Is Sora's AI Video Ready for Public Media?

Hi all. Welcome to Public Media Innovators Weekly. This week the big news is that generative video tool Sora, from OpenAI, is finally available, but we also look at how YouTube ate podcasting, and finally, we hear the story of an AI “granny” developed to waist scammers time.

But First…

There’s still time this morning to register and join us for our next webinar today at 1pET/10aPT, a full session with Talia Rosen, entitled Unleashing Creativity and Applying Standards to Generative AI. And we will soon be opening registration to our January webinar, our annual collaboration with Current. This year we’ll be looking at the future of public media.

Is Sora’s AI Video Ready for Public Media?

Back in Exploration #98, I asked if generative video was ready for public media. It was February and the sizzle reel for Sora had just been released. This week, 10 months later, the world finally got to meet Sora, and the question remains.

For the 411 on Sora, this article by James O’Donnell in MIT Technology Review is a good place to start. But I can tell you that a ChatGPT Plus account will get you video quality up to 720p and clips up to 10 seconds (which could have already changed by the time you read this). You can get 1-4 variations of the video (output as 16:9, 1:1 or 9:16), but every test video you do burns credits. You automatically get 1000 credits a month and you'll blow through that in no time. That's at best 40 videos (meaning 10 experiments, if you generate 4 variations from a prompt). OpenAI provided some useful tables to help you budget your credit burn. Whenever and however you get access, I recommend starting with 1 variation prompts, at 480p (25 credits) until you get the hang of what Sora can and can’t do.

And there’s a lot it can’t do. So, if you are having FOMO over not having access, I’m here to offer some peace of mind: we are a long way from the highly curated magic on display in Sora’s sizzle real from February.

In my experience so far, a lot of the artifacts that I’ve seen in other generative video tools (e.g., a strange morphing between shapes, a gravity-free quality to character movement) are also quite present in Sora. And that’s a bit disappointing. To me, OpenAI oversold Sora but, as we often say, it’s only going to get better from here. The general consensus seems to be if you want to imitate generic high wide drone shots, this is a pretty solid tool for multi-second clips. Otherwise, you’re likely going to only get 15-45 frames of useable video. And maybe that’s enough. I always like to test with Bison, and it created footage that could have been usable for a quick cut in an image spot.

But I found, when it came to specifics, Sora left me wanting. I asked for a high wide shot of New Orleans, and I got a generic city scape that hinted at the architecture but was clearly not New Orleans (a prompt for a lower shot of Bourbon Street faked it a little better.

Sora can do a passable job at people (better, at least than dinosaurs, as my tests showed). But we need to be hyper vigilant on issues of bias here. I usually test any new visual tool by asking it to generate images of Native Americans and powwows (because we are mid-production on a video game set during a powwow). Sora wasn’t the worst offender we’ve seen since the advent of generative AI. But, as our video game producer (and member of the Menominee tribe) Jacob Schwitzer pointed out, all the spectators of the powwow in the video are white.

There are already plenty of Sora review videos on YouTube. But if you’d like to see a few experiments I’ve done you can access them here. Additionally, Talia Rosen, with PBS Standards & Practices, also got in before they lowered the boom on new accounts. She has generously shared access to her experiments (i.e. these are not official PBS test videos…though, my $.02, as a system we’re lucky to have someone innovative in Standards & Practices that wants to spend time exploring new tools like these to understand their potential and limits).

If you have a chance, and a ChatGPT Plus account, you should definitely give it a try over the holidays. But any public media organization serious about experimenting with Sora will want to subscribe to the Pro tier ($200/month). This gets you 10,000 credits a month, but more importantly you get access to “relaxed” videos, which essentially sit longer in the queue but don’t burn credits. You also get access to 1080p videos and can generate clips up to 20 seconds in length without the Sora watermark. Those extra credits will be especially useful in freeing you up to experiment with Sora’s editing tools (also available to Plus users), which allow you to remix and recut footage.

Is Sora ready for broadcast? For now, I’d say that if you’re broadcast production workflow is based around 4K video, the answer is “no.” Where I can see its potential, however, is for video destined for social media. In the near term, I think that’s where Sora generates the greatest artistic potential for public media. But the devil is truly in the details with Sora, so carefully review your Sora footage for bias and accuracy. The more detailed you try to be with your footage, the more strict a review you’ll need to conduct on that footage.

Okay, on to the links….

Focus…

How to use Sora, OpenAI’s new video generating tool (James O'Donnell - MIT Technology Review) - O'Donnell does an excellent job breaking down the essential information about Sora, including the fundamental question: "How much can you do with it? You can generate videos from text prompts, change the style of videos and change elements with a tool called Remix, and assemble multiple clips together with Storyboard. Sora also provides preset styles you can apply to your videos, like moody film noir or cardboard and papercraft, which gives a stop-motion feel. You can also trim and loop the videos that you make."

Learn…

Unleashing Creativity and Applying Standards to Generative AI (Talia Rosen - PBS Standards & Practices) - Join us, Thursday, December 12, 1:30 p.m. ET for a lively presentation and discussion surrounding the application of standards while using generative AI. We will explore specific use cases and share best practices. Bring your station’s insights, dilemmas, experiences, and questions to make the most of our 90-minutes together. This is our first co-presentation with Executive Content Managers group. All are welcome, and you can register for the webinar here.

Earn…

Open Call on Infrastructure (Press Forward) - This one has a tight turnaround (deadline is noon ET, 1/15) but if you have an idea already baking in your shop then give this one a look.

• Going forward, when we find them, we'll be publishing grant opportunities here in the newsletter. If you see one that we should share with the community, send it my way!

GNI AI Lab for News Sustainability (Google) - H/t to Liz Maestri at PMVG for making me aware of this one. The following is from their application form. “The program will begin in March 2025 and run approximately six months…. This program will help organizations use AI to drive the business of news, giving participants a structured way to test AI tools in pursuit of key business goals such as: audience development, revenue growth, product development, operational efficiency, data analytics, or other focuses as sought by the participants. The AI Lab is a perfect fit for operational leaders inside news organizations who have dabbled in generative AI but haven’t yet found a way to put it to work for their major business goals.”

• It continues. “Your organization is a good fit if: You are a local news organization (digital, print, broadcast, etc.), nonprofit or for-profit, focused on public service for audiences in the United States or Canada; You have an interest in experimenting with AI applications to better support your business in any of the areas listed above, and you are committed to a test-and-learn mindset to pursue it; There’s a particular focus on applicants who self identify in the following ways: Your news organization is owned, led by, or serves Black, Indigenous, Latinx, Asian and/or other people of color; your news organization is owned, led by, or serves women or the LGBTQ+ community; and/or your news organization serves a defined local community.”

Think…

Public media reconciles its past with the present (Kristen Muller - NiemanLab) - Into NiemanLab's predictions for 2025, Muller throws a prescription "that civic leaders, educators, artists, and scientists of all political stripes can convene the modern-day equivalent of the Carnegie Commission to reimagine and redesign a new, networked media space, accessible to all, to serve the informational and educational needs of both individuals and communities."

• For more thought provoking predictions, see all of NiemanLab’s 2025 list.

To survive, pubmedia must embrace personality, co-ownership and connection — before it’s too late (Andrew Ramsammy - Current) - The op-ed from the former Chief Impact Officer at the Local Media Association, is radio-focused, but his advice can play just as well for television-based media companies. There are too many quotable lines here for me to highlight just one but I do like his section headlines: "The old gatekeeper model is dead," "Making creators stakeholders," and "The new public media model: personality + ownership + connection." There's good food for thought here.

5 common misconceptions about women and entrepreneurship (Ingrid Chadwick & Alexandra Dawson) - While the study referenced here was of a Canadian dataset, the misconceptions listed are just as applicable south of the border.

Know…

How YouTube Ate Podcasting (John Herrman - New York Magazine) - How do you make audio go viral? Turns out, you refashion it as video. But lest you think this is match point for video, it turns out that seeming to be good at audio may make viewers take you more seriously as a creator. To me this reinforces this idea that being radio or TV is no longer a useful way to separate what we do. Maybe you're better conceiving and executing video, maybe you're better at conceiving and executing audio. But it's all merging and the successful public media organization of the future needs to be good at both, regardless of your origins.

Tuning out TV news might be behind the decline in media trust. (No, really!) (Sarah Scire - NiemanLab) - The researchers are careful to point out that their findings come from societies with higher degrees of press freedom, and that they the study doesn't definitively prove causation, but there seems to be some type of connection here.

The AI Reporter That Took My Old Job Just Got Fired (Guthrie Scrimgeour - Wired) - A little story out of the Hawaiian island of Kauai. This seems like an interesting experiment that should have just stayed an experiment. The development company behind the tech called it a success, and from an R&D perspective it probably was. But that doesn't make it a business success, nor does it make it good for the community.

• And here’s how the story began, courtesy of Matthew Gault at 404 Media: Historic Newspaper Uses Janky AI Newscasters Instead of Human Journalists

Getting started with AI: Good enough prompting (Ethan Mollick - One Useful Thing) - It's been a while since I've highlighted Mollick's writing here. His advice is as sage as ever and is remarkably consistent from when he was the main guest on our February webinar: "The single most useful thing you can do to understand AI is to use AI."

Xreal One Smart Glasses Have Entered the Vision Pro Zone, and I Wore Them (Scott Stein - CNET) - A lot of the discussion around the augmented reality end of the xR space is focused on META and Apple, but Xreal has been impressing testers with various iterations of their tech for well over a year now. Put this on your list of companies to watch.

And finally…

The AI granny scamming the scammers (Will Chalk & Joe Tidy - BBC) - And finally, a lite bit of just desserts for those preying on the elderly.

Have a creative, productive week!