Reviving Retro Legends: The Power of Generative AI
Wynk Music, a division of Airtel Digital Limited, stands at the forefront of India’s music streaming apps, leading in both downloads and daily active users. We aim to boost artists’ following, ensuring that users remain closely connected to their favourite artists. At the same time, we are focussed at enriching user experiences through a deep understanding of our users’ listening patterns and actionable insights.
However, we faced a challenge… a challenge that demanded attention to guide users towards listening to their favourite artists.
The primary challenge was that artist images were not meeting expectations; they appeared cropped and of low quality, thereby diminishing the user experience. As a result, users did not feel connected to their favourite artists.
We wanted to revive the nostalgic charm of the past by featuring legendary artists such as Kishore Kumar, Lata Mangeshkar, Dev Anand, Mohammed Rafi, and others.
We understand the profound emotional connection users have with their favourite artists, and we are determined to enhance this experience through our platform.
To address this, we embarked on a journey to explore potential solutions.
Challenges we faced
Our aim was to present the iconic artists on our platform in the best possible way. However, many artist images were either deformed, incomplete, or lacked clarity. We needed to do something quickly.
We meticulously outlined the obstacles, enabling us to tackle them one by one and achieve our desired result.
The primary observation was that every image looked different, with each having multiple issues. These issues were:
- Lack of high resolution: Many artist images in our collection lacked high resolution, adversely affecting their display quality and making it difficult for users to discern details.
- Pixelated images: In some cases, while the resolution may be correct, the images were pixelated, resulting in a lack of clarity.
- Outdated appearance: For most of the legendary artists, their images were in sepia tone and contained noise, which gave them an outdated and less visually appealing appearance.
- Varied colour tone: The colour tone of images varied drastically and there was no standard, resulting in lack of consistency and authenticity.
- Cropped images: Some artist images were improperly cropped, cutting off heads and shoulders, leading to an incomplete and unprofessional look.
Finding a solution
We explored various tools and conducted extensive research to find a solution. Subsequently, we discovered Generative AI. The Generative AI tools had the capability to enhance picture quality and refine images by providing precise prompts and offering multiple templates to choose from. It was like having a fleet of professionals at our disposal.
We utilised Generative AI to generate cropped images of heads, shoulders, hands, hair, and more.
Generating faces, hands, and other details:
- While trying to generate the entire image at once, we faced several issues, such as improper hand and body proportions. However, despite these challenges, it instilled hope and encouraged us to progress further in our journey of image enhancement.
- We opted to generate the image in parts, starting with the shoulder, then the body, followed by the hands, and finally the hair. This approach led us to the desired results.
- After employing Generative AI to enhance our images, we used image editing tools to further refine images to our desired level.
Enhancing facial features:
- After generating images with Generative AI, we encountered issues with colour range, saturation, exposure, and clarity. To correct the facial features of the artist image, we manually fine-tuned the colour range and exposure, applied noise reduction for improved clarity, and adjusted the saturation and texture of the face to achieve a more natural appearance.
- We use JPEG images for editing, which are compressed files. Compression can result in blockiness, noise around edges, and colour degradation in images. By removing artefacts from subjects, we ensure they stand out prominently, even in extreme close-ups, through these enhancements.
Adding colour to old black-and-white photos: We utilised Generative AI filters to convert black-and-white photos into vibrant, eye-catching colour compositions. Additionally, we implemented a uniform tone for artist images to ensure consistency throughout the app, thereby enhancing user experience and maintaining coherence.
Outcomes
Following the update of artist images in the app, there was a significant increase in their sharing across social media platforms, specially during the 2023 Rewind campaign.
Throughout the campaign, artist playlists accumulated millions of clicks, and the new enhanced artist images played a significant role. This also helped Wynk improve its position, bringing it on par with other competitors in the music app industry.
Future goals
Despite the assistance of Generative AI in enhancing artist images, the task remains tedious due to the large volume of images we handle in the app. Therefore, we are in the process of developing an in-house tool to expedite this process and enhance the quality of artist images across the entire app.
In collaboration with our development team, we’ve devised automation guidelines and processes to improve artist images at scale, and hence allow our listeners to connect with their favourite artists.
Conclusion
In conclusion, leveraging Generative AI to enhance artist images in the app has yielded positive feedback from users, motivating us to persist in our efforts and strive for continuous improvement. Currently, we are developing a tool to expedite this process and enhance images more efficiently.
Our commitment to enhancing artist images remains steadfast, aiming to strengthen the bond between our users and their favourite artists. This journey of improvement is ongoing and will persist into the future.
