VEO 3:
I’ve been using Veo 3 to build out video sequences through structured JSON prompts. By using timestamps, I’m able to trigger specific visual events exactly when they need to happen. It gives me a level of control over the timing and composition that you just can't get with standard text prompting.
Text to Video: An ad I created for a fake airline, Aurora Air, using Veo 3.
I using structured JSON prompts with timestamps it really helps you get you closer to your vision.
One cool takeaway from this project: you can incorporate branding colors as accent details throughout the videos. For example, in the opening scene, the chairs is turquoise (#00C2CB) and the suitcase in the second scene. It’s a subtle touch, but it really ties the visuals together.
WORK PROJECTS:
This video was created for the o2 scambaiter project using audio to video lip sync with KlingAI. These videos were created for socail media and tv.
These image-to-video animations were created using Midjourney. Cadbury invited people to submit vintage photos of themselves with Cadbury products. I selected the best photos and animated them. The quality isn’t great since the source images were low resolution. This project was done in February 2024, and both quality and animation have improved significantly since then.
ComfyUI:
I've been focusing on creating high-resolution, realistic videos of people and nature. This video was generated using the HunYuan model with ComfyUI at 1280x720 resolution and 24fps. (March 2025)
This video was generated using a three-step workflow: text-to-image creation, face swapping, and image-to-video conversion with the LTXV video model, all completed in ComfyUI. (Febuary 2025)
Created with the LTXV video model, these videos use image-to-video at 768x512px and 24fps. The aim was to generate highly realistic, high-quality visuals of people, all within ComfyUI. (Febuary 2025)
Animate Diffusion:
All the videos below were created using the Animate Diffusion model with a technique called prompt traveling. This method allows you to set prompts to start and end on different frames while adjusting the overlap between them for smooth morphing effects.