Drawn to Walk

 

One Drawing, One Playblast

When Kling O1 came out on December 1, 2025, I tried it right away. One day later I rendered this clip, and for the first time an AI video model did exactly what I had been trying to get out of these tools for months.

The setup was simple. I took two things from my own archive: an elephant I had drawn, and an older 3D elephant model of mine with a walk cycle I had animated. I exported the walk as a playblast and gave both to Kling O1 with one clear instruction: use image 1 as the style reference and the video as the motion reference.

What came out is the clip above. My drawn elephant walks with the timing and weight of the original animation. For this video, I cut the original playblast and the result together with a wipe, with the drawing in the corner. That way you can see that the model follows the movement precisely, step by step.

Months of Mush

That sounds obvious today, but it wasn't back then. I had run many experiments like this before, and none of them really worked. The results were always a strange mush, full of odd perspectives that existed neither in the drawing nor in the animation. The models simply never understood what the two inputs were supposed to do together.

Kling O1 was the first video model that could make sense of it and turn it into what I had in mind. For me, that was a milestone.

Early Days

As far as I could see on X and Instagram at the time, hardly anyone was using video models this way yet, feeding in their own animation as motion and their own artwork as style. Since then, the approach has become part of how I work. In later experiments like The Fighter, Maya still provides the choreography while the AI takes care of the look.

https://chris-garcia.art/