Add images optional in nano banana node and remove gcs dependency for r2v

modules/python/src/custom_nodes/google_genmedia/gemini_flash_image_api.py

@@ -70,7 +70,7 @@ def generate_image(
         sexually_explicit_threshold: str,
         dangerous_content_threshold: str,
         system_instruction: str,
-        image1: torch.Tensor,
+        image1: Optional[torch.Tensor] = None,
         image2: Optional[torch.Tensor] = None,
         image3: Optional[torch.Tensor] = None,
     ) -> List[Image.Image]:
@@ -89,7 +89,7 @@ def generate_image(
               content.
             dangerous_content_threshold: Safety threshold for dangerous content.
             system_instruction: System-level instructions for the model.
-            image1: The primary input image tensor for image-to-image tasks.
+            image1: An optional primary input image tensor for image-to-image tasks.
             image2: An optional second input image tensor. Defaults to None.
             image3: An optional third input image tensor. Defaults to None.
 

modules/python/src/custom_nodes/google_genmedia/gemini_flash_image_node.py

@@ -60,7 +60,6 @@ def INPUT_TYPES(cls) -> Dict[str, Dict[str, Any]]:
                         "default": "A vivid landscape painting of a futuristic city",
                     },
                 ),
-                "image1": ("IMAGE",),
                 "aspect_ratio": (
                     [
                         "1:1",
@@ -87,6 +86,7 @@ def INPUT_TYPES(cls) -> Dict[str, Dict[str, Any]]:
                 "top_k": ("INT", {"default": 32, "min": 1, "max": 64}),
             },
             "optional": {
+                "image1": ("IMAGE",),
                 "image2": ("IMAGE",),
                 "image3": ("IMAGE",),
                 # Safety Settings
@@ -145,12 +145,12 @@ def generate_and_return_image(
         temperature: float,
         top_p: float,
         top_k: int,
-        image1: torch.Tensor,
         hate_speech_threshold: str,
         harassment_threshold: str,
         sexually_explicit_threshold: str,
         dangerous_content_threshold: str,
         system_instruction: str,
+        image1: Optional[torch.Tensor] = None,
         image2: Optional[torch.Tensor] = None,
         image3: Optional[torch.Tensor] = None,
         gcp_project_id: Optional[str] = None,
@@ -175,7 +175,7 @@ def generate_and_return_image(
               content.
             dangerous_content_threshold: Safety threshold for dangerous content.
             system_instruction: System-level instructions for the model.
-            image1: The primary input image tensor for image editing tasks.
+            image1: An optional primary input image tensor for image editing tasks.
             image2: An optional second input image tensor. Defaults to None.
             image3: An optional third input image tensor. Defaults to None.
             gcp_project_id: The GCP project ID.

modules/python/src/custom_nodes/google_genmedia/utils.py

@@ -483,13 +483,15 @@ def generate_video_from_image(
 
 
 @api_error_retry
-def generate_video_from_gcs_references(
+def generate_video_from_references(
     client: genai.Client,
     model: str,
     prompt: str,
-    gcs_uris: List[str],
     image_format: str,
     aspect_ratio: str,
+    image1: torch.Tensor,
+    image2: Optional[torch.Tensor],
+    image3: Optional[torch.Tensor],
     output_resolution: Optional[str],
     compression_quality: Optional[str],
     person_generation: str,
@@ -530,16 +532,18 @@ def generate_video_from_gcs_references(
         "generate_audio": generate_audio if generate_audio is not None else False,
     }
 
-    if output_gcs_uri:
-        temp_config["output_gcs_uri"] = output_gcs_uri
-
     reference_images = []
-    for uri in gcs_uris:
-        image_part = Image(gcs_uri=uri, mime_type=mime_type)
-        reference_image = types.VideoGenerationReferenceImage(
-            image=image_part, reference_type="asset"
-        )
-        reference_images.append(reference_image)
+
+    for image_tensor in [image1, image2, image3]:
+        if image_tensor is not None:
+            image_part = Image(
+                imageBytes=tensor_to_pil_to_base64(image_tensor, image_format),
+                mime_type=mime_type,
+            )
+            reference_image = types.VideoGenerationReferenceImage(
+                image=image_part, reference_type="asset"
+            )
+            reference_images.append(reference_image)
 
     temp_config["reference_images"] = reference_images
 
@@ -966,64 +970,6 @@ def process_video_response(operation: Any) -> List[str]:
     return video_paths
 
 
-def upload_images_to_gcs(
-    images: List[Optional[torch.Tensor]], bucket_name: str, image_format: str
-) -> List[str]:
-    """
-    Uploads a list of image tensors to a GCS bucket.
-
-    Args:
-        images: A list of torch.Tensor images, which can contain None.
-        bucket_name: The name of the GCS bucket.
-        image_format: The format of the images (e.g., "PNG", "JPEG").
-
-    Returns:
-        A list of GCS URIs for the uploaded images.
-    """
-    prefix = "gs://"
-    if bucket_name.startswith(prefix):
-        bucket_name = bucket_name[len(prefix) :]
-
-    gcs_uris = []
-    storage_client = storage.Client(
-        client_info=ClientInfo(user_agent=STORAGE_USER_AGENT)
-    )
-    bucket = storage_client.bucket(bucket_name)
-
-    if not bucket.exists():
-        raise APIInputError(
-            f"GCS bucket '{bucket_name}' does not exist or is inaccessible."
-        )
-
-    for i, image_tensor in enumerate(images):
-        if image_tensor is not None:
-            try:
-                timestamp = int(time.time())
-                unique_id = random.randint(1000, 9999)
-                object_name = f"temporary-reference-images/ref_{timestamp}_{i+1}_{unique_id}.{image_format.lower()}"
-                blob = bucket.blob(object_name)
-
-                # VEO expects single images, not batches. Take the first from any potential batch.
-                single_image_tensor = image_tensor[0].unsqueeze(0)
-                image_bytes = tensor_to_pil_to_bytes(
-                    single_image_tensor, format=image_format.upper()
-                )
-
-                blob.upload_from_string(
-                    image_bytes, content_type=f"image/{image_format.lower()}"
-                )
-
-                gcs_uri = f"gs://{bucket_name}/{object_name}"
-                gcs_uris.append(gcs_uri)
-                logger.info(f"Successfully uploaded reference image {i+1} to {gcs_uri}")
-            except Exception as e:
-                raise APIExecutionError(
-                    f"Failed to upload image {i+1} to GCS: {e}"
-                ) from e
-
-    return gcs_uris
-
-
 def validate_gcs_uri_and_image(
     gcs_uri: str, check_object: bool = True
 ) -> Tuple[bool, str]:

modules/python/src/custom_nodes/google_genmedia/veo3_api.py

@@ -231,7 +231,6 @@ def generate_video_from_references(
         self,
         model: str,
         prompt: str,
-        bucket_name: str,
         image1: torch.Tensor,
         image_format: str,
         aspect_ratio: str,
@@ -249,12 +248,11 @@ def generate_video_from_references(
         seed: Optional[int],
     ) -> List[str]:
         """
-        Uploads reference images to GCS and then generates a video.
+        Generates a video from the references.
 
         Args:
             model: Veo3 model.
             prompt: The text prompt for video generation.
-            bucket_name: The GCS bucket to upload reference images to.
             image1: The first reference image as a torch.Tensor.
             image_format: The format of the input images.
             aspect_ratio: The desired aspect ratio of the video.
@@ -278,27 +276,15 @@ def generate_video_from_references(
             raise APIInputError(
                 "Image1 is required. At least reference image must be provided."
             )
-        if not bucket_name:
-            raise APIInputError(
-                "bucket_name is required for uploading reference images."
-            )
-
-        gcs_uris = utils.upload_images_to_gcs(
-            images=[image1, image2, image3],
-            bucket_name=bucket_name,
-            image_format=image_format,
-        )
-
-        if not gcs_uris:
-            raise APIExecutionError("Failed to upload any reference images to GCS.")
-
         model_enum = Veo3Model[model]
 
-        return utils.generate_video_from_gcs_references(
+        return utils.generate_video_from_references(
             client=self.client,
             model=model_enum,
             prompt=prompt,
-            gcs_uris=gcs_uris,
+            image1=image1,
+            image2=image2,
+            image3=image3,
             image_format=image_format,
             aspect_ratio=aspect_ratio,
             output_resolution=output_resolution,

modules/python/src/custom_nodes/google_genmedia/veo3_nodes.py

@@ -522,13 +522,6 @@ def INPUT_TYPES(cls) -> Dict[str, Dict[str, Any]]:
                     [model.name for model in Veo3Model],
                     {"default": Veo3Model.VEO_3_1_PREVIEW.name},
                 ),
-                "bucket_name": (
-                    "STRING",
-                    {
-                        "default": "",
-                        "tooltip": "GCS bucket name to temporarily store reference images.",
-                    },
-                ),
                 "image1": ("IMAGE",),
                 "image_format": (
                     ["PNG", "JPEG"],
@@ -591,7 +584,6 @@ def INPUT_TYPES(cls) -> Dict[str, Dict[str, Any]]:
     def generate_from_references(
         self,
         model: str,
-        bucket_name: str,
         image1: torch.Tensor,
         image_format: str,
         prompt: str,
@@ -622,7 +614,6 @@ def generate_from_references(
             video_paths = api.generate_video_from_references(
                 model=model,
                 prompt=prompt,
-                bucket_name=bucket_name,
                 image1=image1,
                 image2=image2,
                 image3=image3,

platforms/gke/base/use-cases/inference-ref-arch/terraform/comfyui/gcs.tf

@@ -199,10 +199,9 @@ resource "google_storage_bucket_object" "workflow_veo3_itv" {
 }
 
 resource "google_storage_bucket_object" "workflow_veo3_r2v" {
-  bucket     = google_storage_bucket.comfyui_workflow.name
-  name       = "veo3-reference-to-video.json"
-  source     = "src/comfyui-workflows/veo3-reference-to-video.json"
-  depends_on = [local_file.workflow_veo2_ttv]
+  bucket = google_storage_bucket.comfyui_workflow.name
+  name   = "veo3-reference-to-video.json"
+  source = "src/comfyui-workflows/veo3-reference-to-video.json"
 }
 
 resource "google_storage_bucket_object" "workflow_veo3_ttv" {
@@ -270,13 +269,3 @@ resource "local_file" "workflow_veo3_ttv" {
   )
   filename = "${path.module}/src/comfyui-workflows/veo3-text-to-video.json"
 }
-
-resource "local_file" "workflow_veo3_r2v" {
-  content = templatefile(
-    "${path.module}/src/comfyui-workflows/veo3-reference-to-video.tftpl.json",
-    {
-      output_bucket_uri = google_storage_bucket.comfyui_output.url
-    }
-  )
-  filename = "${path.module}/src/comfyui-workflows/veo3-reference-to-video.json"
-}