A lot of the most intriguing updates coming to the AI space are coming from researchers based at Google. One of the most recent reveals that they have made is Vlogger, an AI that has the ability to turn still images into videos with all things having been considered and taken into account. Using cutting edge machine learning models, Vlogger can create gestures and movement from just a single image, and what’s shocking is just how realistic those videos might look.
With all of that having been said and now out of the way, it is important to note that there will be some artifacts here or there that would reveal the AI origins of a given video. In spite of the fact that this is the case, if you were to add an audio with the image, the person in that image would match the words and gestures that correspond to the audio with a high degree of accuracy.
Enric Corona is the researcher that headed the team behind this impressive new machine learning tool, and he’s specified that diffusion models are the specific type of machine learning algorithms they used. They’ve previously been used for the purposes of turning text prompts into detailed images. An enormous new set of data was input into the models thereby allowing them to create video from still images.
This dataset is called MENTOR, and it contains 2,200 hours of video along with 800,000 distinct identities. As a result of the fact that this is the case, people from a wide range of ethnicities can be synthesized with the use of Vlogger.
What sets this new tool apart is that it doesn’t require separate training for each individual image. It doesn’t just focus on the lips either, with the entire individual being synthesized to create a more believable end result.
Such a tool might be useful because of the fact that this is the sort of thing that could potentially end up taking the content creation industry several steps forward. More progress will be required before it reaches its final form, of course.
Read next: Breaking Boundaries: Elon Musk's Grok AI Embraces Open Source, Redefines Chatbot Standards
With all of that having been said and now out of the way, it is important to note that there will be some artifacts here or there that would reveal the AI origins of a given video. In spite of the fact that this is the case, if you were to add an audio with the image, the person in that image would match the words and gestures that correspond to the audio with a high degree of accuracy.
Enric Corona is the researcher that headed the team behind this impressive new machine learning tool, and he’s specified that diffusion models are the specific type of machine learning algorithms they used. They’ve previously been used for the purposes of turning text prompts into detailed images. An enormous new set of data was input into the models thereby allowing them to create video from still images.
This dataset is called MENTOR, and it contains 2,200 hours of video along with 800,000 distinct identities. As a result of the fact that this is the case, people from a wide range of ethnicities can be synthesized with the use of Vlogger.
What sets this new tool apart is that it doesn’t require separate training for each individual image. It doesn’t just focus on the lips either, with the entire individual being synthesized to create a more believable end result.
Such a tool might be useful because of the fact that this is the sort of thing that could potentially end up taking the content creation industry several steps forward. More progress will be required before it reaches its final form, of course.
Read next: Breaking Boundaries: Elon Musk's Grok AI Embraces Open Source, Redefines Chatbot Standards
