I've been testing the new Gen2 text to video AI model from Pika which is available for the public. Bare in mind this sample I made is text to video, may work far better on image to video.
The prompt for these clips were variations of the following:
"extreme close-up of a chameleon gently moving, very detailed and realistic skin texture, keep subject in sharp focus, national geographic style"
Negative prompt:
"morphing, erratic fluctuation in motion, noisy, bad quality, distorted, poorly drawn, blurry, grainy, low resolution, oversaturated, lack of detail, inconsistent lighting"
Varied animation settings to see how the outcome changes.
We can see that there still is a lot to evolve on Pika Labs text to video AI when it comes to anatomy and texture detail. Continuity and animation is somewhat better than Runway ML Gen2. I did manage to get a couple of decent shots. At least with Pika I tried to make a moving chameleon and managed even though it looks more like a lizard. May work better using an image other than text, or image + text.
I will make more tests focused on other subjects such as human portraits, landscapes etc. during the following weeks. I believe it may work well with landscapes especially image to video, other than text to video.