I started my morning watching a short movie, knowing the whole thing was made with AI. But it only took me a few seconds to forget that it was AI-generated. Then it took me a few more seconds to remember why I was watching the clip in the first place. It was to see how believable this particular AI-generated video truly was. With that goal in mind, I started paying attention to what was happening in the short movie. I couldn’t tell what was real and what was AI-generated… except for the giant robot.
Once I saw it, along with the explanation of how The Wall Street Journal’s Joanna Stern pulled it off, I had to watch it again to try spotting the AI-generated content and inconsistencies. Spoiler alert: I rarely could.
I’m not surprised. I’ve been covering the AI industry since the early days of ChatGPT, so I know how we got here. Just a few days ago, we saw Google unveil Veo 3 and Flow, and the web was flooded with amazing Veo 3 AI clips that looked like they were shot with real cameras.
I said at the time that this kind of technology would make you distrust anything you see online unless it clearly states that AI was involved in creating it.
The WSJ clip, which you’ll see below, proves that point.
Stern and video and audio producer Jarrard Cole used about 1,000 AI-generated clips to create their story featuring the fictitious Optimax 5000 robot. The entire production cost around $1,000, which included some free AI credits offered by the companies whose products they tested.
Considering the result, a 3-minute clip that’s almost indistinguishable from a mini-movie made with traditional tools and effects, that’s a remarkable deal. It wasn’t just about money, though. A lot of time went into generating those clips.
Shooting the whole thing with a real crew would have taken much longer and been much more expensive. Then there’s all the post-production work and the special effects needed to make a robot move.
The best part of the video is Stern explaining how it was all done.
Veo 3 and Runway, a powerful AI video generation tool similar to Google’s, still can’t maintain character and scene consistency across clips on their own.
They started by using Midjourney to create Optimax 5000. They took photos of Stern to feed into Runway Gen-4 for the first scene. The hose and neighborhood street were made in Runway using a text prompt.
The first frame from Runway was then moved into Veo 2 and Veo 3, where a different prompt was used to animate it.
Some scenes in the robot movie don’t show the characters’ faces, so character permanence wasn’t needed. Those were animated with text-to-video prompts in Veo 3.
Cloning Stern’s gestures for an AI-generated scientist was done with Runway ReStyle. Except for Stern’s lines, the WSJ team used a range of AI tools for audio and sound.
Veo 3 handled some of the audio. ElevenLabs provided voices for the robot and scientist, while Suno AI generated the song at the end of the movie.
Once they had everything, they used Adobe Premiere Pro to piece it all together. It would have been even more impressive to make the whole film inside an AI-only tool like Google’s Flow. That might not be possible yet, but it’s on the horizon.
Yes, AI like this will replace some jobs in the movie industry. It will start gradually, since the tools still need work. But it wouldn’t make sense for creators to avoid using AI like Veo 3 and Runway to help generate content for their projects.
Don’t be fooled. AI won’t do all the work for you. They spent a lot of time generating the right scenes, tweaking prompts, and cutting out the bloopers.
Most importantly, Stern makes a great point in the video about one thing that won’t disappear when making movies with AI: Creativity. The silly robot movie came from humans who imagined it all. “These tools are nothing without human input, creativity, and original ideas,” she says in the Wall Street Journal story that goes with the AI movie.
