Ada Zhao
The Honor of Kings Auto-generated Game Highlight Video
Intelligent Auto Video-composing Design
Product Manager, Tencent, Shanghai
September 2022
Honor of Kings is the most popular MOBA mobile game in mainland China.
The auto-generated game report abstracts game highlights by analyzing the raw game data and edits according to the timestamp to present users with well-polished videos to publish.
The feature has acquired over one million authorized users with over 10 thousand videos published everyday.
Background
Little World is a short-video platform integrated into the QQ mobile application, one of the biggest social platforms in China, with teenagers as the main users. The objective of our team is to raise the number of daily video-post on the platform.
Data analysis showed that the game category had the most unbalanced ratio of video producers over video consumers among all content categories, which indicated a great potential growth in game-video production.
The question is:
what blocks our users from making game videos on their own?
User Research and Pain Points Diagnose
A user research was done among Honor of Kings game players. The survey showed that even though most players had the motivation to make some videos as a record or to make friends, they had great difficulties producing one.
Once a game is finished, the only way a player can retrieve the video footage is by screen recording while replaying the whole game. In terms of editing, a great short video consists of filters, slow motion, effects, background music, a cover, and so on, which blocks many beginners from starting.
Auto-generate Mechanism Design
I reached out to the Honor of Kings team and found they save game data that contains information about every frame of every player. With the information, we can set rules and retrieve precise video footage.
Once a game is finished, the game info will be used as follows:
To retrieve the precise footage, I define footage with a key factor and optional side factors. The key factor decides the timestamps of footage while side factors make sure the footage is a special one.
By combining the factors in different ways, we can easily form the logic to retrieve different footage as we want.
With more storylines being developed, we are building a library of factors for future reference.
Prioritization
To measure the ROI of each storyline and set priority, I broke down the ROI into a formula:
input = rules to cut footage * editing narrative
revenue = storyline coverage * sharing rate
User Experience Design
Considering privacy issues, a user must authorize their game data at first to enjoy this service. Since rendering an auto-edit video can take up to 1 hour on the server, the whole experience is an asynchronous experience.
Outcomes
In the past half a year, four versions have been released to the public with more than one million authorized users. Daily, 10 thousand auto-generated game videos are published on the platform.
In the future, I am thinking about optimizing the editing design process. Instead of me writing down when and how to add a cover, I am thinking about a tool with GUI to facilitate content designers to define the storytelling.