Image AI, also known as visual recognition, refers to the use of artificial intelligence and machine learning techniques to analyse and understand digital images and videos. It involves the training of models to automatically identify patterns, objects, and features within images and videos. On Pulsar our Image AI enhances your ability to process and interpret visual information present in diverse datasets, particularly via vertical image classification. This enables brands across various sectors and use cases to efficiently analyse visual content, extract insights about customer preferences, and gain an understanding of the type of visual content that their target audience associates with their brand.

On Pulsar, we provide Vertical Image Analysis, a technology that utilises different Image AI models that have been specifically trained to analyse and detect images for particular domains and use cases. This enables us to offer more accurate results compared to the broad categorisation of images through general image tagging. Our models provide more detailed insights about the subject matter in the images.

For instance, while general image tagging may only identify the presence of food in an image, our Food model can accurately recognise specific food items and dishes in the picture, even down to individual ingredients. Restaurants or food and beverage brands can leverage the Food image analysis model to analyse food images and extract insights about trending recipes, cuisines, and restaurants. Similarly, retail businesses can utilize other models such as the Apparel or Make-up looks models to identify fashion and beauty trends.

Furthermore, image analysis can assist brands in making informed decisions about their creative and content strategy by providing more insight into the type of media that resonates with their audience and drives higher engagement.

In addition to general image tagging, which is trained to analyse and classify a variety of concepts and common objects, we currently offer the following Vertical Image AI models on TRAC: Food, Food Beauty, Travel, Interior Styles, Apparel, Make-up Looks, Logo Detection, Colour Detection, Celebrity Identification, Demographics Detection, Instagram Popular Shots and Video Analysis

Our Food image analysis model allows for the accurate identification of specific food items and dishes depicted in images, even detecting individual ingredients. Food and beverage businesses can utilize this technology to analyse food images and gain insights into trending recipes and cuisines, providing inspiration to create more effective marketing campaigns and produce content that better connects with their audience. Restaurants can also benefit from these insights by using them as inspiration for creating new menus, seasonal menus, and pairing food in creative and innovative ways that their target customers would enjoy.

With this model you can group food images by how staged or appealing the food looks. Unlike the Food model, which returns a list of food items identified in a picture, the Food Beauty model is a binary model, which returns a tag telling you whether the food layout is appealing, or not. How can you use this model? Eating, for so many people, is an experience best captured visually. And if you're a brand manager for a food company, you can use this model to understand the brand perception around your products. It's not just the ingredients or recipe that matters, the appearance of the food is in itself a form of art, and people are more likely to share pictures of their food if it is “beautiful” to look at. Access to this user generated content can provide you with valuable insights around your food products or campaign hashtags, for example whether or not it’s associated with highly curated or simpler, more “natural” food shots.

The travel model recognises travel and leisure-related features of hotels, resorts, residencies, and travel properties such as an outdoor pool, lounge, or spa. It’s a great model for anyone building travel and hospitality-related apps, or for travel and holiday companies to create campaigns that resonate with their target customer. In addition to effective marketing campaigns, this model can be useful for helping companies to create enjoyable experiences and travel itineraries that their customers would love.

With this model, you can recognise and classify fashion-related and clothing concepts, hats, jewellery, handbags, etc. in images and videos. It can identify over 100 concepts in a wide range of categories, ranging from clothing and accessories to jewellery, patterns, and fabric texture. This is a great model for brands and magazines to help identify the latest fashion trends, inform new product lines and put together looks that appeal to their target audience as part of their marketing strategy.

With this model, you can recognise different makeup styles in your dataset. The model returns the following tags: bold, bold eye, bold lashes, bold lip, bridal, cat eye, festival, glamour, glitter, luminous, matte, red lip, shimmer, smoky, subtle. An art director at a cosmetics company may use this model to quickly understand the trends, styles, and vibe of a given community, and see what is trending within it, creating a data informed mood board based on large collections of authentic images.

This model recognises interior decor styles of the pictures in your dataset. It returns the following tags: bohemian, classic, coastal, contemporary, eclectic, farmhouse, glam, global, mid-century modern, minimalist, modern, preppy, rustic, Scandinavian, southwestern, traditional. This model might be used by a furniture company to recognise popular decor in a given audience or around a trending hashtag. The brand can discover how to display its furniture in a showroom in a way that is going to resonate with its target audience. This model is also useful to spot emerging trends in interior design, and uncovering unexpected elements and accents for creative inspiration.

This model is capable of detecting the presence and location of brand logos in images, catalog and white backgrounds, as well as in user generated content (UGC) images. Please note that the model will not be able to detect logos outside the companies and industries that it was trained on.

The Colour model returns density values for dominant colours present in images. Colour predictions are returned in hexadecimal format and also mapped to their closest W3C counterparts. This model is great for brands in the design or interior deco space where colour palettes are a big feature. Fashion and Design brands can leverage this model and use it to analyze fashion trends and colour palettes used in clothing and accessories. It can also help in designing products that match the desired colour schemes of their target customer.

The Celebrity model detects whether images contain celebrity faces. Trained with over 10,000 recognised celebrities, the ‘Celebrity’ model analyses images and returns probability scores on the likelihood that the image contains the face of a recognisable celebrity. This could be especially valuable for brands that want to collaborate with celebrities on marketing campaigns, as they can use the model to quickly and easily identify which images contain celebrities who use their products.

The demographics model utilises image analysis to predict the age, gender, and cultural appearance of individuals featured in images and videos, classifying them into multicultural categories such as White, Middle Eastern, Latino_Hispanic, and Indian, as well as into specific age ranges. The model can be used to provide brands with insights into the demographics of their product consumers and online followers who share their affinity for the brand by sharing pictures of themselves. This can be particularly valuable for brands seeking to target specific demographics with their marketing efforts.

With this model you can detect and classify typical social media shots in your dataset. The model returns the following tags: fitness, food, from where I stand, interior design, makeup shot, manicure shot, mirror selfie, party, pet shot, selfie, sky shot, workplace shot. Using Instagram Popular Shots can help you get an overview of your audience’s main behaviours during specific moments. If you were running a campaign for a festival, or a sports event, what are the popular shots that people tend to take when they are in that particular moment? By spotting, for instance, a high percentage of selfies in a certain dataset you might use that insight to craft a campaign strategy, or a brief centred around that behaviour.

Firstly, you'll need to enable your desired Image AI analysis at Search Setup as shown below.

Once an image analysis model has been selected and enabled in a Search, you will be able to see the concepts identified in each image in the Content section under Most Shared Images.

In the Content section, we also visualise the top image concepts, segmented and organised into a network. The chart displays related images grouped together into distinct segments. By using this chart, you can distinguish different images that are associated with your brand, campaign, or a specific subject. Our clustering algorithm analyses each image segment and determines the relevance and importance of each image within it helping you to uncover the dominant and most important visual themes present in a given dataset.

Lastly, you can find all the list of concepts identified in a single image when in the Feed Results, as shown below. A single image can have multiple tags, and each tag will have a confidence score on a scale of 0 to 1.

​