Stability can make videos now
Stable Video Diffusion is now a thing, allowing the company’s generative AI art to be animated, Stability AI has announced. Video Diffusion is available as part of a research preview, and works by creating a video from a single still image. “This state-of-the-art generative AI video model represents a significant step in our journey toward creating models for everyone of every type,” Stability said.
Video Diffusion is available in the form of two image-to-video models, capable of generating 14 and 12 frames “at customizable frame rates between 3 and 30 frames per second,” it said. You won’t get HD videos longer than 5 seconds, with Stability capping the time and resolution to 576 × 1024. According to its website, videos should take around 2 minutes (or less) to process.
Stability didn’t waste any time patting itself on the back, noting that “at the time of release in their foundational form, we have found these models surpass the leading closed models in user preference studies.” Yeah, we’re sure we’d say the same in a world filled to the brim with ChatGPTs, Llamas, Copilots, and Bards.
Despite surpassing “the leading closed models in user preference studies,” Video Diffusion does have some limitations, such as the 5-second cap, very slow camera pans, and the inability to properly render legible text or normal-looking human faces. Even so, the cherry-picked videos used in the launch trailer seem to have a relatively high quality compared to other models available.
Not anybody can just get access – though potential users can sign up for the waitlist here.
Chatting up ChatGPT for free
OpenAI may have had a topsy-turvy few weeks with the firing and subsequent rehiring of its ex- and now current CEO, Sam Altman, but that didn’t stop the company from pushing out updates. It’s a big one, too, as “ChatGPT with voice” makes its way onto the ChatGPT app for both Android and iOS users. This isn’t anything wholly new – OpenAI launched ChatGPT’s voice feature in September, though it was locked behind the company’s Plus and Enterprise paywalls.
The update is making one of ChatGPT’s best features a free one, though it might take a minute before it reaches your device. When it does, you’ll be able to tap the ‘headphone’ symbol and ask it a question – just like you would Siri or Samsung’s Bixby (before shutting it off forever). The model in question will only answer direct questions, however. It won’t be getting access to your device’s contacts or calendars to schedule appointments or start a call.
ChatGPT’s voice power comes from a text-to-speech model that can generate “human-like audio from just text and a few seconds of sample speech.” The company worked with multiple voice actors to create a catalogue of five different voices to choose from.
There are some concerns around ChatGPT’s voice synthesis, with OpenAI admitting that the technology could present big risks if abused. It’s an issue that’s currently plaguing social media, with deepfake videos, voices and images breeding a pit of uncertainty and misinformation that can’t be easily tracked and disproven.
Ocarina of Time x Studio Ghibli
We’re about to make some of you feel really old. The Legend of Zelda: Ocarina of Time is officially twenty-five years old. You’d think Nintendo might release a direct Switch port that doesn’t require a subscription to Online+ Expansion, but no. We’re not that lucky. All we’re left with is the thought that the company is making a live-action film and not what the fans want: a Zelda film with Studio Ghibli-like animation. Nintendo might be denying us, but a YouTuber by the name of RwanLink isn’t.
In honour of the game’s anniversary, the YouTuber has recreated Castle Town in a Studio Ghibli style using Unreal Engine 5 that took over four months. And look at him now – making it into Stuff’s Light Start. According to Eurogamer, most of the job was done by the YouTuber himself, with the music being the only outsourced piece of the project.
The 30-second trailer is nothing short of gorgeous and only makes us resent the Nintendo/Sony partnership even more. If 30 seconds isn’t enough to sate your appetite, RwanLink has a nearly 20-minute video up, showing actual “gameplay” of the world and its surrounds. It’s a technical marvel and does a brilliant job of showing off Unreal Engine 5’s capabilities.
Whether it’ll ever turn into anything more remains to be seen. It’ll never be released – free or not – thanks to Nintendo’s lawyers, but it’s a YouTube series that we’d watch the hell out of.
EA Sports FC 24 is going to the Euros
EA Sports FC 24 might not be our favourite FIFA game of all time, but it’s always nice when the company puts in a modicum of effort, which it’s doing through a massive (and free) Euro 2024 update being released alongside the start of the European tournament.
It’s not just limited to FC 24, either. The update will be coming to FC Mobile and FC Online in the Summer of 2024 (that’s Winter, for us South Africans). EA also said that anyone who logs in and plays FC 24 again before 16 January 2024 will unlock “one of Europe’s biggest stars” as an untradeable Ultimate Team player item. Playing now won’t do much, as the promotion only kicks off on 18 December.
The players include: Jack Grealish, Ousmane Dembélé, Federico Chiesa, Florian Wirtz, Virgil Van Djik, and Alvaro Morata.