Blog #34 The Consequences of A.I. Generative Video
Recently, Sora 2 was announced and it got a lot of attention. According to OpenAI, Sora 2 is: “…our flagship video and audio generation model.“
These are the same guys behind ChatGPT, so you know it’s serious.
I’ve seen the demo video they posted: a series of clips of their CEO—Sam Altman—in a wide range of ridiculous scenarios. One thing was striking—they all looked really good. Is realistic AI video here? Well…it’s not far.
There is still some fine tuning to do but the technology is capable of creating incredible imagery, that easily passes for realistic footage.
Check this still out. It’s from a video they generated in-house, link here (https://openai.com/index/sora-2/):
Prompt: in the style of a japanese anime, a melancholy scene under the fireworks of a night sky. the world is so happy, but not the two star-crossed protagonists of this gorgeous japanese town in the middle of a festival. film-caliber sakuga japanese animation, close-up shots of the characters having a conversation in japanese, beautiful fluid hand-drawn animation
This capability raises a lot of questions, especially from a safety standpoint. Here is OpenAI’s policy on deepfakes:
“Preventing deception and knowing the source
Deep Fakes, or unauthorized AI-generated recreations, involve deliberate misrepresentation of people and ideas, and are unacceptable.
Verifying Authenticity: We partner with the Coalition for Content Provenance and Authenticity (C2PA) Steering Committee to help people know where images came from and what made them. This includes adding C2PA metadata to all images created and edited by DALL•E 3 in ChatGPT and OpenAI API. We are also testing the DALL·E 3 Detection Classifier to predict likelihood if an image originated with our AI.
Preset Voices: ChatGPT is designed to speak in only one of our preset voices.
Creative Controls: DALL·E 3 is designed to refuse image requests for a public figure by name.“
My take away from this is: if you are looking for the source of a video you suspect to be deepfake, open up the metadata of the file. The information you are looking for may be in there. Of course, it may have already been deleted, I doubt it is hard to do so…There are ways around this, no doubt.
Will someone think of the children?
Children can be cruel and unpredictable. What they do with a technology like this, cannot be predicted. However, they may use it to bully each other in new and creative ways. They could create videos of other kids in compromising situations and use it to spread rumours.
There are also other ways it can be used, as pointed out by OpenAI themselves. In their demo, they created a video of Sam Altman on CCTV robbing a grocery store. Will criminals in the present/future use AI video to recuse themselves of crimes they commited? Or how about crimes they didn’t commit? Could you create an AI video in order to frame someone? Or both? The possibilities are endless.
My workflow
This may surprise you but I don’t generate AI video into my workflow at all. There are a few reasons for this:
1. Time: generating videos, from my experience, takes a lot of time. It is not a time-efficient use of my workflow and thus I would rather stick to A-roll/B-roll, instead of trying to use AI video.
2. The look: AI video tends to have a “fake“ look, which makes the video seem low-quality/low-effort. This is something I want to avoid because I put a lot of work into the content I make and I wouldn’t want AI video to take away from that.
3. The cost: From my experience, creating AI video costs. If you use free sites, you get what you pay for—very little. If you want to generate video in a reasonable timeframe, you have to pay. To me, it’s not worth it.
4. Enjoyment: I don’t enjoy it. AI is never fun to use in my video workflow—it’s clunky, ugly and fake. It takes away the enjoyment of crafting together a project from scratch. Sure, I use stock footage and fair-use clips a lot, but there’s a big difference between those and just generating a video from prompts.
Will this change in the future? Undoubtedly so. If I could have a free, quick AI video generator that creates quality imagery, I would use it in a heartbeat. It would improve the product and my workflow, allowing me to focus on other areas.
Then we will enter into another world: Real vs. fake
What is real? What is fake? If this technology becomes indistinguishable from reality, will it change anything? Yes. I believe, once it’s good enough, people will watch what they enjoy most—whether AI or real. We’ve seen AI live-action trailers for The Simpsons (1989) blow up on Youtube. It shows that there is a huge market for such a product. And…if studios won’t give it to us, who will? Some AI artist from outside the US will. They’ll do it again and again because if you build it, people will come. There’s so much potential and still so much yet to come. It’s exciting, frightening and a little crazy.
Those are my thoughts. Hope you found something useful in here.
Thanks for reading and if you’ve any comments, let me know down below.
Good luck out there.
-D.C.