Facebook says its new artificial intelligence can mimic your writing just by seeing a word
After the deepfakes from videos, voices And till satellite images, it is the turn of writing. A new Facebook project manages to imitate most of the writing just by reading a word. From this word, artificial intelligence learns what typography is like and is in charge of writing any text with it.
TextStyleBrush is the name given to the new Facebook AI project. It is an artificial intelligence that can copy the style of a text finding in a photograph. What undoubtedly makes it impressive compared to other similar artificial intelligences is the small amount of data it requires for this: just a few letters from a word. That is, a theoretically could write his name and the AI would know how to imitate his writing.
Facebook explains that while most AI systems can do this with specific tasks, in this case TextStyleBrush is flexible enough to understand all kinds of real-world fonts. That is, it understands without apparent problems texts with all kinds of calligraphies, applied styles, thicknesses, colors, shading or rotations.
They indicate that to achieve this their artificial intelligence tries to understand the typography you analyze from a fucking more holistic point of view and taking into account how it is in its fullness. Generally, AIs that mimic writing look for specific parameters such as serifs or thickness, not in this case, which tries to understand writing more as a set.
The real utility of an AI copying your writing
As is often the case with these kinds of advances in artificial intelligence, it can have both positive and potentially negative uses. In theory an AI like this can be used to fake someone’s handwriting and “speak on their behalf”, for example. But also for positive applications. For example, you can used to instantly translate real-world texts just by photographing or pointing the camera.
Either way, at the moment it’s just a public Facebook investigation. We will see if in the next iterations improvement can become popular when applied to a real app or tool. They say that there are still limitations and concrete points to overcome, for example when the AI has to deal with metallic texts and reflections.
More information | Facebook