Artificial Intelligence (AI) has transformed numerous industries, including the media industry. AI is about automating repetitive tasks and facilitating better decision-making by providing data insights. With AINAR V4, The Next Level of Cognitive AI Technology, we have enhanced our ability to offer AI-powered services to our clients. AINAR V4 is the most advanced AI product we have created and a game-changer in the world of cognitive AI technology.
AINAR V4, our latest Fingerprint+ product, is based on a powerful 1088-dimensional content embedding, making it the most advanced AI product created by Vionlabs. The content embedding is a representation of an audio/video asset that maps each asset to a vector in a high-dimensional space. The dimensions of the embedding represent various features of the video extracted by our proprietary Deep Learning algorithms. The exact interpretation of each dimension is not immediately obvious, and we perform several downstream tasks, such as keyword prediction or movie similarity clustering, to understand what information the embedding contains. We ensure that our embedding contains information about the emotional state of the content asset by adding specific features and labels to the training of our embeddings.
AINAR V4 has several new features that differentiate it from its predecessor, AINAR V3. We have increased the number of dimensions from 16 to 1088, which allows us to encode more information about each content asset into the embedding. AINAR V4 can predict and produce over 40 genres, 1500 keywords, 40 mood categories, 70 mood tags, similar content clustering, contextual ad cue point detection, contextual scene-level data for targeted advertising, intro detection, recap detection, credit detection, thumbnail generation, preview clip generation, and short-form to long-form content connection.
| Version | V3 | V4 |
| Dimensions | 16 | 1088 |
| Capabilities | Content Similarity, genre prediction, Mood Categories, Keywords | Mood tags, highlight generation, binge marker detection, thumbnail selection, contextual ad breaks + 4x more tags predicted per title, short-form to long-form connections |
| Performance | Performed well on mainstream long-form content | Handles all the worlds content including gameplay, anime, Bollywood, short-form, reality-tv, cooking, sport match, concerts, etc |
| Potential Applications | Enhance recommendations and editorial list creation | Creating a “shorts” like experience with highlight clips and metadata, contextual targeting for scene-level ads, personalized artworks and much more |
The Next Level of Cognitive AI Technology
AINAR V4 can recognize diverse types of content, including sports, concerts, Bollywood movies, animated content, anime, reality TV, cooking shows, talk shows, video games, game shows, documentaries, and more. It can identify the correct mood of each piece of content with over 40 mood categories and 70 mood tags. AINAR V4 can detect ad breaks that do not interrupt the storyline and provide contextual ad data, allowing media companies to optimize their ad campaigns and increase their ROI. It can accurately detect and track different segments of media content, including Binge Markers such as Intro, Recap, and Credit detection. It can automatically generate thumbnails and preview clips, which is useful for streaming companies as content libraries grow larger. AINAR V4 can connect short-form to long-form content, providing streaming services with data on which preview clips people watch to optimize their content recommendations.
In conclusion, as our datasets grow, as the research field of AI keeps growing at a rapid pace, and as we learn more about how AI can create value for our customers, AINAR will keep growing bigger, better, and smarter every day. This is only a sneak peek into some of the capabilities we are launching in our v4 release, but in the next blog we’ll give an introduction to how we can use json commands to ask AINAR to find specific scenes in a content asset based on desired emotional parameters e.g. “give me a high-energy, joyful clip, that’s 60 seconds and that includes the main character”. Stay tuned!
