Animation and AI

[johngrundy]

Jul 9, 2024 21:07:19 GMT Natalie Sinead said:

Moffat has a point about AI not being able to truly replicate Troughton's skills and comic timing.

Where can I find this interview with Steven Moffat, please, Natalie? I would be interested to read it if he has things to say about AI.

[Stuart Douglas]

Jul 9, 2024 19:00:20 GMT johngrundy said:

This is a tricky one with Star Wars here. Any film (Rogue One) and series (Andor Season 2) that is set around the Death Star being constructed and tested out on a planet (i.e. blow it up) must feature Grand Moff Tarkin, since he plays a large part in these events. To have any actor other than 'the late great Peter Cushing' recreating the role might seem disrepectful. Surely an accurate-as-possible AI version - with the blessing of his estate - is preferable to a living actor who looks nothing like him doing an imitation performance?

Surely both AI and another actor are exactly as unlike him as each other, given neither is him in any way?

It's a role in a made-up world - if people feel the need to revisit that role after the original actor is dead, recast it.

[johngrundy]

Jul 10, 2024 7:46:26 GMT Stuart Douglas said:

Jul 9, 2024 19:00:20 GMT johngrundy said:

This is a tricky one with Star Wars here. Any film (Rogue One) and series (Andor Season 2) that is set around the Death Star being constructed and tested out on a planet (i.e. blow it up) must feature Grand Moff Tarkin, since he plays a large part in these events. To have any actor other than 'the late great Peter Cushing' recreating the role might seem disrepectful. Surely an accurate-as-possible AI version - with the blessing of his estate - is preferable to a living actor who looks nothing like him doing an imitation performance?

Surely both AI and another actor are exactly as unlike him as each other, given neither is him in any way?

It's a role in a made-up world - if people feel the need to revisit that role after the original actor is dead, recast it.

I think Peter Cushing made the Grand Moff Tarkin role his own. It's hard to imagine any recast actor playing it. An AI lookalike coupled with a voicealike works better for me than recasting. The 'spirit' of PC just wasn't there in the recast role in Revenge of the Sith like it was in the AI version in Rogue One.

[George D]

The key to me is respect.
Respect of the actors reputation.
Respect for the actors memory.

To use an actor in a piece that one would logically think he would not approve of, is desecration of his memory and we, as consumers, should not support it. the usage of hartnell in some new who I think used him to promote a value system that he would not condone

However keeping someone's memory alive in a dignified fashion is one of the greatest legacy a man could have.. that people, after his death would honor him. And appreciate him and want more of him. Not just so a trademark holder could make millions.

Hopefully simple respect protects people's legacy.

[Ken Jacowitz]

At first I thought returned missing episodes.

But reconstructions.

Ian Levine just announced on twitter:

In the next few days expect...
FURY FROM THE DEEP EPISODE FIVE
MARCO POLO EPISODE TWO
THE REIGN OF TERROR EPISODE FOUR
DALEKS MASTER PLAN EPISODE ONE -
REMADE
THE SPACE PIRATES EPISODE FIVE.

[stevehoare61]

Having seen a couple of episodes its all good news, but despite many attempts to offer contributions etc, I dont seem to be able see any other or help finance....so while its great news, isnt going to make much difference to me sadly.

[Steve Hamilton]

I have seen some of Ian's AI reconstructions. I think they are very good.

I particularly liked "The Massacre" and honestly could have been watching recovered footage. I think there are several teams working on individual stories so the style does vary between them. Perfect? No, but this is a vast improvement on static telesnap reconstructions and can be improved upon as techniques develop.

Indeed, Ian has already 're-made' episodes of "The Dalek Masterplan" and "The Savages" for example in order to better them.

I would prefer this sort of reconstruction to the released animations (at least the recent ones). I think this has a great future for completing lost episodes. My thoughts anyway.

[jcoleman]

Perhaps this thread should be merged with the "Animation and AI" thread?

I too have seen a few of these AI recons. The biggest issue with them is Ian and others exaggerating the results, which has opened both the recons and Ian himself up to a lot of what would otherwise be unwarranted criticism.

These recons are nowhere close to convincing as realistic replacements for missing episodes and the quality is not remotely suitable for a commercial release. They do however provide an interesting indication of how technology is advancing and provide a glimpse into what might be possible in the not too distant future. I for one appreciate the endeavour if not necessarily the outcome.

There are a number of glaring issues with the current AI-based recons. Ian has said that it is going to be an ongoing iterative process, so it's good to see some of the episodes already being redone. DMP1 was one of those that was shared more widely than just Facebook. I wonder if the second version will similarly be made available. It would be interesting to see how they compare.

Ian is obviously determined to get these recons made for all the missing episodes. It's very much a case of quantity over quality. Given Ian's age and health that's understandable. If these recons are bringing him and others joy then that's great.

As far as the recons go, there are occasionally shots that look pretty good but they are few and far between. Personally I would prefer a focus on quality. I'd rather see 25 seconds of convincing footage than a 25 minute episode of extremely variable quality. As the technology improves, along with the skill of the recon makers, more will be possible. This is just my opinion however; I don't have the knowledge and skills nor I am the one stumping up the money to produce these recons.

The single biggest issue with these recons seems to me to be the over-processing of the still images upon which much of the footage is based. Attempts have been made to upscale them but the result is the complete removal of detail from the actors' faces giving them a bizarre unnatural smoothness and sending them straight into the uncanny valley even before the movement has been generated. There are numerous occasions when the actor/character likenesses are lost completely. One would think this would be a relative simple issue to address.

The lip-syncing appears to be getting better, but there is still little if any expression of emotion. Wider movement remains a significant challenge too, which is as to be expected. There are lots of examples where everything appears to be slowed down and/or run forwards and back on a loop both of which are very unnatural and distracting.

The most effective shots are those where surviving footage has been used with characters' mouth movements changed to match the dialogue being spoken.

There's a long way to go, but every journey has to start somewhere. If these recons were not being spoken about or promoted as being "just like the real thing" and were seen as another step towards what one day being possible I believe they'd be viewed in a much more positive light.

[George D]

One thing I give Ian credit for. He noticed the ai opportunity, and in spite of nay sayers, created not just an episode, but complete stories and at a pace faster than both classical reconstructions and traditional animation.

He opened our eyes to what is possible.

I view them as in the animation class. And just like regular animation, there are times of irregular movements, characters not looking right, and artistic license. Also, the choice between higher quality clips compared to using the more authentic lower quality telesnaps can be debated. Just like animation, one can be careless and not have scenes created accurately or with errors.


So far, it appears scenes in the toymaker recon have used ai, and I think it's impressive when used correctly.

I just hope the cognitive and emotional biases and obstacles are able to be overcome and not a bias for future endeavors. So they can work together to utilize this technology even better in the future.

[nathangeorge]

The TARDIS Cam was once considered impressive technology

[Natalie Sinead]

Tech only marches on.

But then, Bigon's words to Tegan Jovanka....