Core Insights - Lemon Slice is developing a new diffusion model that generates digital avatars from a single image, aiming to enhance video interaction in AI applications [1][5] - The model, named Lemon Slice-2, has 20 billion parameters and can generate video streams at 20 frames per second, allowing businesses to integrate virtual human functionalities easily [5][6] - The company has completed a seed funding round of $10.5 million to support technology development and business expansion [6][9] Company Overview - Lemon Slice was co-founded in 2024 by Lina Koluchy, Sidney Plamas, and Andrew Wetz, focusing on creating differentiated virtual human products through self-developed models [6][8] - The company aims to address the shortcomings of existing virtual human solutions, which often fail to provide a satisfactory user experience [6][8] - Lemon Slice has established a risk control mechanism to prevent unauthorized cloning of faces and voices, ensuring compliance in content generation [3][7] Technology and Innovation - The diffusion model allows for the generation of both human-like and non-human virtual characters, catering to diverse application scenarios [2][5] - The technology is based on reverse learning from noisy training data, enabling the creation of new data content [6][9] - The model's capabilities are expected to surpass the limitations of existing technologies, potentially overcoming the "uncanny valley" effect in virtual human interactions [9] Market Position and Competition - Lemon Slice faces competition from various startups in the digital video generation space and established virtual human developers [8] - Industry experts believe that the technology will see significant growth in video-centric applications, with Lemon Slice's approach providing a competitive edge [8][9] - The company plans to expand its team and invest in computational resources for model training, indicating a commitment to scaling its operations [9]