Do you have a dumb question that you're kind of embarrassed to ask in the main thread? Is there something you're just not sure about?
This is your opportunity to ask questions. No question too simple or too silly.
Culture war topics are accepted, and proposals for a better intro post are appreciated.
Jump in the discussion.
No email address required.
Notes -
Posting less as a question and mostly for self-accountability.
I made a prediction that we might see a feature length film produced by a small team using AI by the end of this year.
Well, the year has ended and I can't find any such releases that have been made publicly available. So comfortable saying my specific call is a bust.
But.
In the 11th hour, one of the creators (Gossip Goblin) I've been tracking since like July published something that at least validates my logic.
Woodnuts
If it were 80 minutes instead of 10, I'd argue it adequately fulfills prediction. Instead, I'll just argue that it proves my point that in principle a small team could have built out a feature film, insofar as its just a matter of repeating the efforts that produced the original 10 minutes to add to the length.
It avoids the standard AI 'tells.' The character's appearances are consistent throughout. There's no weird physics or physical deformities (that aren't intentional), the SFX quality is arguably a step above modern CGI in many cases (Avatar movies notwithstanding). There are some truly impressive cinematic shots in there.
Now the main hints are the short length of the individual shots, the lack of 'action' scenes to speak of, the general surreality of the environment, and the fact that they relied on narration rather than characters actually speaking dialogue. Don't think that dialogue isn't mostly solved, though.
The previous top contender was Kira (still extremely impressive on its own).
So I'm still betting on us seeing that first feature-lengther in fairly short order. And not TOO far after that, the ability to produce feature-length films from a single, fairly-detailed prompt.
Anyone else have a guess as to when such a film drops? (again, I don't say it has to be released on streaming or broadly viewed, just that it has to exist and be released in a publicly-reachable way)
Bonus Question:
When will we see an existing movie completely reworked via AI? Or perhaps just a couple of characters recast.
P.S. My other longer term prediction about AI replacing newly minted lawyers is still in play, and I did get some validation on that one.
If nothing else (and that's an if that won't hold); AI is to CGI as CGI was to stop-motion (and many other practical effects). CGI is soon to be over as the state of the art way to produce special effects. It will be reduced tremendously in it's purpose
I think that is a correct analogy.
My guess is that there might be an opening where very low-fidelity renderings are used to map out the action on screen, but AI is doing the work of dozens of other animators in texturing, lighting, simulating and 'rendering' the actual image on screen, with a human just nudging it along and rejecting outputs as they go.
The missing step seems to be fine-grained control over the details, but creators like Gossip Goblin have been able to keep an extremely consistent style, so either that's a solved problem or they've got their prompts refined to a point that they aren't having to toss out much.
The quality available at what has to be a fraction of the cost of traditional FX is going to lead to rapid uptake.
Something very much like this will be a near certainty because trying to prompt detailed poses, positions, proportions, movement paths and so on is a fool's errand. Pure written language is a horrible inefficient way to do such things while a 3D modeler uses an interface optimized for that and provides realtime feedback to the user.
To a large extent, these tools already exist. They're just limited: SCAIL struggles for movement paths with more than three characters or over nine seconds, ControlNet Pose has to be tuned for each model and sometimes even each finetune, and LoRA can uniquely handle three or four style/character/event/motion per output before they start getting funky interactions.
But even assuming that these problems can be fixed - plausible, but not a given! - there's a fundamental tradeoff between what you let the model do, and what you don't. Sometimes expressed as a double! And still hard to manage.
More options
Context Copy link
I mean if we actually get human-level AI in the picture, isn't this pretty much how traditional animation is done? Some storyboards plus a bunch of pure written language?
More options
Context Copy link
Yep. Unless you can hook the thing straight up to the animator's brain (hi there, Neuralink!) the fidgety little details will be hard to keep perfect and consistent, let alone going back and making minute changes without 'redoing' the whole shebang.
It still might beat having to go in and do all the detailed work manually, bur I know way to little about digital animation to give a real guess.
I note that this isn't all that different from standard live-action filmmaking, where you would have actors give multiple 'takes' on a scene and edit in the best ones. You're still 'prompting' actors, and refining your instructions based on the 'output' they produce, then choosing which ones you like and discarding the rest.
In fact, that might be the way to think of it, a return from the sheer tedious craftmanship of computer animation to the more 'organic' style of a Director/Prompter eliciting their ideal performance and massaging it into the final product.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link