Added support for Veo 3 (9:16 aspect ratio, 4s/6s durations) and Veo 2 R2V.

backend/app.py

@@ -35,4 +35,4 @@ def create_app():
 if __name__ == '__main__':
     print(f"Starting Flask app with SQLite persistence.")
     print(f"Database will be stored at: {app.config['SQLALCHEMY_DATABASE_URI']}")
-    app.run(debug=False, host='0.0.0.0', port=5001)
+    app.run(debug=False, host='0.0.0.0', port=5000)

backend/google_veo.py

@@ -42,6 +42,7 @@ def _compose_videogen_request(
         camera_control: str = "",
         generate_audio: bool = False,
         resolution: Optional[str] = None,
+        reference_images: Optional[list] = None,
     ):
         if self.model_name.startswith("veo-3.0"):
             if "durationSeconds" in parameters and isinstance(parameters["durationSeconds"], (int, float)) and parameters["durationSeconds"] > 90:
@@ -64,6 +65,11 @@ def _compose_videogen_request(
             instance["video"] = {"gcsUri": video_uri, "mimeType": "video/mp4"}
         if last_frame_uri:
             instance["lastFrame"] = {"gcsUri": last_frame_uri, "mimeType": last_frame_mime_type}
+
+        # Add referenceImages support for veo-2.0-generate-exp
+        if reference_images and self.model_name == "veo-2.0-generate-exp":
+            instance["referenceImages"] = reference_images
+
         # Only add cameraControl if the model supports it, it's provided, AND it's not a video extension task
         if self.model_name != "veo-3.0-generate-001" and camera_control and not video_uri:
             instance["cameraControl"] = camera_control
@@ -98,6 +104,7 @@ def generate_video(
         camera_control: str = "",
         generate_audio: bool = False,
         resolution: Optional[str] = None,
+        reference_images: Optional[list] = None,
     ):
         req = self._compose_videogen_request(
             prompt=prompt,
@@ -110,6 +117,7 @@ def generate_video(
             camera_control=camera_control,
             generate_audio=generate_audio,
             resolution=resolution,
+            reference_images=reference_images,
         )
         print(f"Sending video generation request: {req}")
         resp = self._send_request_to_google_api(self.prediction_endpoint, req)

backend/migrate_db.py

@@ -156,6 +156,9 @@ def setup_database():
         # The model uses db.Boolean, so SQLAlchemy handles the abstraction.
         # When adding manually, 'BOOLEAN' should be acceptable for both via SQLAlchemy's engine.
         migrate_schema_add_column(engine, 'video_generation_task', 'generate_audio', 'BOOLEAN')
+
+        # Add reference_images_data column for storing reference images JSON data
+        migrate_schema_add_column(engine, 'video_generation_task', 'reference_images_data', 'TEXT')
 
         # Backfill data
         migrate_data_backfill_user_column(engine)

backend/models.py

@@ -23,6 +23,7 @@ class VideoGenerationTask(db.Model):
     last_frame_filename = db.Column(db.String(255), nullable=True) # Filename of the uploaded last frame image
     last_frame_gcs_uri = db.Column(db.String(1024), nullable=True) # GCS URI of the uploaded last frame image
     video_uri = db.Column(db.String(1024), nullable=True) # User-added: new video_uri attribute
+    reference_images_data = db.Column(db.Text, nullable=True) # JSON string for reference images data
     error_message = db.Column(db.String(1024), nullable=True)
     user = db.Column(db.String(255), nullable=True) # New field for user email
     generate_audio = db.Column(db.Boolean, default=False)

backend/routes/video.py

@@ -25,7 +25,7 @@ def generate_video_route():
     model = request.form.get('model', DEFAULT_VIDEO_MODEL)
     aspect_ratio = request.form.get('ratio', '16:9')
     camera_control = request.form.get('camera_control', 'FIXED') # Get camera_control
-    duration_seconds = int(request.form.get('duration', 5))
+    duration_seconds = int(request.form.get('durationSeconds', 8))
     resolution = request.form.get('resolution', None)
     gcs_output_bucket = request.form.get('gcs_output_bucket', None)
     generate_audio = request.form.get('generateAudio', 'false').lower() == 'true'
@@ -52,6 +52,41 @@ def generate_video_route():
         last_frame_filename_to_save = last_frame_img_filename
         print(f"Saved uploaded last frame image to: {last_frame_image_path}")
 
+    # Handle reference images for veo-2.0-generate-exp
+    reference_images_data = None
+    if model == 'veo-2.0-generate-exp' and aspect_ratio == '16:9':
+        reference_images_list = []
+        reference_type = request.form.get('reference_type', 'asset')  # Default to 'asset'
+
+        # Determine max images based on type: asset=3, style=1
+        max_images = 3 if reference_type == 'asset' else 1
+
+        # Handle multiple reference image files
+        for i in range(max_images):
+            ref_file_key = f'reference_image_{i}'
+            ref_file = request.files.get(ref_file_key)
+            if ref_file and allowed_file(ref_file.filename):
+                original_extension_ref = os.path.splitext(ref_file.filename)[1]
+                ref_img_filename = secure_filename(f"{uuid.uuid4()}_ref_{i}{original_extension_ref}")
+                ref_image_path = os.path.join(uploads_dir, ref_img_filename)
+                ref_file.save(ref_image_path)
+
+                reference_images_list.append({
+                    'filename': ref_img_filename,
+                    'type': reference_type  # Use the same type for all images in this request
+                })
+                print(f"Saved uploaded reference image {i} to: {ref_image_path}")
+
+        # Validate image count based on type
+        if reference_images_list:
+            if reference_type == 'style' and len(reference_images_list) > 1:
+                return jsonify({"error": "Style type supports maximum 1 reference image"}), 400
+            elif reference_type == 'asset' and len(reference_images_list) > 3:
+                return jsonify({"error": "Asset type supports maximum 3 reference images"}), 400
+
+            import json
+            reference_images_data = json.dumps(reference_images_list)
+
     new_task = VideoGenerationTask(
         prompt=prompt_text,
         model=model,
@@ -62,6 +97,7 @@ def generate_video_route():
         gcs_output_bucket=gcs_output_bucket,
         image_filename=image_filename_to_save,
         last_frame_filename=last_frame_filename_to_save,
+        reference_images_data=reference_images_data,
         user=user_email,
         generate_audio=generate_audio
     )

backend/tasks.py

@@ -31,20 +31,12 @@ def _run_video_generation(app, task_id):
 
         try:
             # Model specific checks based on user feedback
-            # User feedback: "veo-3.0-generate-preview dosen't support lart frame image and 9:16 ratio"
-            # Assuming "lart frame" means "last frame"
-            TARGET_MODEL_FOR_CHECKS = "veo-3.0-generate-001" # Or the correct model name if this is a typo
-
-            if task.model == TARGET_MODEL_FOR_CHECKS:
+            # Both veo-3.0-generate-001 and veo-3.0-fast-generate-001 now support 9:16 aspect ratio
+            # Only last frame image is still not supported for veo-3.0 models
+            if task.model.startswith('veo-3.0'):
                 if task.last_frame_filename:
                     task.status = "failed"
-                    task.error_message = f"Model {TARGET_MODEL_FOR_CHECKS} does not support last frame images."
-                    db.session.commit()
-                    print(f"Task {task_id} failed: {task.error_message}")
-                    return
-                if task.aspect_ratio == "9:16":
-                    task.status = "failed"
-                    task.error_message = f"Model {TARGET_MODEL_FOR_CHECKS} does not support 9:16 aspect ratio."
+                    task.error_message = f"Model {task.model} does not support last frame images."
                     db.session.commit()
                     print(f"Task {task_id} failed: {task.error_message}")
                     return
@@ -154,6 +146,48 @@ def _run_video_generation(app, task_id):
                 else:
                     print(f"Last frame image file {task.last_frame_filename} not found for task {task_id}")
 
+            # Process reference images for veo-2.0-generate-exp
+            reference_images_for_api = None
+            if task.model == "veo-2.0-generate-exp" and hasattr(task, 'reference_images_data') and task.reference_images_data:
+                reference_images_for_api = []
+                try:
+                    import json
+                    reference_images_list = json.loads(task.reference_images_data) if isinstance(task.reference_images_data, str) else task.reference_images_data
+
+                    for ref_img in reference_images_list:
+                        if 'filename' in ref_img and 'type' in ref_img:
+                            ref_img_path = os.path.join(uploads_dir, ref_img['filename'])
+                            if os.path.exists(ref_img_path):
+                                # Upload reference image to GCS
+                                if DEFAULT_OUTPUT_GCS_BUCKET:
+                                    storage_client_ref = storage.Client()
+                                    ref_bucket_name = DEFAULT_OUTPUT_GCS_BUCKET.replace("gs://", "")
+                                    bucket_ref = storage_client_ref.bucket(ref_bucket_name)
+                                    base_ref_filename = os.path.basename(ref_img['filename'])
+                                    ref_blob_name = f"reference_images/{task.id}/{base_ref_filename}"
+                                    blob_ref = bucket_ref.blob(ref_blob_name)
+
+                                    # Determine MIME type
+                                    ref_mime_type = "image/jpeg"
+                                    filename_lower = ref_img['filename'].lower()
+                                    if filename_lower.endswith(".png"):
+                                        ref_mime_type = "image/png"
+                                    elif filename_lower.endswith(".gif"):
+                                        ref_mime_type = "image/gif"
+
+                                    blob_ref.upload_from_filename(ref_img_path, content_type=ref_mime_type)
+                                    ref_gcs_uri = f"gs://{ref_bucket_name}/{ref_blob_name}"
+
+                                    # Add to API format
+                                    reference_images_for_api.append({
+                                        "image": {"gcsUri": ref_gcs_uri, "mimeType": ref_mime_type},
+                                        "referenceType": ref_img['type']  # "asset" or "style"
+                                    })
+                                    print(f"Successfully uploaded reference image {ref_img['filename']} to {ref_gcs_uri}")
+                except Exception as e_ref:
+                    print(f"Error processing reference images for task {task_id}: {e_ref}")
+                    # Continue without reference images if there's an error
+
             # Call GoogleVeo to generate video
             # Note: model_to_use (task.model or DEFAULT_VIDEO_MODEL) is not used here as GoogleVeo class has a hardcoded model.
             # This might be a point of future enhancement if model selection is needed with GoogleVeo.
@@ -167,7 +201,8 @@ def _run_video_generation(app, task_id):
                 last_frame_mime_type=current_last_frame_mime_type,
                 camera_control=task.camera_control, # Pass camera_control directly
                 generate_audio=task.generate_audio,
-                resolution=task.resolution
+                resolution=task.resolution,
+                reference_images=reference_images_for_api
             )
 
             # Process the result from GoogleVeo

frontend/public/locales/en/translation.json

@@ -142,5 +142,21 @@
   "videoEditingUnderDevelopmentNoticeTitle": "Video Editing Feature - Under Development",
   "videoEditingUnderDevelopmentNoticeBody": "The video editing (Weave) mode is currently under active development. You may experience performance issues or encounter bugs.",
   "videoEditingUnderDevelopmentNoticeSuggestion": "This feature is CPU intensive. If you find it very slow, please consider upscaling your server resources.",
-  "page": "Page"
+  "page": "Page",
+  "referenceTab": "Reference",
+  "referenceLabel": "Reference Images",
+  "referenceTypeLabel": "Type",
+  "referenceTypeAsset": "Asset",
+  "referenceTypeStyle": "Style",
+  "addReferenceImageButton": "Add Reference",
+  "removeReferenceImageButton": "Remove",
+  "referenceImagePreviewAlt": "Reference image preview",
+  "referenceNotSupportedMessage": "Reference images are not supported by this model.",
+  "maxReferenceImagesReached": "Maximum {{max}} reference images allowed.",
+  "maxAssetImagesReached": "Asset type supports maximum 3 reference images",
+  "maxStyleImagesReached": "Style type supports maximum 1 reference image",
+  "referenceImagesOnly16x9": "Reference images only support 16:9 ratio and do not support 9:16 resolution",
+  "uploadReferenceImageButtonTitle": "Upload reference image",
+  "pasteReferenceImageFromClipboardButtonTitle": "Paste reference image from clipboard",
+  "clearReferenceImageButtonTitle": "Clear reference image"
 }

frontend/public/locales/zh-CN/translation.json

@@ -141,5 +141,21 @@
   "videoEditingUnderDevelopmentNoticeTitle": "视频编辑功能 - 开发中",
   "videoEditingUnderDevelopmentNoticeBody": "视频编辑（织梦）模式目前正在积极开发中。您可能会遇到性能问题或错误。",
   "videoEditingUnderDevelopmentNoticeSuggestion": "此功能占用大量CPU资源。如果运行非常缓慢，请考虑升级您的服务器配置。",
-  "page": "页"
+  "page": "页",
+  "referenceTab": "参考图",
+  "referenceLabel": "参考图像",
+  "referenceTypeLabel": "类型",
+  "referenceTypeAsset": "资产",
+  "referenceTypeStyle": "风格",
+  "addReferenceImageButton": "添加参考图",
+  "removeReferenceImageButton": "移除",
+  "referenceImagePreviewAlt": "参考图像预览",
+  "referenceNotSupportedMessage": "此模型不支持参考图像功能。",
+  "maxReferenceImagesReached": "最多只能添加 {{max}} 张参考图像。",
+  "maxAssetImagesReached": "资产类型最多只能添加 3 张参考图像",
+  "maxStyleImagesReached": "风格类型最多只能添加 1 张参考图像",
+  "referenceImagesOnly16x9": "参考图功能仅支持16:9比例，且不支持9:16分辨率",
+  "uploadReferenceImageButtonTitle": "上传参考图像",
+  "pasteReferenceImageFromClipboardButtonTitle": "从剪贴板粘贴参考图像",
+  "clearReferenceImageButtonTitle": "清除参考图像"
 }