News
Person
Technology, Artificial IntelligenceJuly 14, 2023Meta unveils CM3leon: A new AI image generation model with enhanced efficiencyMeta has made significant strides in the field of generative AI models with the introduction of CM3leon. This groundbreaking model focuses on text-to-image and image-to-text generation, employing a token-based autoregressive approach instead of the commonly used diffusion models. Meta's research indicates that CM3leon outperforms previous transformer-based methods in text-to-image generation, even though it was trained using only a fraction of the computational resources. A notable aspect of Meta's approach is its commitment to ethical image training, relying solely on licensed images from Shutterstock to avoid concerns related to image ownership and attribution. The CM3leon model undergoes a retrieval-augmented pre-training stage followed by supervised fine-tuning, resulting in optimized resource utilization and high-quality image generation. While CM3leon is currently a research effort, the impressive results and increased efficiency suggest that Meta may eventually integrate this technology into their platforms, further revolutionizing the field of generative AI. With its ability to understand complex prompts and generate extremely high-resolution images, CM3leon showcases Meta's dedication to advancing the capabilities of generative AI models while addressing ethical concerns in image data sourcing.
Source: VentureBeat.com
;