According to Meta, this is the best template for generating images.
Compared to other models, CM3Leon is more efficient, uses five times as much computation, and has a smaller training data set.
The majority of AI image generators also use diffusion technology, which is slow, expensive, and computationally heavy.
In addition, it relies on a different technique known as attention, which makes it less expensive to operate and faster to train and deliver results.
Reportedly, CM3Leon is capable of producing image descriptions, which opens the door for future models with even greater potential.
To improve the performance of the template, Meta uses millions of images from Shutterstock. The model now includes 7 billion operands, more than double OpenAI's DALL-E 2 model.
Meta has deployed supervised fine-tuning techniques to improve form performance in creating images, writing comments, and comments on them, allowing users to ask questions and make changes to images using text commands.
CM3Leon applications include virtual reality, augmented reality, robotics, and advertising and marketing.
It is not clear when the CM3Leon model will be released by Meta, and Meta is expected to provide more information regarding the model in the future.
Comments
Post a Comment