According to the 2023 Generative AI market Research Report, the global image to video ai technology market size has reached 4.7 billion US dollars, but the number of related copyright dispute cases has increased by 320% year-on-year. For instance, in 2023, the United States Copyright Office ruled that if the videos generated by the ai video generator rely entirely on copyrighted input images (such as Disney characters or National Geographic photographic works), the output content does not enjoy independent copyright, resulting in a start-up company training a model using 12,000 unauthorized news images. Facing a compensation lawsuit of up to 18 million US dollars. From a technical perspective, the training datasets of mainstream ai video generator tools such as Runway ML and Pika Labs contain over 250 million online images. Among them, approximately 34% of the sources do not clearly indicate the copyright status, which may violate Section 512 of the Digital Millennium Copyright Act (DMCA).
From the perspective of commercial applications, the average cost of generating a 10-minute video using image to video ai is $45, which is only 1/60 of that of professional film and television production. However, enterprises need to pay an additional 15%-30% of the budget for copyright review. For instance, Shutterstock launched a “Copyright Liquidation Fund” in 2024 to address compliance issues with its training data, paying creators who contribute images a licensing fee of $0.02 per image. The cumulative expenditure has exceeded $8 million. Adobe’s Firefly system has reduced the infringement risk from the industry average of 22% to 4.7% by training the model using only its own image library (approximately 300 million compliant materials). For creators, a certain TikTok user used ai video generator to convert 300 Van Gogh paintings into animated videos. The single video had more than 240 million views. However, due to not obtaining authorization from the Van Gogh Museum in the Netherlands, the account was permanently banned and all the related advertising revenue of 127,000 US dollars was frozen.
In terms of technical standards, the EU’s “Artificial Intelligence Act” requires that the ai video generator platform must record the metadata of all input images (including the model of the shooting device, GPS coordinates and creation time), and the storage period must be no less than 10 years, which increases the operating cost of the system by 18%-25%. In May 2024, OpenAI was fined 3.7 million euros by the European Union for failing to implement the standard in the Sora model. It is worth noting that research by the Computer Vision Laboratory of Stanford University has found that among the current videos generated by image to video ai, 14.6% of the frame images have a similarity of more than 90% to the images of the training data source. This high-density replication behavior may trigger the infringement determination criterion of “substantial similarity” stipulated in Article 9 of the Berne Convention. Market monitoring data shows that the user growth rate of platforms adopting copyright filtering technology (such as Artbreeder) has reached 67%, much higher than the 23% of those not adopting it, proving that compliance has become a key factor affecting the market competitiveness of ai video generator.