Meta releases new AI model, ‘CM3leon’ to generate text & images

• Meta’s CM3leon excels in vision-language tasks such as visual question answering and long-form captioning.
• Meta introduces CM3leon, an innovative AI model for text-to-image and image-to-text generation.
• This AI model is designed to perform both text-to-image and image-to-text generation tasks.

Additionally, Meta stated that CM3leon demonstrates exceptional performance in various vision-language tasks, including visual question answering and long-form captioning. Despite being trained on a smaller dataset of only three billion text tokens, CM3leon’s zero-shot performance is comparable to larger models trained on larger datasets.

Meta, formerly known as Facebook, has launched an innovative generative AI model named “CM3leon” (pronounced chameleon). This AI model is designed to perform both text-to-image and image-to-text generation tasks.

In a recent blog post, Meta announced that “CM3leon” is the first multimodal model to be trained using a recipe adapted from text-only language models. The training process involves a large-scale retrieval-augmented pre-training stage, followed by a multitask-supervised fine-tuning (SFT) stage.

Meta releases new AI model, ‘CM3leon’ to generate text & images

Meta has highlighted that CM3leon’s advanced capabilities enable the generation of more coherent and contextually aligned imagery based on input prompts. In comparison to previous transformer-based methods, CM3leon requires only five times the computing power and a smaller training dataset to achieve its impressive results.

Meta’s newly introduced AI model, CM3leon, has achieved remarkable success in text-to-image generation, surpassing Google’s Parti model and establishing a new state-of-the-art FID (Frechet Inception Distance) score of 4.88 when compared to the widely used MS-COCO benchmark.

CM3leon’s exceptional performance extends beyond text-to-image generation. It showcases excellence in various vision-language tasks, including visual question answering and long-form captioning. Remarkably, CM3leon’s zero-shot performance remains competitive even when compared to larger models trained on significantly larger datasets, despite being trained on a dataset comprising only three billion text tokens.

Also Read

WhatsApp’s newest feature improves group chats

WhatsApp enhances group chat feature for a better user experience. The latest...

Meta expressed its belief that CM3leon’s strong performance across multiple tasks represents a significant advancement toward achieving higher-fidelity image generation and understanding. The tech giant emphasizes the potential impact of models like CM3leon on boosting creativity and facilitating improved applications in the metaverse.

Meta looks forward to further exploration of the boundaries of multimodal language models, aiming to release more advanced models in the future. The introduction of CM3leon is seen as a significant step forward in the development of generative models, fueling excitement for the possibilities they hold in enhancing creativity and expanding the potential applications within the metaverse.