Date Started: June 17th, 2025
Date ended: July 24th, 2025.
TOOLS: ChatGPT & Gemini for Prompt development, VEO3 Text to Video, AfterEffects/Photoshop for cleanup, and Premiere as Editor.
This is an Ai Study proposed to me by a friend, a professor at NYU. It required a deep dive into Ai to recreate a real scenario; one in which has 4 characters at an office space, and he wanted to find out if this could be well directed __this implies a push to realism.
Very quickly I found out that to create a video like this, enact a real scene, block it and structured as a real shoot is needless to say complex. For this test I worked with Veo3, currently the platform that proposed the best results in the market.
The process.
I should say, this was my first attempt at working with AI Video creation, and aside of ChatGPT, Perplexity, and Firefly, this is the first time I fully engaged Ai as the building block for a Production and/or Post. In the past, I have used Ai in Post-Production to create background extensions, patches, replacement elements, rather small tasks that could take some time to create. I never got perfection but I was able to use it successfully for the purpose intended.
I started building my prompting with ChatGPT, and later during the process I switched and kept with Gemini with the hopeful reasoning of directly tailored prompts, hence better results. At the end this proved to be inconsistent as well __in my conclusion I will tell you why.
It took me at least 3 weeks on-and-off to wrap my head around prompting and learning to pivot from my results. Also I got acquainted to the limitations of an Ai video creator: It performs great with 1-2 characters, with short dialogue and simple language (we have to take into account Audio Sync it is still considered a new feature), also building interactions and solid matter is requires iteration and a different level of complexity at prompting. This is one of the reasons why Ai excels at fantastic and "out-of-the-box", anything is visually possible. For someone that made a career at fixing production mistakes where detail is the goal, it is particularly difficult to navigate through the process.
Of course I am being ambiguous and not listing all the limitations.