For the past year we’ve been working extensively with both Google VEO and OpenAI Sora (with an emphasis on Sora 2 Pro the last couple of months). We wanted to see how confidently we can tell our clients that some concepts can be achieved completely with AI.
A year ago, the answer would have been one hundred percent…not confident. There were still issues with video quality, audio, and prompt execution. Fast forward one year to 2026, and Generative AI has taken leaps and bounds. So, can we now tell clients that AI can achieve the same results as a live shoot for just a fraction of the budget?
Unfortunately, the answer is still no.
Generative AI is still not a reliable replacement for live action video production
We attempted to leverage AI for some small tasks (mainly b-roll generation) and pre-visualization. In most cases, it took several attempts with multiple versions of prompts in order to generate a usable clip. Cost-wise, we spent several hundred dollars more on extra credits for a simple clip than we initially expected. We also burned a lot more hours on prompt development and generations, each of which takes about 5 to 10 minutes. The cost would still be more efficient than a live action budget, even for a 30 to 90-second spot, but we wouldn’t be confident that we could achieve the desired results even after multiple attempts.
As an additional exercise, we took some of our past commercials and tried to replicate the scene with Sora 2 Pro.
On our first attempt, we simply used the script as the prompt. Based on the results, we modified the prompt and ran it again. For this exercise, we capped the number of generations to five.
Here’s a video showcasing these attempts.
Problems with Sora 2 Pro in creating commercials
The main problem with Sora 2 Pro is that it freelances with a great deal of unwanted things. As you can see in several of the clips, our characters ad lib quite a bit. It’s okay for Marlon Brando, but doesn’t quite work for these spots.
Sora also adds scenes, graphics and edits that are not in the prompts. We usually just want a clean steady clip or a single shot that we can edit ourselves, but Sora seems to think that we want it to edit for us and tries to deliver almost a fully produced clip. In some cases, like social media content, this might be okay, but not workable for a video production company trying to put a spot together.
What GenAI video capabilities can we promise our clients in 2026?
Considering how fast AI is developing, this might completely change by the end of the year. But as of now, we can’t promise that we will be able to make every concept come alive and follow a fixed vision.
We can create simple spots as long as there is flexibility in the final product. For this spot for Primo’s Donuts, we had to change a few things during the production process. For example, originally there was an aerial shot of the construction worker with the drone quickly pulling away as he took a bite of a donut. We couldn’t get the desired result even after multiple generations so we simplified it.
We can also promise small clips, like simple b-roll or low-res shots. Even then, considering the quality of the images, we would add disclaimers, especially for broadcast.
Generative AI will eventually take a bigger role in video production. We’ll eventually see it replace traditional filmmaking in our lifetime, but for now it’s still a work-in-progress.
To learn more about Picturelab’s generative AI or traditional video production capabilities, contact us today at info@picturelab.com.
We are based in the San Francisco Bay Area (Mountain View, CA) and the greater Los Angeles/Orange County region (Irvine, CA), but serve clients all over the world.