Deepfakes have been around for quite a while and like most other technologies in existence can fundamentally change the world as we know it by making it impossible to know what’s real or fake. Not all are convinced it is a force for good and we are on the fence.
But is deepfake tech currently good enough to make it to TV shows and movies from the biggest studios in existence? Well, the answer to that appears to be yes, at least according to Disney, which is committed to making deepfakes as realistic as possible by all means necessary.
Disney has just released a new demo this week that helps illustrate the impressive advances in deepfake technology it has made. The demo was made at a computer graphics conference and saw Disney researchers showing off what is reportedly the first photo-realistic deepfake in existence shot at what appears to be a megapixel resolution.
The Disney demo was quite impressive, at least to anyone with a working pair of eyes. That doesn’t however mean the deepfakes being showcased were good enough to make it to the next big-budget movie from Disney or other studios, but they are a decided improvement from the deepfakes of yesteryear.
It’s All In The Megapixel Count
From the results of Disney’s research, it seems likely that megapixels hold the key to making deeply convincing deepfakes that would fool experts at every level. Before now, the focus was on smooth facial transfers, instead of boosting the pixel count, but that could change now that Disney has shown the way.
You have probably seen some deepfake videos on your mobile and been impressed, but the flaws in such videos get more apparent when viewed on bigger screens than the phone you can hold in your palm. For example, using DeepFakeLab, one of the most popular open-source deepfake tools in existence, Disney researchers could only create deepfakes with a max resolution of 256 x 256 pixels. On the other hand, with their new model, they could get a maximum possible resolution of 1024 x 1024, which is a very appreciable increase indeed and would enable the creation of compelling deepfakes that can be watched on big screens.
Disney’s new deepfake model does a good job overall. It can, for example, exchange the facial structure and appearance of one individual with another while maintaining the unique facial expressions of the targeted individual.
However, the technology is still limited in reach and capability. In the video demo released by Disney, the best deepfakes turned out to be of individuals in well-lit backgrounds looking directly or almost at the camera. Poor lighting and improper camera angles tended to produce less than impressive results, but that could be remedied in the future.
Disney Forges Ahead
Sure, Disney’s new tech has limitations, but it does mark an important step and will sooner rather than later lead to the making of deepfakes so good they get used for an endless array of commercial projects. Till that time, when film companies like Disney and others want to swap faces for a project, they will have to utilize traditional VFX.
VFX is however very time and resource-consuming and is expensive as hell. Usually, producing mere seconds of content with VFX can take months of hard and expensive work.
Deepfakes are far easier and more convenient to craft. In contrast with VFX, with enough computing power at hand, and the construction of the original model, results can be expected in hours.
Disneys’ new model has the potential to relegate expensive and complicated visual effects to the trash heap. It will also enable the company and any other studio to cast any individual they want to, whether living, dead or incapacitated in any movie with near-absolute fidelity.
And that ladies and gents is a very remarkable achievement, that is nearly on par with raising the dead!