',
+            'veo-3.1-lite-generate-001',
+        }
+
+
+class TestNoImagenSurface:
+    """Imagen ids have no constructor, no export, and no module."""
+
+    def test_no_imagen_exports(self) -> None:
+        """Neither the package nor the plugin classes expose an Imagen name."""
+        assert not {name for name in genkit_google_genai.__all__ if 'Imagen' in name}
+        assert not hasattr(GoogleAI, 'imagen_model')
+        assert not hasattr(VertexAI, 'imagen_model')
+
+    def test_no_imagen_module(self) -> None:
+        """The Imagen model module is gone."""
+        with pytest.raises(ModuleNotFoundError):
+            importlib.import_module('genkit_google_genai.models.imagen')
+
+    @pytest.mark.parametrize('plugin', [GoogleAI, VertexAI])
+    def test_every_constructor_rejects_imagen_ids(self, plugin: type[GoogleAI] | type[VertexAI]) -> None:
+        """An imagen- id is refused by every family constructor with the image hint."""
+        for method in ('gemini_model', 'gemini_tts_model', 'gemini_image_model', 'gemma_model'):
+            with pytest.raises(GenkitError, match=r'for image generation use \w+\.gemini_image_model\(\)'):
+                getattr(plugin, method)('imagen-4.0-generate-001')
+        with pytest.raises(GenkitError, match=r'for image generation use \w+\.gemini_image_model\(\)'):
+            plugin.embedding('imagen-4.0-generate-001')
+
+    @pytest.mark.parametrize('bad_id', ['imagegeneration@006', 'imagetext@001', 'virtual-try-on-001'])
+    def test_other_unsupported_ids_omit_the_image_hint(self, bad_id: str) -> None:
+        """Only imagen- ids are redirected to image generation."""
+        with pytest.raises(GenkitError) as exc_info:
+            GoogleAI.gemini_model(bad_id)
+        assert 'is not a supported model' in str(exc_info.value)
+        assert 'gemini_image_model' not in str(exc_info.value)
diff --git a/py/packages/genkit-google-genai/tests/google_genai_plugin_test.py b/py/packages/genkit-google-genai/tests/google_genai_plugin_test.py
index ed43d937c9..a2aa5abcb3 100644
--- a/py/packages/genkit-google-genai/tests/google_genai_plugin_test.py
+++ b/py/packages/genkit-google-genai/tests/google_genai_plugin_test.py
@@ -47,7 +47,6 @@
     GeminiTtsConfigSchema,
     GemmaConfigSchema,
 )
-from genkit_google_genai.models.imagen import ImagenConfigSchema
 from genkit_google_genai.models.veo import VeoConfig, VeoModel
 
 from genkit import ActionKind, Genkit, GenkitError, Message, ModelRequest, Part, Role, TextPart
@@ -83,7 +82,7 @@ def test_googleai_name() -> None:
 def test_vertexai_name() -> None:
     """Test vertexai_name helper function."""
     assert vertexai_name('gemini-2.0-flash') == 'vertexai/gemini-2.0-flash'
-    assert vertexai_name('imagen-3.0-generate-001') == 'vertexai/imagen-3.0-generate-001'
+    assert vertexai_name('gemini-2.5-flash-image') == 'vertexai/gemini-2.5-flash-image'
 
 
 def test_plugin_names() -> None:
@@ -179,7 +178,6 @@ def test_genai_models_container() -> None:
     """Test GenaiModels container initialization."""
     models = GenaiModels()
     assert models.gemini == []
-    assert models.imagen == []
     assert models.embedders == []
     assert models.veo == []
 
@@ -202,24 +200,10 @@ async def test_googleai_resolve_model(mock_list_models: MagicMock, mock_client:
 @patch('genkit_google_genai.google.genai.client.Client')
 @patch('genkit_google_genai.google._list_genai_models')
 @pytest.mark.asyncio
-async def test_googleai_resolve_imagen_model(mock_list_models: MagicMock, mock_client: MagicMock) -> None:
-    """Test GoogleAI plugin resolves Imagen image generation models."""
-    mock_list_models.return_value = GenaiModels()
-
-    plugin = GoogleAI(api_key='test-key')
-    action = await plugin.resolve(ActionKind.MODEL, 'googleai/imagen-3.0-generate-002')
-
-    assert action is not None
-    assert action.kind == ActionKind.MODEL
-    assert action.name == 'googleai/imagen-3.0-generate-002'
-    assert _custom_options(action) == to_json_schema(ImagenConfigSchema)
-
-
-@patch('genkit_google_genai.google.genai.client.Client')
-@patch('genkit_google_genai.google._list_genai_models')
-@pytest.mark.asyncio
-async def test_googleai_resolve_gemini_image_is_not_imagen(mock_list_models: MagicMock, mock_client: MagicMock) -> None:
-    """Native Gemini image models must not route through Imagen."""
+async def test_googleai_resolve_gemini_image_uses_image_config(
+    mock_list_models: MagicMock, mock_client: MagicMock
+) -> None:
+    """Native Gemini image models validate the image config schema."""
     mock_list_models.return_value = GenaiModels()
 
     plugin = GoogleAI(api_key='test-key')
@@ -241,7 +225,6 @@ async def test_googleai_resolve_gemini_image_is_not_imagen(mock_list_models: Mag
         ('googleai/gemini-2.5-flash-preview-tts', GeminiTtsConfigSchema),
         ('googleai/gemini-2.5-flash-image', GeminiImageConfigSchema),
         ('googleai/gemma-3-12b-it', GemmaConfigSchema),
-        ('googleai/imagen-3.0-generate-002', ImagenConfigSchema),
     ],
 )
 async def test_googleai_resolve_types_family_config(
@@ -271,7 +254,6 @@ async def test_googleai_resolve_types_family_config(
         ('vertexai/gemini-2.5-flash-preview-tts', GeminiTtsConfigSchema),
         ('vertexai/gemini-2.5-flash-image', GeminiImageConfigSchema),
         ('vertexai/gemma-3-12b-it', GemmaConfigSchema),
-        ('vertexai/imagen-3.0-generate-002', ImagenConfigSchema),
     ],
 )
 async def test_vertexai_resolve_types_family_config(
@@ -321,8 +303,8 @@ async def test_veo_start_types_family_config(mock_list_models: MagicMock, mock_c
     mock_list_models.return_value = GenaiModels()
 
     for plugin, name in (
-        (GoogleAI(api_key='test-key'), 'googleai/veo-3.0-generate-001'),
-        (VertexAI(project='test-project'), 'vertexai/veo-3.0-generate-001'),
+        (GoogleAI(api_key='test-key'), 'googleai/veo-3.1-generate-preview'),
+        (VertexAI(project='test-project'), 'vertexai/veo-3.1-generate-001'),
     ):
         action = await plugin.resolve(ActionKind.BACKGROUND_MODEL, name)
         assert action is not None
@@ -363,7 +345,7 @@ async def test_veo_action_run_dumps_leftover_and_stamps(mock_list_models: MagicM
     mock_client.return_value.aio.models.generate_videos = AsyncMock(return_value=op)
 
     plugin = VertexAI(project='test-project')
-    action = await plugin.resolve(ActionKind.BACKGROUND_MODEL, 'vertexai/veo-3.0-generate-001')
+    action = await plugin.resolve(ActionKind.BACKGROUND_MODEL, 'vertexai/veo-3.1-generate-001')
     assert action is not None
 
     started = await action.run({
@@ -378,7 +360,7 @@ async def test_veo_action_run_dumps_leftover_and_stamps(mock_list_models: MagicM
     assert cfg.duration_seconds == 5
     assert cfg.http_options is not None
     assert cfg.http_options.extra_body == {'parameters': {'fooBar': 1}}
-    assert started.response.action == '/background-model/vertexai/veo-3.0-generate-001'
+    assert started.response.action == '/background-model/vertexai/veo-3.1-generate-001'
 
 
 @patch('genkit_google_genai.google.genai.client.Client')
@@ -389,7 +371,7 @@ async def test_veo_action_run_rejects_bad_duration(mock_list_models: MagicMock,
     mock_list_models.return_value = GenaiModels()
 
     plugin = VertexAI(project='test-project')
-    action = await plugin.resolve(ActionKind.BACKGROUND_MODEL, 'vertexai/veo-3.0-generate-001')
+    action = await plugin.resolve(ActionKind.BACKGROUND_MODEL, 'vertexai/veo-3.1-generate-001')
     assert action is not None
 
     with pytest.raises(GenkitError) as exc_info:
@@ -410,46 +392,69 @@ async def test_veo_check_is_typed(mock_list_models: MagicMock, mock_client: Magi
     mock_list_models.return_value = GenaiModels()
 
     plugin = VertexAI(project='test-project')
-    action = await plugin.resolve(ActionKind.CHECK_OPERATION, 'vertexai/veo-3.0-generate-001/check')
+    action = await plugin.resolve(ActionKind.CHECK_OPERATION, 'vertexai/veo-3.1-generate-001/check')
     assert action is not None
     hints = get_type_hints(action._fn)  # noqa: SLF001
     assert hints['op'] is Operation
     assert hints['return'] is Operation
 
 
-@patch('genkit_google_genai.google.genai.client.Client')
-@patch('genkit_google_genai.google._list_genai_models')
-@pytest.mark.asyncio
-async def test_googleai_init_registers_imagen_models(mock_list_models: MagicMock, mock_client: MagicMock) -> None:
-    """Test GoogleAI init registers Imagen models from dynamic discovery."""
-    models = GenaiModels()
-    models.imagen = ['imagen-3.0-generate-002']
-    mock_list_models.return_value = models
+def test_list_genai_models_googleai_skips_imagen() -> None:
+    """A predict-only ``imagen-`` entry lands in no bucket."""
 
-    plugin = GoogleAI(api_key='test-key')
-    actions = await plugin.init()
+    def _model(name: str, actions: list[str]) -> MagicMock:
+        item = MagicMock()
+        item.name = name
+        item.supported_actions = actions
+        item.description = ''
+        return item
 
-    imagen_actions = [a for a in actions if 'imagen' in a.name]
-    assert len(imagen_actions) == 1
-    assert imagen_actions[0].name == 'googleai/imagen-3.0-generate-002'
-    assert imagen_actions[0].kind == ActionKind.MODEL
+    client = MagicMock()
+    client.models.list.return_value = [
+        _model('models/gemini-2.5-flash', ['generateContent']),
+        _model('models/imagen-4.0-generate-001', ['predict']),
+        _model('models/imagen-4.0-ultra-generate-001', ['predict', 'generateContent']),
+    ]
+    catalog = _list_genai_models(client, is_vertex=False)
+    assert vars(catalog) == {'gemini': ['gemini-2.5-flash'], 'embedders': [], 'veo': []}
 
 
 @patch('genkit_google_genai.google.genai.client.Client')
-@patch('genkit_google_genai.google._list_genai_models')
 @pytest.mark.asyncio
-async def test_googleai_list_actions_includes_imagen(mock_list_models: MagicMock, mock_client: MagicMock) -> None:
-    """Test GoogleAI list_actions includes Imagen models."""
-    models = GenaiModels()
-    models.imagen = ['imagen-3.0-generate-002']
-    mock_list_models.return_value = models
+@pytest.mark.parametrize('backend', ['googleai', 'vertexai'])
+async def test_list_actions_never_advertise_imagen(mock_client: MagicMock, backend: str) -> None:
+    """An ``imagen-`` id served by the API reaches neither list_actions nor init."""
 
-    plugin = GoogleAI(api_key='test-key')
-    actions_list = await plugin.list_actions()
+    def _model(name: str, actions: list[str] | None) -> MagicMock:
+        item = MagicMock()
+        item.name = name
+        item.supported_actions = actions
+        item.description = ''
+        return item
 
-    imagen_actions = [a for a in actions_list if 'imagen' in a.name]
-    assert len(imagen_actions) == 1
-    assert imagen_actions[0].name == 'googleai/imagen-3.0-generate-002'
+    if backend == 'googleai':
+        plugin: GoogleAI | VertexAI = GoogleAI(api_key='test-key')
+        listing = [
+            _model('models/gemini-2.5-flash', ['generateContent']),
+            _model('models/imagen-4.0-generate-001', ['predict']),
+        ]
+    else:
+        plugin = VertexAI(project='test-project')
+        listing = [
+            _model('publishers/google/models/gemini-2.5-flash', None),
+            _model('publishers/google/models/imagen-4.0-generate-001', None),
+        ]
+    mock_client.return_value.models.list.return_value = listing
+
+    listed = await plugin.list_actions()
+    registered = await plugin.init()
+
+    # The Gemini id proves the listing reached discovery, so the imagen
+    # assertions below are not passing on an empty catalog.
+    assert any('gemini-2.5-flash' in a.name for a in listed)
+    assert any('gemini-2.5-flash' in a.name for a in registered)
+    assert not any('imagen' in a.name for a in listed)
+    assert not any('imagen' in a.name for a in registered)
 
 
 @patch('genkit_google_genai.google.genai.client.Client')
@@ -503,6 +508,8 @@ async def test_vertexai_resolve_model(mock_list_models: MagicMock, mock_client:
         'virtual-try-on-001',
         'imagegeneration@006',
         'imagetext@001',
+        'imagen-3.0-generate-002',
+        'imagen-4.0-generate-001',
         'lyria-002',
         'deep-research-pro-preview',
         'gemini-embedding-001',
@@ -531,6 +538,8 @@ async def test_vertexai_unroutable_ids_fail_closed(
         'virtual-try-on-001',
         'imagegeneration@006',
         'imagetext@001',
+        'imagen-3.0-generate-002',
+        'imagen-4.0-generate-001',
         'deep-research-pro-preview',
         'gemini-embedding-001',
         'models/deep-research-pro-preview',
@@ -557,7 +566,7 @@ async def test_googleai_resolve_veo_as_model_returns_none(mock_list_models: Magi
     mock_list_models.return_value = GenaiModels()
 
     plugin = GoogleAI(api_key='test-key')
-    action = await plugin.resolve(ActionKind.MODEL, 'googleai/veo-3.0-generate-001')
+    action = await plugin.resolve(ActionKind.MODEL, 'googleai/veo-3.1-generate-preview')
 
     assert action is None
 
@@ -570,7 +579,7 @@ async def test_vertexai_resolve_veo_as_model_returns_none(mock_list_models: Magi
     mock_list_models.return_value = GenaiModels()
 
     plugin = VertexAI(project='test-project')
-    action = await plugin.resolve(ActionKind.MODEL, 'vertexai/veo-3.0-generate-001')
+    action = await plugin.resolve(ActionKind.MODEL, 'vertexai/veo-3.1-generate-001')
 
     assert action is None
 
@@ -583,11 +592,11 @@ async def test_resolve_model_finds_veo_as_background(mock_list_models: MagicMock
     mock_list_models.return_value = GenaiModels()
 
     ai = Genkit(plugins=[GoogleAI(api_key='test-key')])
-    action = await ai.registry.resolve_model('googleai/veo-3.0-generate-001')
+    action = await ai.registry.resolve_model('googleai/veo-3.1-generate-preview')
 
     assert action is not None
     assert action.kind == ActionKind.BACKGROUND_MODEL
-    assert action.name == 'googleai/veo-3.0-generate-001'
+    assert action.name == 'googleai/veo-3.1-generate-preview'
 
 
 @patch('genkit_google_genai.models.veo.genai.Client')
@@ -613,7 +622,7 @@ async def test_generate_and_check_operation_apply_veo_context_and_config(
     context = {'secrets': {'api_key': 'tenant-key'}}
 
     operation = await ai.generate_operation(
-        model='googleai/veo-3.0-generate-001',
+        model='googleai/veo-3.1-generate-preview',
         prompt='a cat walking',
         config={'aspectRatio': '16:9', 'baseUrl': 'https://request.example', 'apiVersion': 'v1'},
         context=context,
@@ -650,18 +659,18 @@ async def test_vertexai_resolve_veo_background_model(mock_list_models: MagicMock
     mock_list_models.return_value = GenaiModels()
 
     plugin = VertexAI(project='test-project')
-    start = await plugin.resolve(ActionKind.BACKGROUND_MODEL, 'vertexai/veo-3.0-generate-001')
-    check = await plugin.resolve(ActionKind.CHECK_OPERATION, 'vertexai/veo-3.0-generate-001/check')
+    start = await plugin.resolve(ActionKind.BACKGROUND_MODEL, 'vertexai/veo-3.1-generate-001')
+    check = await plugin.resolve(ActionKind.CHECK_OPERATION, 'vertexai/veo-3.1-generate-001/check')
 
     assert start is not None
     assert start.kind == ActionKind.BACKGROUND_MODEL
-    assert start.name == 'vertexai/veo-3.0-generate-001'
+    assert start.name == 'vertexai/veo-3.1-generate-001'
     model_meta = cast('dict[str, object]', start.metadata['model'])
     supports = cast('dict[str, object]', model_meta['supports'])
     assert supports['longRunning'] is True
     assert check is not None
     assert check.kind == ActionKind.CHECK_OPERATION
-    assert check.name == 'vertexai/veo-3.0-generate-001/check'
+    assert check.name == 'vertexai/veo-3.1-generate-001/check'
 
 
 @patch('genkit_google_genai.google.create_vertex_evaluators')
@@ -673,7 +682,7 @@ async def test_vertexai_init_registers_veo_as_background(
 ) -> None:
     """Vertex init registers Veo start/check, never a blocking MODEL action."""
     models = GenaiModels()
-    models.veo = ['veo-3.0-generate-001']
+    models.veo = ['veo-3.1-generate-001']
     mock_list_models.return_value = models
     mock_evaluators.return_value = []
 
@@ -691,7 +700,7 @@ async def test_vertexai_init_registers_veo_as_background(
 async def test_list_actions_advertises_veo_as_background(mock_list_models: MagicMock, mock_client: MagicMock) -> None:
     """Both plugins list Veo with the kind that resolve() actually serves."""
     models = GenaiModels()
-    models.veo = ['veo-3.0-generate-001']
+    models.veo = ['veo-3.1-generate-001']
     mock_list_models.return_value = models
 
     googleai_actions = await GoogleAI(api_key='test-key').list_actions()
@@ -700,12 +709,12 @@ async def test_list_actions_advertises_veo_as_background(mock_list_models: Magic
     for actions, plugin_name in ((googleai_actions, 'googleai'), (vertexai_actions, 'vertexai')):
         veo_entries = [a for a in actions if 'veo' in a.name]
         assert len(veo_entries) == 1
-        assert veo_entries[0].name == f'{plugin_name}/veo-3.0-generate-001'
+        assert veo_entries[0].name == f'{plugin_name}/veo-3.1-generate-001'
         assert veo_entries[0].action_type == ActionKind.BACKGROUND_MODEL
 
 
-def test_list_genai_models_vertex_skips_substring_veo_and_retired_image() -> None:
-    """Discovery buckets on the ``veo-`` prefix, not a ``veo`` substring."""
+def test_list_genai_models_vertex_skips_substring_veo_and_unsupported_image() -> None:
+    """Discovery buckets on the ``veo-`` prefix and drops unsupported image ids."""
 
     def _model(name: str) -> MagicMock:
         item = MagicMock()
@@ -717,17 +726,20 @@ def _model(name: str) -> MagicMock:
     client = MagicMock()
     client.models.list.return_value = [
         _model('publishers/google/models/gemini-2.5-flash'),
-        _model('publishers/google/models/veo-3.0-generate-001'),
+        _model('publishers/google/models/veo-3.1-generate-001'),
         _model('publishers/google/models/braveo-lab'),
         _model('publishers/google/models/imagegeneration@006'),
         _model('publishers/google/models/virtual-try-on-001'),
         _model('publishers/google/models/imagetext@001'),
+        _model('publishers/google/models/imagen-3.0-generate-002'),
+        _model('publishers/google/models/imagen-4.0-generate-001'),
     ]
     catalog = _list_genai_models(client, is_vertex=True)
-    assert catalog.veo == ['veo-3.0-generate-001']
-    assert catalog.imagen == []
-    assert 'imagetext@001' not in catalog.gemini
-    assert 'braveo-lab' not in catalog.gemini
+    assert vars(catalog) == {
+        'gemini': ['gemini-2.5-flash'],
+        'embedders': [],
+        'veo': ['veo-3.1-generate-001'],
+    }
 
 
 @patch('genkit_google_genai.google.genai.client.Client')
@@ -737,19 +749,19 @@ async def test_veo_start_stamps_background_action_key(mock_list_models: MagicMoc
     """Start and check stamp ``/background-model/{name}`` so a later check can resolve."""
     mock_list_models.return_value = GenaiModels()
     plugin = VertexAI(project='test-project')
-    start = await plugin.resolve(ActionKind.BACKGROUND_MODEL, 'vertexai/veo-3.0-generate-001')
-    check = await plugin.resolve(ActionKind.CHECK_OPERATION, 'vertexai/veo-3.0-generate-001/check')
+    start = await plugin.resolve(ActionKind.BACKGROUND_MODEL, 'vertexai/veo-3.1-generate-001')
+    check = await plugin.resolve(ActionKind.CHECK_OPERATION, 'vertexai/veo-3.1-generate-001/check')
     assert start is not None
     assert check is not None
 
     request = ModelRequest(messages=[Message(role=Role.USER, content=[Part(root=TextPart(text='a clip'))])])
     with patch.object(VeoModel, 'start', new=AsyncMock(return_value=Operation(id='ops/1'))):
         started = await start.run(request)
-    assert started.response.action == '/background-model/vertexai/veo-3.0-generate-001'
+    assert started.response.action == '/background-model/vertexai/veo-3.1-generate-001'
 
     with patch.object(VeoModel, 'check', new=AsyncMock(return_value=Operation(id='ops/1'))):
         checked = await check.run(Operation(id='ops/1'))
-    assert checked.response.action == '/background-model/vertexai/veo-3.0-generate-001'
+    assert checked.response.action == '/background-model/vertexai/veo-3.1-generate-001'
 
 
 @patch('genkit_google_genai.google.genai.client.Client')
diff --git a/py/packages/genkit-google-genai/tests/veo_test.py b/py/packages/genkit-google-genai/tests/veo_test.py
index 0d44063b45..5fc9cf2e61 100644
--- a/py/packages/genkit-google-genai/tests/veo_test.py
+++ b/py/packages/genkit-google-genai/tests/veo_test.py
@@ -24,7 +24,6 @@
 from genkit_google_genai.models.veo import (
     VeoConfig,
     VeoModel,
-    VeoVersion,
     _from_veo_operation,
     is_veo_model,
 )
@@ -77,11 +76,11 @@ class TestIsVeoModel:
 
     def test_veo_model_name(self) -> None:
         """Veo model names are recognized."""
-        assert is_veo_model('veo-2.0-generate-001') is True
+        assert is_veo_model('veo-3.1-generate-001') is True
 
     def test_veo_uppercase(self) -> None:
         """Case-insensitive matching works."""
-        assert is_veo_model('VEO-2.0-generate-001') is True
+        assert is_veo_model('VEO-3.1-generate-001') is True
 
     def test_non_veo_model(self) -> None:
         """Non-Veo model names are rejected."""
@@ -89,30 +88,13 @@ def test_non_veo_model(self) -> None:
 
     def test_namespaced_veo_model(self) -> None:
         """Plugin prefixes are stripped before the ``veo-`` check."""
-        assert is_veo_model('googleai/veo-3.0-generate-001') is True
+        assert is_veo_model('googleai/veo-3.1-generate-preview') is True
 
     def test_substring_veo_is_rejected(self) -> None:
         """A bare ``veo`` substring is not enough; the id has to start with ``veo-``."""
         assert is_veo_model('devotional-hymn') is False
 
 
-class TestVeoVersion:
-    """Tests for VeoVersion enum convenience constants."""
-
-    @pytest.mark.parametrize(
-        'version',
-        [
-            VeoVersion.VEO_3_1_PREVIEW,
-            VeoVersion.VEO_3_1_FAST_PREVIEW,
-            VeoVersion.VEO_3_0,
-            VeoVersion.VEO_3_0_FAST,
-        ],
-    )
-    def test_new_googleai_models_are_recognized(self, version: VeoVersion) -> None:
-        """New Veo 3.0/3.1 model constants map to valid Veo names."""
-        assert is_veo_model(version.value) is True
-
-
 class TestFromVeoOperation:
     """``_from_veo_operation`` reads the SDK operation and resolves playable media."""
 
@@ -341,7 +323,7 @@ async def test_start_passes_generate_videos_config_and_returns_ticket(self) -> N
         """A typed Veo config dumps aspectRatio / durationSeconds onto generate_videos."""
         client = MagicMock()
         client.aio.models.generate_videos = AsyncMock(return_value=_sdk_op(name='operations/1', done=False))
-        veo = VeoModel('veo-3.0-generate-001', client)
+        veo = VeoModel('veo-3.1-generate-001', client)
         request = _text_request(
             config=VeoConfig.model_validate({'aspectRatio': '16:9', 'durationSeconds': 5, 'fooBar': 1}),
         )
@@ -361,7 +343,7 @@ async def test_start_no_config_sends_none(self) -> None:
         """No config is a valid start; generate_videos gets no knobs."""
         client = MagicMock()
         client.aio.models.generate_videos = AsyncMock(return_value=_sdk_op(name='operations/1', done=False))
-        veo = VeoModel('veo-3.0-generate-001', client)
+        veo = VeoModel('veo-3.1-generate-001', client)
 
         await veo.start(_text_request(), ActionRunContext())
 
@@ -384,7 +366,7 @@ async def test_check_polls_operation_by_sdk_name_and_returns_updated_operation(s
                 ),
             ),
         )
-        model = VeoModel('veo-3.0-generate-001', client)
+        model = VeoModel('veo-3.1-generate-001', client)
         updated = await model.check(Operation(id='operations/123'), ActionRunContext())
 
         assert updated.done is True
@@ -395,7 +377,7 @@ async def test_check_wraps_api_error_into_genkit_error(self) -> None:
         """A 503 on the poll must stay retryable UNAVAILABLE, not collapse to INTERNAL."""
         client = MagicMock()
         client.aio.operations.get = AsyncMock(side_effect=APIError(503, {'error': {'message': 'overloaded'}}))
-        model = VeoModel('veo-3.0-generate-001', client)
+        model = VeoModel('veo-3.1-generate-001', client)
 
         with pytest.raises(GenkitError) as raised:
             await model.check(Operation(id='operations/abc'), ActionRunContext())
@@ -403,7 +385,7 @@ async def test_check_wraps_api_error_into_genkit_error(self) -> None:
 
     def test_invalid_sdk_field_raises_invalid_argument(self) -> None:
         """SDK type errors become a named INVALID_ARGUMENT."""
-        veo = VeoModel('veo-3.0-generate-001', MagicMock())
+        veo = VeoModel('veo-3.1-generate-001', MagicMock())
         request = _text_request(config=VeoConfig.model_construct(duration_seconds='nope'))
 
         with pytest.raises(GenkitError) as exc_info:
@@ -447,7 +429,7 @@ async def test_empty_context_uses_plugin_client(self) -> None:
         plugin = MagicMock()
         plugin.aio.models.generate_videos = AsyncMock(return_value=_pending_sdk_op())
         plugin.aio.operations.get = AsyncMock(return_value=_pending_sdk_op())
-        veo = VeoModel('veo-3.0-generate-001', plugin)
+        veo = VeoModel('veo-3.1-generate-001', plugin)
 
         with patch('genkit_google_genai.models.veo.genai.Client') as ctor:
             started = await veo.start(_text_request(), ActionRunContext())
@@ -463,7 +445,7 @@ async def test_secrets_api_key_builds_request_client(self) -> None:
         plugin.aio.models.generate_videos = AsyncMock(return_value=_pending_sdk_op())
         override = MagicMock()
         override.aio.models.generate_videos = AsyncMock(return_value=_pending_sdk_op(name='operations/tenant'))
-        veo = VeoModel('veo-3.0-generate-001', plugin)
+        veo = VeoModel('veo-3.1-generate-001', plugin)
 
         with patch('genkit_google_genai.models.veo.genai.Client', return_value=override) as ctor:
             op = await veo.start(
@@ -486,7 +468,7 @@ async def test_start_config_routes_client_and_stays_out_of_model_parameters(self
         plugin.aio.models.generate_videos = AsyncMock(return_value=_pending_sdk_op())
         override = MagicMock()
         override.aio.models.generate_videos = AsyncMock(return_value=_pending_sdk_op())
-        veo = VeoModel('veo-3.0-generate-001', plugin, client_kwargs={'api_key': 'plugin-key'})
+        veo = VeoModel('veo-3.1-generate-001', plugin, client_kwargs={'api_key': 'plugin-key'})
         request = _text_request(
             config=VeoConfig.model_validate({
                 'aspectRatio': '16:9',
@@ -517,7 +499,7 @@ async def test_start_config_routes_vertex_location(self) -> None:
         override = MagicMock()
         override.aio.models.generate_videos = AsyncMock(return_value=_pending_sdk_op())
         veo = VeoModel(
-            'veo-3.0-generate-001',
+            'veo-3.1-generate-001',
             plugin,
             client_kwargs={'vertexai': True, 'project': 'p', 'location': 'us-central1'},
         )
@@ -536,7 +518,7 @@ async def test_check_without_tenant_context_returns_to_plugin_client(self) -> No
         plugin.aio.operations.get = AsyncMock(return_value=_pending_sdk_op())
         override = MagicMock()
         override.aio.models.generate_videos = AsyncMock(return_value=_pending_sdk_op())
-        veo = VeoModel('veo-3.0-generate-001', plugin)
+        veo = VeoModel('veo-3.1-generate-001', plugin)
 
         with patch('genkit_google_genai.models.veo.genai.Client', return_value=override):
             ticket = await veo.start(
@@ -553,7 +535,7 @@ async def test_secrets_apikey_alias(self) -> None:
         plugin = MagicMock()
         override = MagicMock()
         override.aio.models.generate_videos = AsyncMock(return_value=_pending_sdk_op())
-        veo = VeoModel('veo-3.0-generate-001', plugin)
+        veo = VeoModel('veo-3.1-generate-001', plugin)
 
         with patch('genkit_google_genai.models.veo.genai.Client', return_value=override) as ctor:
             await veo.start(
@@ -570,7 +552,7 @@ async def test_check_uses_secrets_and_does_not_write_key_on_op(self) -> None:
         plugin.aio.operations.get = AsyncMock(return_value=_pending_sdk_op())
         override = MagicMock()
         override.aio.operations.get = AsyncMock(return_value=_pending_sdk_op())
-        veo = VeoModel('veo-3.0-generate-001', plugin)
+        veo = VeoModel('veo-3.1-generate-001', plugin)
         ticket = Operation(id='operations/1', done=False)
 
         with patch('genkit_google_genai.models.veo.genai.Client', return_value=override) as ctor:
@@ -591,7 +573,7 @@ async def test_secrets_and_base_url_together(self) -> None:
         override = MagicMock()
         override.aio.models.generate_videos = AsyncMock(return_value=_pending_sdk_op())
         override.aio.operations.get = AsyncMock(return_value=_pending_sdk_op())
-        veo = VeoModel('veo-3.0-generate-001', plugin)
+        veo = VeoModel('veo-3.1-generate-001', plugin)
         ctx = ActionRunContext(
             context={
                 'secrets': {'api_key': 'sk-tenant'},
@@ -617,7 +599,7 @@ async def test_vertex_secrets_drop_project_and_location(self) -> None:
         override = MagicMock()
         override.aio.models.generate_videos = AsyncMock(return_value=_pending_sdk_op())
         veo = VeoModel(
-            'veo-3.0-generate-001',
+            'veo-3.1-generate-001',
             plugin,
             client_kwargs={
                 'vertexai': True,
@@ -645,7 +627,7 @@ async def test_googleai_location_is_ignored(self) -> None:
         plugin = MagicMock()
         plugin.vertexai = False
         plugin.aio.models.generate_videos = AsyncMock(return_value=_pending_sdk_op())
-        veo = VeoModel('veo-3.0-generate-001', plugin, client_kwargs={'api_key': 'plugin-key'})
+        veo = VeoModel('veo-3.1-generate-001', plugin, client_kwargs={'api_key': 'plugin-key'})
 
         with patch('genkit_google_genai.models.veo.genai.Client') as ctor:
             await veo.start(
@@ -663,7 +645,7 @@ async def test_vertex_location_rewrites_base_url(self) -> None:
         override = MagicMock()
         override.aio.operations.get = AsyncMock(return_value=_pending_sdk_op())
         veo = VeoModel(
-            'veo-3.0-generate-001',
+            'veo-3.1-generate-001',
             plugin,
             client_kwargs={
                 'vertexai': True,
@@ -690,7 +672,7 @@ async def test_vertex_regional_location_clears_rep_url(self) -> None:
         override = MagicMock()
         override.aio.operations.get = AsyncMock(return_value=_pending_sdk_op())
         veo = VeoModel(
-            'veo-3.0-generate-001',
+            'veo-3.1-generate-001',
             plugin,
             client_kwargs={
                 'vertexai': True,
@@ -713,7 +695,7 @@ async def test_vertex_regional_location_clears_rep_url(self) -> None:
     @pytest.mark.asyncio
     @pytest.mark.parametrize('bag', ({'api_key': 'sk-wrong'}, {'apiKey': 'sk-wrong'}))
     async def test_config_api_key_is_invalid_argument(self, bag: dict[str, str]) -> None:
-        veo = VeoModel('veo-3.0-generate-001', MagicMock())
+        veo = VeoModel('veo-3.1-generate-001', MagicMock())
 
         with pytest.raises(GenkitError) as raised:
             await veo.start(
@@ -726,7 +708,7 @@ async def test_config_api_key_is_invalid_argument(self, bag: dict[str, str]) ->
 
     @pytest.mark.asyncio
     async def test_secrets_must_be_a_dict(self) -> None:
-        veo = VeoModel('veo-3.0-generate-001', MagicMock())
+        veo = VeoModel('veo-3.1-generate-001', MagicMock())
 
         with pytest.raises(GenkitError) as raised:
             await veo.start(
@@ -738,7 +720,7 @@ async def test_secrets_must_be_a_dict(self) -> None:
 
     @pytest.mark.asyncio
     async def test_secret_api_key_must_be_a_string(self) -> None:
-        veo = VeoModel('veo-3.0-generate-001', MagicMock())
+        veo = VeoModel('veo-3.1-generate-001', MagicMock())
 
         with pytest.raises(GenkitError) as raised:
             await veo.start(
@@ -755,7 +737,7 @@ async def test_api_version_overlay(self) -> None:
         plugin.vertexai = False
         override = MagicMock()
         override.aio.models.generate_videos = AsyncMock(return_value=_pending_sdk_op())
-        veo = VeoModel('veo-3.0-generate-001', plugin, client_kwargs={'api_key': 'plugin-key'})
+        veo = VeoModel('veo-3.1-generate-001', plugin, client_kwargs={'api_key': 'plugin-key'})
 
         with patch('genkit_google_genai.models.veo.genai.Client', return_value=override) as ctor:
             await veo.start(
@@ -769,7 +751,7 @@ async def test_api_version_overlay(self) -> None:
 
     @pytest.mark.asyncio
     async def test_empty_secrets_pocket_is_invalid_argument(self) -> None:
-        veo = VeoModel('veo-3.0-generate-001', MagicMock())
+        veo = VeoModel('veo-3.1-generate-001', MagicMock())
         for pocket in ({}, {'api_key': None}, {'api_key': ''}):
             with pytest.raises(GenkitError) as raised:
                 await veo.start(
@@ -780,7 +762,7 @@ async def test_empty_secrets_pocket_is_invalid_argument(self) -> None:
 
     @pytest.mark.asyncio
     async def test_top_level_api_key_is_invalid_argument(self) -> None:
-        veo = VeoModel('veo-3.0-generate-001', MagicMock())
+        veo = VeoModel('veo-3.1-generate-001', MagicMock())
         for bag in ({'api_key': 'sk-wrong'}, {'apiKey': 'sk-wrong'}):
             with pytest.raises(GenkitError) as raised:
                 await veo.start(_text_request(), ActionRunContext(context=bag))
@@ -789,7 +771,7 @@ async def test_top_level_api_key_is_invalid_argument(self) -> None:
 
     @pytest.mark.asyncio
     async def test_request_config_api_key_is_invalid_argument(self) -> None:
-        veo = VeoModel('veo-3.0-generate-001', MagicMock())
+        veo = VeoModel('veo-3.1-generate-001', MagicMock())
         cfg = VeoConfig.model_validate({'api_key': 'sk-gemini-habit'})
 
         with pytest.raises(GenkitError) as raised:
@@ -802,7 +784,7 @@ async def test_request_config_api_key_is_invalid_argument(self) -> None:
     async def test_client_ctor_failure_is_invalid_argument(self) -> None:
         plugin = MagicMock()
         plugin.vertexai = False
-        veo = VeoModel('veo-3.0-generate-001', plugin, client_kwargs={'api_key': 'plugin-key'})
+        veo = VeoModel('veo-3.1-generate-001', plugin, client_kwargs={'api_key': 'plugin-key'})
 
         with (
             patch(
@@ -826,7 +808,7 @@ async def test_vertex_tenant_key_clears_plugin_base_url(self) -> None:
         override = MagicMock()
         override.aio.models.generate_videos = AsyncMock(return_value=_pending_sdk_op())
         veo = VeoModel(
-            'veo-3.0-generate-001',
+            'veo-3.1-generate-001',
             plugin,
             client_kwargs={
                 'vertexai': True,
diff --git a/py/packages/genkit/tests/genkit/core/registry_test.py b/py/packages/genkit/tests/genkit/core/registry_test.py
index 7118122f63..9a83b9324c 100644
--- a/py/packages/genkit/tests/genkit/core/registry_test.py
+++ b/py/packages/genkit/tests/genkit/core/registry_test.py
@@ -538,8 +538,8 @@ async def resolve(self, action_type: ActionKind, name: str) -> Action | None:
             return Action(name=name, kind=ActionKind.BACKGROUND_MODEL, fn=_bg_start)
 
     ai = Genkit(plugins=[VeoPlugin()])
-    got = await ai.registry.resolve_model('plug/veo-2.0-generate-001')
+    got = await ai.registry.resolve_model('plug/veo-3.1-generate-preview')
 
     assert got is not None
     assert got.kind == ActionKind.BACKGROUND_MODEL
-    assert got.name == 'plug/veo-2.0-generate-001'
+    assert got.name == 'plug/veo-3.1-generate-preview'