|
Post by Richard Tipple on May 31, 2019 8:48:42 GMT
Because it's only going to be a short clip, and probably not very convincing, and people are going to be trying to rip people off by trying to convince them it's an excerpt from the complete episode which they'll try and convince people they're hoarding. Those fakes are convincing enough when it's someone standing still and talking. You could do it with a telesnap. Once you try and make things move a bit more, a few pans, a bit of action, it's not going to look very convincing because you have no information for that part and it's a stretch too far for the software to extrapolate something it has no information for, it's going to be all too obvious it's a fake. The Hartnell Bradley deep fake is amazing and shows what could be achieved - tinyurl.com/HartnellBradley. There are calls for them to try Hartnell over Hurndall in The Five Doctors. With a bit more sophistication this could be a nice dvd extra. I'm currently working with this guy to produce better results and have sourced a lot of source material which should improve the process significantly. For example, there's a scene from 'Double Confession' which shows Hartnell in various mid-shots, with a wide array of facial expressions and, most importantly left/right profile shots.
|
|
RWels
Member
Posts: 2,857
|
Post by RWels on May 31, 2019 9:12:19 GMT
Because it's only going to be a short clip, and probably not very convincing, and people are going to be trying to rip people off by trying to convince them it's an excerpt from the complete episode which they'll try and convince people they're hoarding. Those fakes are convincing enough when it's someone standing still and talking. You could do it with a telesnap. Once you try and make things move a bit more, a few pans, a bit of action, it's not going to look very convincing because you have no information for that part and it's a stretch too far for the software to extrapolate something it has no information for, it's going to be all too obvious it's a fake. The Hartnell Bradley deep fake is amazing and shows what could be achieved - tinyurl.com/HartnellBradley. There are calls for them to try Hartnell over Hurndall in The Five Doctors. With a bit more sophistication this could be a nice dvd extra. No, I don't call that amazing. I'm sure it's miraculous that this is now within the realm of possibility, as they say. But at the same time it's just a face transplant. And not to put too fine a point on it, it's still quite obvious. Issues: 1. This is only possible because the rest is already there. A condition that can't be met for missing episodes without really extensive efforts. 2. The assumption is that a face is all that an actor is. What about voice, movements and gestures and other body language, etcetera?
|
|
|
Post by Dan S on Jun 2, 2019 1:29:33 GMT
Issues: 1. This is only possible because the rest is already there. A condition that can't be met for missing episodes without really extensive efforts. Yes, for the missing episodes all we have to represent most episodes is around 60 (at best, for some it's much less) quite low quality photos, and the audio. I'd like to see Deep Fakes try to extrapolate an entire episode from a handful of photos - it couldn't be done. Maybe one way would be to re-film the entire episode and then transplant the faces afterwards. Somehow I don't think it'd look authentic.
|
|
|
Post by scotttelfer on Jun 2, 2019 11:48:27 GMT
Issues: 1. This is only possible because the rest is already there. A condition that can't be met for missing episodes without really extensive efforts. Yes, for the missing episodes all we have to represent most episodes is around 60 (at best, for some it's much less) quite low quality photos, and the audio. I'd like to see Deep Fakes try to extrapolate an entire episode from a handful of photos - it couldn't be done. Maybe one way would be to re-film the entire episode and then transplant the faces afterwards. Somehow I don't think it'd look authentic. Just to be clear, Deep Fake refers to the "face transplant", it isn't going to produce new footage.
|
|
|
Post by Dan S on Jun 3, 2019 0:57:51 GMT
Just to be clear, Deep Fake refers to the "face transplant", it isn't going to produce new footage. Oh, I didn't know that. It's never going to be able to re-create any lost Dr Who then, making this thread utterly pointless. I suppose you could use a telesnap and make the people in it talk, but if they're rooted to the spot in that one location it's not much use apart from novelty value.
|
|
|
Post by Alan Hayes on Jun 3, 2019 11:20:59 GMT
The Hartnell Bradley deep fake is amazing and shows what could be achieved - tinyurl.com/HartnellBradley. There are calls for them to try Hartnell over Hurndall in The Five Doctors. With a bit more sophistication this could be a nice dvd extra. While I'm not claiming I could produce something a tenth as convincing, that's not "amazing". It's terribly artificial looking, soft and lacking in detail compared to the quality of the surrounding footage and most of the time it doesn't even look like Hartnell. Full marks for trying it (and I'm sure the technology and techniques will improve), but if that's what can be achieved currently with Doctor Who, it's going to be no more than a novelty. EDIT: Sorry - hadn't seen the previous responses.
|
|
|
Post by scotttelfer on Jun 3, 2019 12:31:57 GMT
Just to be clear, Deep Fake refers to the "face transplant", it isn't going to produce new footage. Oh, I didn't know that. It's never going to be able to re-create any lost Dr Who then, making this thread utterly pointless. I suppose you could use a telesnap and make the people in it talk, but if they're rooted to the spot in that one location it's not much use apart from novelty value. The idea would be to have body doubles act out the scene and then Deep Fake the original actors onto their faces. There is still a hypothetical use, but the software is limited by its purpose.
|
|
|
Post by jamesvincent on Jun 3, 2019 20:40:19 GMT
There is hopefully a good chance Josh Snares is going to produce something soon to show just how cool it could be.
|
|
|
Post by jcoleman on Jun 3, 2019 23:58:47 GMT
Oh, I didn't know that. It's never going to be able to re-create any lost Dr Who then, making this thread utterly pointless. I suppose you could use a telesnap and make the people in it talk, but if they're rooted to the spot in that one location it's not much use apart from novelty value. The idea would be to have body doubles act out the scene and then Deep Fake the original actors onto their faces. There is still a hypothetical use, but the software is limited by its purpose. The Bradley/Hartnell Deep Fake video doesn't work very well at all. However, if you take a look at the Star Trek Deep Fake videos made by the same person then you can really see what the potential is. These are done using footage from episodes of Star Trek Continues, a fan-made continuation of the original Star Trek series which is available on YouTube. Whilst the results remain variable, the Vic Mignogna/William Shatner Captain Kirk Deep Fake in particular is really quite remarkable. Much of the time if it wasn't for the voice you could easily believe it was Shatner. The recent Samsung AI demonstration appears to enable still images of an individual to be input (with more images leading to better results - the demonstration went up to 32 images) and a head and shoulders image of the individual to then be animated in a convincing fashion. In theory, unless I have completely misunderstood, this could allow a segment or segments of surviving footage of say Hartnell to be used to teach the AI how his face moves, allowing an 'actor' to then lip-synch dialogue from a missing episode to animate an image of Hartnell. Obviously there are ethical concerns about these new software developments, but from a recreation of missing episodes standpoint they do have extraordinary potential. At the very least they would provide another tool in the toolbox for photo-realistic reconstructions. As technology stands, with CGI, model work, rotoscoping, body doubles for shots where the character's face isn't seen, clever re-use of surviving footage etc. (all of which have been used in efforts to reconstruct missing footage already) it is technically possible to recreate pretty much everything already apart from the actors speaking... which of course is what the bulk of any episode consists of. The likes of Deep Fake and Samsung AI open the door to this final hurdle one day being overcome. I can't imagine it ever being done commercially by the BBC, but as an evolution of fan-made efforts maybe. In lieu of more footage ever being found, which seems increasingly unlikely the more time passes, I for one would enjoy seeing say a convincingly animated CGI Dalek having a conversation with a Deep Fake/Samsung AI rendering of the Doctor in The Daleks' Master Plan or The Evil of the Daleks!
|
|
|
Post by Ed Brown on Jun 4, 2019 8:29:29 GMT
It's one thing to map Hartnell's face over Bradley's, where "all" the software has to do is alter a small area of the image, and has cues to guide it, from the positioning of Bradley's eyes, nose and lips.
It would be far beyond the capabilities of this software to generate a scene from nothing. All it does is manipulate a few pixels in an existing image, and track the movement across the frame of a small data area containing those pixels.
Admittedly, it's a remarkable achievement. But it has very rigid limitations, and, as it stands currently, it can't even fake Hartnell's voice, so you still hear Bradley. Now, admittedly, I could easily rebuild Hartnell's dialogue, in a missing episode, from an audio recording made off air in the Sixties. Maybe I could even use some new software I'm familiar with to fake Hartnell's dialogue for 'The Five Doctors', to replace the existing soundtrack.
But we are still decades away from any software which could generate an entire scene from a missing episode. The animated reconstructions prove just how much of a pipedream it is to hope for a realistic presentation. No current software can replicate the 50 years of acting experience which went into William Hartnell's screen performance in those missing episodes.
|
|
|
Post by Natalie Sinead on Jun 4, 2019 8:33:38 GMT
I think, going forward, more actors will follow Robin Williams's example and put serious restrictions on the use of their image and voice after their death. I mean, would Peter Cushing have *really* wanted to star in a film decades after his death?
|
|
|
Post by Ed Brown on Jun 4, 2019 8:43:52 GMT
Before anyone gets too carried away, I'd like to point out that I'm a big fan of 'Star Trek Continues'. The acting performance given by Vic in the part of Kirk is truly astonishing, he actually almost seems to be Shatner in some scenes -- helped, undoubtedly, by the fact that in real life he genuinely resembles Shatner (his hair, his build, and in his physical movement). The faker software is not what is convincing in the test footage, that footage is convincing because of Vic's acting ability. The face-transplant would be far less convincing if the actor playing Shatner was not doing a really convincing impersonation of Shatner to start with. If you didn't have Vic, using the software to rebuild the face of J Fred Muggs in that role would be only marginally effective.
|
|
Sofia Fox
Member
Favourite Missing Stories: Marco Polo, The Crusade, The Myth Makers, The Macra Terror, and Evil.
Posts: 75
|
Post by Sofia Fox on Jun 5, 2019 10:54:04 GMT
Been reading this thread for a bit. I say just leave deepfake alone. It's already ruined people's lives and I'd rather not that happen.
|
|
|
Post by michaeltetzlaff on Jun 5, 2019 13:25:05 GMT
Instead of a "deep fake" where a surrogate actor is required, what missing Doctor Who needs is really this: news.developer.nvidia.com/ai-learns-to-lip-sync-from-audio-clips/It's a couple of years old; I don't think there's a publicly available release, but it basically uses reference video of a particular person and then synthesizes a video that is lip synced to a different voice track while preserving the facial expressions, mannerisms, etc. from the reference. Seems perfect for missing Doctor Who where you have plenty of surviving footage of the main actors and an audio track for it to be synced to. It probably wouldn't be enough for a full episode that looks like the original thing, but would at least be good enough for an "enhanced reconstruction," combined with telesnaps and CGI for sets, props, and monsters. Unlike "deep fake," the video is actually being driven by the original actor's performance, so it's arguably the most respectful way you could do something like this.
|
|
|
Post by Natalie Sinead on Jun 5, 2019 13:31:46 GMT
I still think that as we move forward in time, important (and other) actors will do a Robin Williams. They won't want to end up like Fred Astaire and vacuum cleaners or Cary Grant and Diet Coca-Cola.
|
|