AVA Discovery View: Netflix's Artwork Tool

Netflix has developed a tool called AVA Discovery View, which helps creative teams at Netflix efficiently create artwork for their titles. This tool utilizes machine learning to provide relevant still frames from video content, offering valuable insights for crafting captivating promotional assets.

• AVA Discovery View intelligently categorizes still frames based on storylines, prominent characters, environments, and potential sensitivities.
• The tool provides a curated selection of the best still frames, reducing the need for creatives to manually browse through entire videos.
• This approach allows for a more efficient and scalable artwork creation process, enabling Netflix to create compelling promotional material across various platforms.

AVA Discovery View's Features and Benefits

AVA Discovery View serves as a creative assistant, empowering creatives to produce impactful artwork for Netflix titles. It features a comprehensive set of tools and functionalities designed to streamline the artwork creation process and enhance the user experience.

• Storyline and Tone suggestions: AVA Discovery View offers suggestions based on the title's storyline and tone, allowing creatives to capture the essence of the narrative through visuals.
• Prominent Character identification: By highlighting the most prominent characters in a title, AVA Discovery View facilitates the selection of stills that showcase key talent.
• Sensitivity awareness: The tool provides insights into potential sensitivities within the content, helping creatives ensure the artwork is appropriate for all audiences.
• Environment exploration: AVA Discovery View surfaces stills that represent key environments in the title, giving creatives a visual understanding of the setting.

Addressing Challenges in Algorithm Development

Developing and refining the algorithms behind AVA Discovery View has presented various challenges, including ensuring accuracy, minimizing repetition, and optimizing user experience.

• Visual Search algorithm improvements: The team addressed the influence of text in images by filtering out stills with text results, focusing on visuals.
• Character identification challenges: The algorithm's ability to handle animated content has been improved, leading to more accurate suggestions for animated titles.
• Sensitivity algorithm refinement: False positives in the sensitivity detection have been minimized by increasing the confidence threshold.
• Repetition reduction: The team has implemented ranking and filtering systems to eliminate visually similar frames and promote diversity in suggestions.

Improving Suggestion Ranking and User Feedback

AVA Discovery View continuously seeks to improve its suggestion ranking and user feedback mechanisms to enhance the overall user experience.

• Category prioritization: Suggestions are presented in a hierarchical order based on their relevance to the user's workflow, ensuring the most important categories are shown first.
• Ranking by relevance: Suggestions within each category are ranked based on the number of results and confidence thresholds.
• Explicit feedback: Users can provide explicit feedback on suggestions through a thumbs up or thumbs down system, contributing to algorithm improvement.
• Implicit feedback: The tool tracks the use of suggestions, providing valuable data for algorithm refinement based on real-world utilization.

Technical Architecture and Scalability

AVA Discovery View is built upon a robust technical architecture, ensuring scalability and flexibility.

• Pluggable Architecture: The tool's design allows for easy integration of new algorithms and features, supporting its ongoing development.
• Unified Interface for Discovery: All Discovery View features share a common interface, simplifying extensions and integration with other Netflix platforms.
• Dynamic Category Management: Categories and recommendations are dynamically hidden or displayed based on the availability of suggestions, optimizing the user experience.
• Graceful Failure Handling: Independent loading of Discovery View suggestions ensures a responsive user experience even when encountering temporary issues.
• Asset Feedback Microservice: A dedicated microservice collects feedback on the quality of still frames, facilitating algorithm improvement and providing valuable data.

Integration with the Media Understanding Platform

AVA Discovery View seamlessly integrates with the Media Understanding Platform (MUP), leveraging its capabilities to streamline algorithm onboarding and enhance query features.

• Uniform Query Interface: MUP's uniform interface simplifies integration with various algorithms, enabling consistent query handling.
• Rich Query Feature Set: MUP offers a wide range of query options, allowing for testing of different confidence thresholds, intersection of algorithm suggestions, and flexible sorting options.
• Rapid Algorithm Onboarding: MUP accelerates the onboarding process for new algorithms, ensuring timely availability of suggestions for new Netflix titles.

Impact on Creative Teams and the Future of AVA Discovery View

AVA Discovery View has made a significant impact on Netflix's creative teams, enhancing their workflow and enabling them to create compelling artwork more efficiently.

• Increased Efficiency: By automating the process of finding relevant still frames, AVA Discovery View significantly reduces the time and effort required for artwork creation.
• Enhanced Quality: The tool's ability to suggest high-quality, contextually relevant stills leads to improved artwork quality, ultimately benefitting Netflix's promotional assets.
• Scalability: AVA Discovery View's scalability allows Netflix to handle the growing volume of content and maintain consistent artwork quality across its diverse library.

Netflix continues to invest in refining and expanding the capabilities of AVA Discovery View, focusing on:

• Minimizing Repetition: The team aims to eliminate redundancy in suggestions, ensuring each category presents a unique and valuable set of stills.
• Improving Frame Quality: Continued efforts are underway to ensure that only the highest quality frames are suggested, eliminating poor technical or editorial quality stills.
• Building User Trust: Netflix strives to enhance the user experience and build trust by providing a comprehensive, reliable, and intuitive tool that delivers exceptional results.