Two years ago, I made this video and for its time, it was pretty damn good. Link
I created it by editing this workflow to fit my needs Link
Here's what I did back then: I generated images using my own merged SDXL model, then added details with my own merged SD 1.5 model. After that, I turned those images into video using AnimateDiff and ControlNet IPAdapter.
Fast forward to now - I took those same images and ran them through Wan2.2 with denoise set to 0.75. The results? Way more realistic and visually stunning.
I generated videos using first frame and last frame settings, then stitched everything together in After Effects. The 2-year evolution of AI became crystal clear when you see them side by side.
I'm planning to make this comparison series an annual thing. Let's see what models we'll be playing with next year and what crazy stuff we can create.
A small request: I want to grow my YouTube channel; I would be very grateful if you subscribed.