Last week we explored the latest Generated Video (GV) pipeline. This week Seattle’s Yutao Han, aka Tao Prompts, goes further and illustrates How to Create Ai Videos of Yourself!
The goal here is to consistently end up with the same real person in multiple generated video clips.
“In this tutorial we’ll learn how to use the Flux image generator to train a custom AI model specifically for your own face and generate AI photos of yourself. Then we’ll animate those photos with the Kling AI video generator, which in my opinion generates the best AI videos right now.”
In a nutshell, the process is:
- Create an archive of at least ten photos of your star
- Upload this to the Ostris flux-dev-lora-trainer model on Replicate
- Train the LORA custom image model and use it to generate key frames
- Upscale these images on Magnific, optionally
- Generate six second clips in Kling AI with these images
My take: it seems week by week we’re getting closer to truly usable generated video that rivals (or even surpasses) Hollywood’s CGI/VFX. Imagine being able to train more than one LORA model into Flux for Kling. I have it on good authority that that is just around the corner.