The PFP team FIFA World Cup AI video originally started on the day of the Dragon Boat Festival. I thought to myself: after making the AI Dragon Boat video, why not take on an even more unique challenge?
The AI Dragon Boat video was essentially my first attempt at creating a multi-person AI “transformation” (reincarnation) video. I was personally quite satisfied with it, though during the creative process, I also came to realize the current technical limitations and challenges of AI video generation.
Current Challenges with Image-to-Video
For instance, I found that it is currently still impossible to perform a 1-to-1 AI transformation for multiple people within the same multi-person shot. Because the process of turning static images into video with AI is not like traditional AI face-swapping; instead, the AI learns and understands character features before regenerating a dynamic character. However, when a frame requires multiple characters to be dynamically generated simultaneously, the features of each character end up being weighted and averaged out during the machine learning process. To our human eyes, the characters lose their resemblance to their original real-life counterparts because the feature details get overlooked by the AI.
As I understand it, the more characters sharing the screen, the easier it is for each character’s unique traits to be ignored, making them look less like the original—or rather, everyone starts looking more and more alike. This is because the AI averages out and scatters everyone’s features across the generated characters.
Therefore, after completing the AI Dragon Boat video, I thought of an event that was gaining huge momentum at the time: the kickoff of the 2026 World Cup. For this new creation, I planned a strategy where every candidate would get an opportunity to shine dynamically, rather than relying merely on static images for scene transitions.
So I came up with a solution: design shots where each candidate player gets a solo moment dynamically playing soccer. This allows the AI to focus on generating dynamic movement for a single, unique character—at most paired with fictional rival players generated by the AI, whose appearances weren’t crucial to my requirements. Since Google Gemini Omni generates 10-second videos at a time, I spent several days generating about three separate soccer-playing clips for each candidate to ensure I had enough editing material and a sufficient final video duration.
Another Major Challenge
Although Google Gemini Omni can reference multi-angle photos of a specific character, Google products have always sidestepped sensitive topics, such as violence, pornography, elections, and political figures. Chairman James Soong is a well-known political figure recognized by Google—even a video of calligraphy for Spring Festival couplets can be flagged by the AI as election-related… Of course, I am well aware of this; when you input “James Soong” into Google Translate, the Chinese translation is indeed “宋楚瑜”.
My takeaway is that generating videos of political figures with AI using Google Gemini Omni or Google Veo is currently something that can only be implied, not explicitly stated. When intending to get the AI to generate a video of Chairman James Soong, I can only hint at it subtly—hehe.
The Final Challenge
If a YouTube Short uses music tracks provided by the YouTube platform, the video length must be kept under 60 seconds due to Google’s music copyright regulations. After editing, this World Cup video turned out to be 136 seconds long. Fortunately, this creation didn’t have that constraint because the background track I used was an audio file I owned myself—generated via Google Lyria within Google Gemini. Naturally, during video editing, the soundtrack or song is the soul of the video and is extremely important; the rhythm and atmosphere must align, and the music and visuals must complement each other seamlessly.
Chairman Soong was very pleased with this AI World Cup video and expressed that he was deeply moved, calling it a demonstration of teamwork: “We are a team.”
For me, the greatest encouragement comes when the viewers’ feedback and emotions align with the creator’s intent and initial purpose—that is the highest affirmation for an AI video project.
The above is my personal reflection on applying AI video creation to the People First Party’s (PFP) 2026 election campaign.
As for how I plan to conceptualize the next AI video for the PFP’s 2026 campaign, I haven’t made up my mind yet. This is a continuous journey of learning and challenging creative tasks for me, as well as a process of learning and evolving for the AI itself.