Meta continues the AI revolution as it research team unveiled a new text-to-image creation tool called CM3leon.
The CM3leon — pronounced as “chameleon” — is a text-to-image and image-to-text creation tool, which means that it can also easily generate captions for images.
Related
Its text-to-image function is similar to other tools already available in the market: DALL-E, Stable Diffusion, and Midjourney. But what makes it special is, as per Meta, the techniques the Facebook owner used and the performance the foundation model can achieve.
Instead of using diffusion models like others do, CM3leon utilizes a token-based autoregressive model.
Comparing it to diffusion models, Meta’s research team said that: “In contrast, token-based autoregressive models are known to also produce strong results, with even better global image coherence in particular, but are much more expensive to train and use for inference.”
Basically, the Meta researchers found out that the token-based autoregressive model can be more efficient than the diffusion model-based approach.
You can read the full research here.