Just when you thought that nothing could be completely surprising ever again, how about this?

The new IDLES video is... the iconic video for Coldplay's first hit Yellow, directed by James Frost and Alex Smith, reworked so that the fresh-faced Chris Martin, is lipsyncing the rock band's track Grace (from just released new album Tangk). And it's flawless. It's so convincing it defies belief... well, at the moment at least. 

Helmed by director Jonathan Irwin, this is a feat of visual deception which is so real and believable its genuinely eery. It's a next level Deepfake, a new depth reached in the Uncanny Valley. And it's not only mindblowing in the way its been executed. Its the sheer unexpectedness of it. 

Not surprisingly AI was involved, and also lots of other VFX techniques, to make it happen. But at its core, there is a surprising creative collaboration going on here, primarily between IDLES' Joe Talbot and Coldplay's Chris Martin, that no-one saw coming.

Joe Talbot came up with the concept - in a dream, allegedly, which seems entirely plausible - of using the classic Yellow video, directed by James Frost and Alex Smith back in 2000, to frame the song Grace. The process by which this outlandish idea became possible is currently a mystery to Promonews. But at some point, rather wonderfully, Chris Martin then agreed that this was a splendid idea.

After that, the process began when production studio Joyrider were approached by IDLES commissioner John Moule, with the question: can you make this happen? But the fact that nothing like this had been done before meant that the necessarily level of perfection could only be achieved by a combination of techniques and a good deal of trial and error.

It also required Chris Martin's further involvement beyond giving his go-ahead, as the filmed him in order to provide a mouth reference to 'train' the AI technology. But that was later on, after Jonathan Irwin had tested various AI tech solutions to land on the best route forward. Tests on the mouth, face and full head replacement tests were trialed using AI tools such as wac2lip. But at that stage it did not deliver the required consistency and quality.

The team moved to an approach that used a filmed mouth singing the song, which the AI Deepfake training then used, before this pre-trained face went through around one million iterations to get to a very decent level. To sync the mouth seamlessly with the original video Joyrider approached VFX specialists Stone Dogs to harness their flame compositing expertise. In a cyclic turn of events, Brian Carbin, senior VFX artist and co-founder of Stone Dogs, had worked on the original Yellow video.

Then the team were able to take the production to the next level and film Chris Martin himself in the band’s studio, singing the song at double speed (50fps) with three 4K cameras. The video was then played back and slowed down, adding an additional layer of subtlety to match the original plate. The filmed material was then Deepfaked further, helping to deliver a convincing AI performance of Chris Martin, filmed as if it actually happened in 2000.

Jonathan Irwin led weeks of further research and development, constantly refining machine-learning approaches. “To create the training set, we fed every frame of the original video into DeepFaceLab and ran for five million iterations, approximately one and a half months of processing time, running 24/7," he explains. "At each stage of improvement, there would be backwards and forwards of new AI training runs and flame VFX work, tracking, extensions, clean up, more tracking, test composites, more AI, more clean up, etc.”

Beyond the demands of the lipsyncing, other factors had to be addressed, such as the difference in the length of the tracks while keeping the integrity of the original video, which was filmed in one take (and ironically, was a last minute option taken by Frost and Smith, after bad weather had curtailed their original idea for the video). A time/speed ramp effect was applied to a section of the film to bring the two versions in sync, and additional time warp effects were positioned over certain points of the video, to take the viewer away from the original video and into an IDLES-esque world.

And as a result, we have something never seen before, and something that was frankly unimaginable, not too long ago. But crucially this is not using amazing technology for its own sake. At its heart, it's just a gobsmacking idea.