ABOUT
The AspIyr Brand Advertisement project is a 90 second emotional rollercoaster, developed with the help of various text-to-video, text-to-image, and text-to-audio AI models. Each video clip has been generated using AI. Most of the sound and voice effects are also AI, with only the BGM belonging to one of the best Sci-Fi movies ever - INTERSTELLAR, & composed by the legendary Sir Hans Zimmer.
This Advertisement does not showcase a product or service, but rather a depiction of what humanity has already achieved, is currently achieving, and can achieve in the future, alongside my vision for working in collaboration with this revolutionary creation of ours known as 'Artificial Intelligence'.
WHY?
The idea to create an advertisement was born from my prior novice experience with video editing and a heavy dose of inspiration I took from a particular footwear advertisement which repeatedly showed up in my Youtube feed (Good thing I don't use an adblocker huh)
As someone who learns best via engaging video content, I believe that the best way to showcase your skillset, and the value you can provide/create is by allowing others to see what you see in that head of yours, be it via an interactive powerpoint, a storyboard animation, or as in my case, a powerful and captivating video.
“Creativity is intelligence having fun.” — Albert Einstein
HOW?
PHASE 1 - Planning & Testing Within The OpenAI Ecosystem
Given the fact that the planning for the entire portfolio was done with ChatGPT, it was only obvious to discuss my ideas for the Ad and refine them into scene-by-scene prompts with my trusty AI companion, and test them out using it's SORA model.
This is the very first prompt I pasted in Sora to generate a scene and the result is as shown below - "A burning campfire. Lightning hits the flame and turns it into a glowing light bulb. Black background, cinematic lighting, slow zoom in".
PHASE 2 - Discovering OpenArt & Testing Multiple Models
Although the quality of Sora's video outputs were quite impressive, I quickly realised the model was unable to process multiple scene transitions within a single prompt. After multiple prompt iterations and further breakdown of the scenes, the ouput remianed surprisingly underwhelming.
While watching a few Sora prompting tutorials to improve the output quality, I came across a video showcasing OpenArt.ai, a platform that allows the use of multiple image, audio and video generation tools all under 1 subscription, by allocating a set number of credits as per the chosen tier.
The decision to purchase a subscription and experiment with different models is what allowed me to create the brand advertisement in the matter of a few days. Each model excels at particular functions and generating certain types of outputs. To further demonstrate this, I have included the outputs generated by 3 text-to-video models, while using the same prompt for each.
Google VEO 3: Native audio & ultra-realistic, prompt-accurate results (Most accurate but expensive)
Seedance 1.0: Natively supports narrative videos with multiple cohesive shots (Cheapest)
MiniMax Hailuo 02: Handles complex scenes with extreme physics & motion (Most consistent and value for money)
FUN FACT: 90% of the clips in the final Ad were generated using the Minimax Heiluo 02 model.
PHASE 3 - FINAL EDITING (THE HUMAN TOUCH)
As great as the clips were individually, they didn't make sense unless arranged together by a person, aka ME. This is a clear example of AI's current capabilities and limits. This is where my basic experience with video editing was put to the test, and I couldn't be happier with the overall result.
Video Editing Software : DaVinci Resolve
Result
5 days & 50+ generations later, the AspIyr Brand Advertisement was born. What started off a simple idea, is now a full fledged visual treat - a collaboration between 2 AI's.
This is some award winning stuff right here. Don't believe me? Just WATCH