Hacker News Reader

Comment by gcanyon | Hacker News Reader

Anyone saying:

   - it’s only short videos
   - it’s not fully consistent
   - it is hard to direct to get what you want

…is looking at where it is now, and failing to consider where it will be in literally a year or two.

Five years from now parents will be mashing up Winnie the Pooh and Avatar, starring their children, in immersive 3D, for the Apple Vision Pro 4 or the Quest 7.

Tyler Perry is making a wise choice.

View on HN

>Five years from now parents will be mashing up Winnie the Pooh and Avatar, starring their children, in immersive 3D, for the Apple Vision Pro 4 or the Quest 7.

five years MAYBE expert creators will be able to do this in a satisfactory way after a lot of polish in editing out the edges and hundreds to thousands of prompts. And even then we're talking more about 5-10 minute short, not a feature length film.

five years for a parent with no technical/artistic expertise, sure. some will make some 10 second shorts on par with a current college freshman's early film project. minus understandings of the fundamentals of film, of course.

I really don't know how to reply to this.

Five years ago, there was nothing in existence that could respond well to a prompt like "Write a short story about my six-year-old daughter Emily traveling undersea to rescue her favorite sea horse, who's been kidnapped by the Shrimp King. Make her have several adventures along the way, engaging in side quests and making friends who ultimately help her succeed in her quest"

Today ChatGPT responds to the above with no difficulty, and part of the reason is that it has been trained not just to output reasonable responses, but aesthetically pleasing ones.

There is every reason to believe that video output will be more responsive five years from now than text is right now. Sure, it's possible it will be a little stilted, or a little generic, but it's possible it won't be.

Obviously we're both speculating, but I find it hard to believe that one minute today won't turn into hour-length in five years.

Well there are two issues right now:

1. Text has come car, but I don't think anything other than blogspam sites have heavily leveraged AI as a commercial product. There's a huge gulf of difference between "this looks okay enough that the common user won't question it" and "this is helping professionals make money for products consumers the self seek out". AI is 90% of the way there, and it has another 90% to go, just like any human.

2. With all that noted, Video is a lot more computationally expensive than text generation. Which is multiple magnitudes of more information to edit/refine. It will only magnify the problem since the common user is a lot better at recognizing defects in visuals than in text (where native speakers famously don't process every word while reading).

I'm not saying that this won't massively improve efficiency. But I think we both will agree that this tech won't lead to a large studio making content in months so much as letting a very lean team make content in 3-5 years for much less money when they lay off 95% of the staff.

The only parts of movies or TV that this will be useful for is non-character-driven filler that doesn't move the plot along. It will be great for things like action scenes and car chases, or staring wistfully at the ocean. But they still need actors and a story, and to tie those together seamlessly with whatever is generated. We're very far away from doing that in a way that will look or feel good.

Special effects are nothing new. But we all know special effects don't make a good movie.

You aren't thinking about generative voice AI or LLMs. Putting it all together, it is not surprising that creatives are afraid they will be done in under 5 years.

Very true, today, but how far are we from something that takes a book as script and spits out a full movie? We have a huge library of ultra cheap forgotten novels that will make excellent prompts for killer personal productions once technology allows it. Who will ever go watch a movie if they can upload the book to their server and get the movie in minutes?

Think more about the backgrounds used in shows like Mandalorian. Think about the size of the stage/studio/set that Mandalorian filmed on. It's going to get smaller, and the backgrounds and other details used in music videos, tv shows, commercials and films will be far more virtual than even today.

All the components will still be needed, but the levels and origins of them all will change.

If you go back and look at how the median productions have changed at 5 year intervals (going in the past), it'd be a poor choice to commit $800M to something that's probably going to remain exceedingly fluid for a while.

The SFX industry is probably the most prepared to shift, as it's already screens, pixels, and servers. The days of green dots are already (mostly) over (see the ILM SFX reel for The Creator -- wholesale human detail replacements without tracking dots).

The change will affect everyone else more.

Reply on news.ycombinator.com