Matthew Tang

This commit introduces a **new feature** by adding a `labels` field to the **video generation configuration** within the `google.genai` library. Users can now specify custom labels for billing usage through the updated `GenerateVideosConfig` and `GenerateVideosConfigDict` types. The `google/genai/models.py` module includes new logic to process these labels, ensuring they are correctly converted for Vertex AI (`_GenerateVideosConfig_to_vertex`) while also implementing a check for unsupported usage in the Gemini API. This **enhances billing traceability** for video generation requests and ensures proper backend integration.

This commit introduces **Image Grounding support** to the **GoogleSearch tool**, significantly enhancing its capabilities within the `google.genai` library. This **new capability** allows the tool to process image-based search queries and control specific parameters like `prominent_people` detection in image configurations. The implementation involves defining **new types** for web and image search in `google/genai/types.py` and updating the **conversion logic** for live API, tokenization, batch processing, caching, and model interactions across several `google/genai` modules. This enables more sophisticated **multimodal search functionality** by integrating image understanding directly into the `GoogleSearch` tool's operations.

This commit **updates the model ID** for the **Virtual Try-On feature** to `virtual-try-on-001`. This is a **documentation and maintenance update** that primarily affects **example usage** within the `Models` and `AsyncModels` classes in `google/genai/models.py`. Additionally, an **internal test constant** `VIRTUAL_TRY_ON_IMAGE_MODEL_LATEST` in `google/genai/tests/models/test_recontext_image.py` is updated to reflect this new ID. The change ensures that all references to the Virtual Try-On model are consistent and up-to-date, providing accurate guidance for users and maintaining test integrity.

This commit introduces a **new capability** by adding the `person_generation` field to the **`ImageConfig`** class in `google/genai/types.py`, enabling users to control the presence of people in generated images. This new configuration option is specifically integrated and supported for **Vertex AI** image generation, with mapping logic added to the `_ImageConfig_to_vertex` function in `google/genai/models.py`. To ensure correct usage, a `ValueError` check was implemented in `_ImageConfig_to_mldev` across `google/genai/batches.py` and `google/genai/models.py` to explicitly indicate that `person_generation` is **not supported** when targeting the Gemini API. This enhancement provides more granular control over image content for users leveraging Vertex AI.

This commit **introduces new enum values**, `IMAGE_RECITATION` and `IMAGE_OTHER`, to the `FinishReason` enum within the **Google GenAI library's types module** (`google/genai/types.py`). This **enhances the system's ability to categorize and communicate** specific reasons why an image generation process might stop. The **new capability** provides **more granular control and understanding** of image generation outcomes, particularly for scenarios beyond standard completion or error, improving the robustness of image generation status reporting.

This commit **updates the documentation** in `README.md` to reflect the latest model IDs for **Imagen** and **Veo**. It specifically revises **code examples** to use the correct identifiers and **clarifies details related to video generation**. This **documentation update** ensures users have accurate and current information when working with these models, particularly for **video generation capabilities**.

This commit introduces **new image output configuration options** to the `ImageConfig` class, allowing users to specify `output_mime_type` and `output_compression_quality` for generated images. This **new feature** significantly enhances control over **Vertex AI image generation outputs**, enabling users to define the desired format and quality. It involves updates to the `ImageConfig` and `ImageConfigDict` definitions in `google/genai/types.py`, alongside new conversion logic in `google/genai/models.py` and `google/genai/batches.py` to properly translate these settings for Vertex AI and mldev formats during content generation and batch processing.

This commit performs a **maintenance update** to the **Imagen API test suite** within the `google/genai` library, migrating tests to use the latest **Imagen 4.0 models**. It updates the default model versions in `test_generate_images.py` and `test_upscale_image.py` to `imagen-4.0-generate-001` and `imagen-4.0-upscale-preview` respectively. This ensures that automated tests for **image generation** and **image upscaling** are validated against the most current model versions. Additionally, the update includes removing outdated test cases and adjusting image size parameters, enhancing the **relevance and accuracy of the test coverage**.

This commit delivers a **bug fix** to the **`google.genai.models` module**, specifically addressing the **`recontext_image` functionality**. It **corrects the parameter path** for `base_steps` within the `_RecontextImageConfig_to_vertex` function, which is responsible for converting image recontextualization configurations. This ensures that the `base_steps` parameter is accurately processed and applied when interacting with **Vertex AI integrations**, preventing potential misconfigurations or errors during image recontextualization.

This commit provides a **bug fix** for the **`google.genai.models`** module, specifically addressing an issue with the `list()` method. It **corrects the URL construction logic** within `Models` and `AsyncModels` to properly handle and format the `filter` query parameter, ensuring that API requests are made correctly when filtering models. This allows users to reliably filter model listings, impacting both synchronous and asynchronous operations. A new test case has been added to `google/genai/tests/models/test_list.py` to validate the correct behavior of filtering tuned models.

This commit performs **maintenance updates** to the **video operations** handling and its associated test suite within the `google.genai` library. A **response verification step** has been added to `google/genai/operations.py` to ensure the integrity of video operation responses, enhancing the robustness of this feature. Correspondingly, the `test_create_operation_to_poll` test in `google/genai/tests/models/test_generate_videos.py` has been updated with new operation names to reflect **updated test recordings**. This ensures the **video generation tests** remain accurate and reliable, improving the overall stability of the `genai` client's video capabilities.

This commit introduces a **new capability** by adding an `image_size` field to the **ImageConfig** class and `ImageConfigDict` within `google/genai/types.py`. This enhancement allows users to explicitly specify the desired dimensions for generated images, providing more granular control over the output. As part of an **Early Access Program**, this feature empowers developers to tailor image generation parameters more precisely.

This commit introduces **new configuration parameters** for the **Imagen image upscaling API** within the `google.genai` library. It defines `safety_filter_level` and `person_generation` fields in the `UpscaleImageConfig` types, enabling users to control content safety and the generation of people during the upscaling process. The `upscale_image` and `upscale_image_async` functions in `google/genai/models.py` are updated to accept and correctly convert these new parameters for the Vertex AI backend. This **feature enhancement** provides more granular control over the image upscaling process, with corresponding test updates to ensure functionality.

This commit performs **maintenance** on the **Google GenAI library's video generation tests** by updating them to utilize **Veo 3.1 model versions**. Specifically, `google/genai/tests/models/test_generate_videos.py` is modified to include the `mime_type` parameter for video sources, ensuring compatibility with the latest Veo API. This update also adjusts expected exception messages to align with the behavior of Veo 3.1, ensuring the test suite remains robust and accurate.

This commit **fixes a bug** in the **`google.genai` library's credential precedence logic**, specifically for **Vertex AI API keys**. It ensures that when initializing a client, **explicitly provided API keys or credentials will correctly override any Vertex AI API key found in environment variables**, preventing unintended credential usage. The change modifies the `__init__` method within `BaseApiClient` in `google/genai/_api_client.py` to adjust this behavior. A **new test case** has been added to `google/genai/tests/client/test_client_initialization.py` to validate that explicit credentials are now correctly prioritized during client initialization in Vertex AI mode. This improves the reliability and predictability of credential management for `google.genai` users.

This commit provides a **bug fix** to ensure proper asynchronous video processing when generating **Veo Video Extensions** using the **Gemini Developer API**. It **modifies** the `generate_videos` method in `google/genai/models.py` to **remove `video_bytes`** from video objects before they are sent to the API. This change is crucial because the Gemini Developer API does not support `video_bytes` in this specific context, preventing potential errors during video extension generation. An accompanying asynchronous test case has been added to verify the successful generation of video extensions from a source video with this corrected API interaction.

This commit provides a **documentation update** by adding a new example to the `README.md` file. This **enhancement** specifically demonstrates how to leverage the **Gemini API** for generating image output from text input, showcasing a key capability of the project. The addition improves the clarity and utility of the project's **documentation**, making it easier for users to understand and implement **text-to-image generation** features.

This commit performs a **maintenance chore** to **update the package version** for the `python-genai` library, signifying a new release. It specifically modifies the `google/genai/version.py` file to increment the version string from `1.43.0` to `1.44.0`, and also updates the `CHANGELOG.md` to reflect the changes included in this release. This change officially marks the availability of **version 1.44.0** of the `python-genai` library, making any new features, bug fixes, or improvements from this development cycle accessible to users.

This commit introduces **support for video extensions** within the **Gemini Developer API**, specifically for Veo. It **enhances the `google.genai` library** by updating video data conversion functions and adding new logic to `generate_videos` to properly handle video URIs and encoded video bytes. This **new feature** involves refactoring internal video data extraction in `google/genai/_operations_converters.py` to adjust key access for video data. Comprehensive test cases have also been added in `google/genai/tests/models/test_generate_videos.py` to verify the new video extension functionality. Developers can now leverage advanced video capabilities when interacting with the Gemini API.

This commit **introduces support for Imagen's EditImage functionality** within the **Dotnet SDK**, providing a **new capability** for developers. This allows programmatic access to advanced image editing operations using the Imagen model directly through the SDK. A minor adjustment was made to a prompt string in `test_edit_image.py` to refine the testing of this new feature. This significantly expands the **image generation and manipulation capabilities** available to Dotnet developers, enabling more dynamic and creative applications.