Doctor Who missing episodes seven telesnaps and AI

This is a first attempt to see if AI can reproduce telesnaps in a useable form. It isn't taking the telesnap itself and manipulating it this is combining character references with enhanced telesnaps hoping to reproduce the composition of the original shot but with AI character models inside of it.

This is the seven best (or least worst) shots from seven telesnaps - and runner up and other interesting shots generated to get there...

Where possible I've framed the best result (immediately after the telesnap) in something approximating 4:3. For a few of them the A.I. would over scan and give me 16:9 but it would still be the best result so it went in.

The "best/least worst" are all presented (again) in 16:9 at the end of the video.

Apart from the last telesnap all the other shots were made possible by the telesnap enhancements done by Conga Line Monkey (link below is coming). They give enough information to make the results stable and repeatedly consistent otherwise AI tends to go its own way and invent a lot. The first enhancement of Polly and Kirsty shown is not one of Conga's but I did use his enhancement strongly in the overall mix to stabilise the results.

This wasn't so much about capturing the intent of the shot as there's so many things that can drag AI off course, this was purely about proof of principle that AI could replicate the shot composition near enough and also have character models within the shot at the same time that could move about a bit and make expressions. Sometimes it seems to have captured something of the intent of the shot anyway which is good, but to insist the Doctor does something specific etc would potentially eat a lot of credits. But for next time the intention of shot will be a much higher consideration and these shots can then be added into the mix as a starting point rather than as an end result. It was tight facial AI shots generated for this channel's Fury from the Deep video that really helped get this one over the line.

The other shots scattered throughout are what I've been working on since October. The single character shots for characters who do not have any existing moving footage are not tied to a specific telesnap. The Masterplan shots are the exception to the rule in this video as they are simple frame to frame and extremely easy to do, but the source material for them is so startingly good and I'd seen how powerful VIDU's new frame to frame tool is it was irresistible to put them through it... Alas though really interesting it doesn't really work with the vast majority of unrestored telesnaps as there isn't anywhere near enough information for it to go on... and that's where combining the telesnap enhancements with character references come in...

Until next time...

All shots generated in VIDU.

Music:
Titan by Scott Buckley - released under CC-BY 4.0. www.scottbuckley.com.au
Orchestra Fantasy by Lite Saturation