CM3leon

Discover CM3Leon, Meta's state-of-the-art multi-modal AI model for text and image generation, offering superior performance and efficiency.

AI Assistant Updated 16 minutes ago

Visit Website

CM3leon's Top Features

Retrieval-augmented, token-based, decoder-only multi-modal language model

Uses CM3 multi-modal architecture with advanced scaling and tuning

First multi-modal model with text-only adapted training recipe

Supports both text-to-image and image-to-text generation

Employs self-contained contrastive decoding methods

State-of-the-art text-to-image generation with minimal compute

Enhanced controllability for image editing, generation, segmentation

Frequently asked questions about CM3leon

CM3Leon (pronounced 'Chameleon') is a retrieval-augmented, token-based, decoder-only multi-modal language model developed by Meta AI, capable of generating and infilling both text and images with advanced editing and control features.

CM3Leon is based on the CM3 multi-modal architecture, integrating retrieval-augmented pretraining and multi-task supervised fine-tuning tailored for both image and text tasks.

CM3Leon achieves state-of-the-art performance in text-to-image generation with five times less compute than other models, boasting a zero-shot MS-COCO FID score of 4.88.

CM3Leon employs a combination of large-scale retrieval-augmented pretraining and supervised fine-tuning (SFT) on diverse instruction-style data to enhance multi-task efficiency and controllability.

Yes, CM3Leon supports both text-to-image and image-to-text generation, making it a versatile multi-modal model.

Meta AI provides advanced tools for image and video creation, personalized AI responses, Meta AI Studio for custom AI solutions, and large language models like LLaMA 4 for code generation.

Meta AI's platforms span mobile apps, desktop web, and wearable tech like Ray-Ban Meta glasses, ensuring seamless access to its AI tools.

The CM3Leon team includes Lili Yu, Bowen Shi, Ram Pasunuru, Benjamin Miller, and Yaniv Taigman, among other Meta AI researchers.

CM3Leon supports language-guided image editing, image-controlled generation, segmentation, and uses self-contained contrastive decoding for high-quality outputs.

Visit Meta AI’s website for detailed research publications, blog updates, and subscribe to their newsletter for ongoing updates.