In the short-form video industry, the AI Kissing Generator significantly boosted the efficiency of content production through real-time rendering and facial motion capture technology. For instance, TikTok’s “Virtual Kiss” effect that will be rolled out in 2024. The application is based on a generative adversarial network (GAN) model and can produce a 1080P resolution kiss animation at 60 frames per second within 0.3 seconds. Each generation requires only $0.05, which is 99% lower compared to traditional 3D modeling. Data from Sensor Tower reveals that daily average volume of playing short video content along with this special effect was 470 million times on average. People’s frequency of usage increased 320% vs. the equivalent time last year, and developers’ revenue shares of advertising incomes were up to $1.2 per thousand plays (industrial average of $0.8).
At the business level, AI Kissing Generator presents new opportunities for promoting a brand. In 2023, L ‘Oreal launched the “AI Lip Print Challenge” on Instagram. Members created personalized digital kiss animations with the camera of their mobile phone. Within the first week of the launch of the campaign, 23 million individuals engaged. The brand exposure increased by 78%, and the product conversion rate was up to 6.4% (higher than the 3.1% of normal UGC activities). FaceUnity, a technology provider, said that the AI Kissing Generator module of its SDK allows it to reduce special effects development time from 6 weeks to 3 days, keep facial tracking error rate at 0.2 millimeters, support 98% of Android and iOS devices, and gain a customer renewal rate of up to 89%.
Usage behavior statistics show that the AI Kissing Generator significantly increases content interactivity. The Kuaishou platform Q1 2024 report indicates the average likes per short video empowered by this technology have risen to 12,000 for each video, which is 4.3 times the median of the platform. The comment interaction rate (CTR) rose to 18.7%, and the stay time of users has been extended to 34 seconds per video (the reference value is 19 seconds). The below analysis of the technology reveals that the Emotion recognition Accuracy of the resulting kissing action from AI is up to 92% (tested as per the EmotionNet dataset), and the Lip Sync Accuracy is 97.5%, outstripping the 95% of manual calibration.
Despite that, technical limitations continue to limit the range of its application. A 2024 Stanford University study found that under complicated lighting conditions (i.e., backlighting or heavy shadows), the deviation rate of the AI Kissing Generator’s lip deformation increased to 15%, requiring an additional 3.2 seconds/video of post-production correction. Additionally, under interactive conditions with multiple individuals, the bone collision detection error rate of two-person kissing animation reached 12%, resulting in 35% of users complaining of “limb penetration.”. To meet these challenges, the RTX 5000 series GPUs introduced by NVIDIA, with their ray tracing acceleration, increased the rate of the physical simulation to 240 frames per second and decreased the error rate to 1.8%. Although the hardware investment increased the unit investment of a single creator by $1,200, there are ethical implications too.
Ethical repercussions have also come into focus. The 2023 FTC report includes that the number of cases over the use of ai kissing generator to generate imitation celebrity kiss videos has increased by 470% year-to-year, and 83% of them fall under the case of unauthorized use of portraits. For this, Adobe has introduced the “Deep Watermarking” feature of Premiere Pro 2024, which is able to detect and mark AI-generated content with a confidence level of 99.3% with a false alarm rate of merely 0.07%. Marketer analysis firm eMarketer estimates that by 2025, 63% of short-video websites worldwide will have requirements for use of AI-based content identification systems, and that review costs of compliance will account for 7% to 12% of operational costs of such sites. User confidence, however, is expected to bounce back to 79%.
Despite being in trouble, the AI Kissing Generator is now a “traffic lever” in the short-videos business. From Douyin’s challenge competition to brand cross-border marketing, its characteristics of low cost and high iteration are transforming the chain of content creation. As computing power on the edge side for edge computing hardware (such as the AI calculation power of Qualcomm Snapdragon 8 Gen3 chip up to 75 TOPS) has strengthened, latency on real-time generation has been driven down to 8 milliseconds. In the future, it may extend further into occasions such as live streaming and virtual social communication and redefine the artistic boundaries of “instant intimate interaction”.
