Over the past few months, we’ve seen how large language models like ChatGPT generate text copy, image generators like Stable Diffusion create images on demand, and even perform text-to-speech. . An enterprising developer named Pizza Later combined five different AI models to create a live-action commercial for a fictional pizza restaurant called “Pepperoni Hug Spot.”
The resulting video we’ve embedded below is both terrifying and impressive. There is even some dialogue and decent background music. However, some characters have a little more facial expressions and dead eyes.
Obviously, the quality of the output leaves something to be desired. Sometimes objects seem to blend into each other. My son said it looked like people were eating pizza that had grown out of the plate.
They all seem to live in the uncanny valley. And the somewhat incoherent script reads like text from another language improperly translated into English (which it wasn’t).
However, it’s impressive to see how close these technologies are to full-fledged readiness. In a short time, you will find that photorealistic video images become more convincing.
To be fair, this video required human editing. After using five different models he used to create various assets for the video, Pizza Later took the time to stitch together the video, dialogue, music, and some custom images using Adobe After Effects. said to have spent All in all it took me 3 hours to complete the project.
Pizza Laiter came up with the idea for the commercial Runway Gen-2 (opens in new tab), a text-to-video model in private beta.In an email interview, the developer told me that the video’s first prompt was “Happy man/woman/family eating pizza at a restaurant, TV commercial.” 1st generation (opens in new tab)creates videos based on existing footage and can be tried today for free via the web or the new iOS app. (opens in new tab).
After watching a high-quality video produced by Runway Gen-2, Pizza Later uses GPT-4 (the engine behind ChatGPT and Bing Chat) to name a fictional pizza joint (Pepperoni Hug Spot). I figured it out and scripted it. The developer then used ElevenLabs Prime Voice AI (opens in new tab) Provides realistic narration with a male voice.they used mid journey (opens in new tab) Generate some images that appear in the video, such as the exterior of the restaurant and some pizza patterns.they also used sound low (opens in new tab) Create background music.
Most of the tools Pizza Later uses are paid, but we offer free trials, low-end free accounts, or your first set of free credits. Clearly, the developer had to piece together the end result, so this is far from a plug-and-play operation.
Perhaps in the near future a multi-model tool like Microsoft Jarvis will be able to perform all these tasks in a single chat prompt. Alternatively, autonomous agents such as AutoGPT (see How to use AutoGPT) generate commercials given the broad goal of marketing a restaurant. But for now, this video is really impressive, even knowing it needed human editing.
