We often talk about the "fragmented workflow" problem in generative AI. You generate a character in one place, upscale it in another, and attempt to animate it in a third. It is a friction-filled process that usually results in a desktop full of files and a final video that looks nothing like your original concept.

However, the landscape is shifting. We have been testing a remarkable new workflow that solves two massive headaches—character consistency and the "morphing" problem—by combining the power of Google’s Nano Banana Pro with the latest video capabilities of Kling 2.5.

If you have been struggling to get consistent character videos for fashion or storyboarding, this approach is about to become your new favourite method.

The Consistency Problem

Usually, creating a 3D-consistent video of a specific character involves a lot of luck. You prompt an image generator, hope for a good result, and then cross your fingers that the face doesn't morph into a stranger when the character turns their head in the video stage.

To solve this, we took a different approach using Nano Banana Pro.

Step 1: The "Contact Sheet" with Nano Banana Pro

The secret lies in how you set up your source image. Instead of generating a standard portrait, we used Nano Banana Pro to create a "contact sheet"—a grid showing our subject from multiple angles (front, side, profile) simultaneously.

For our test, we generated a model wearing a Clicky branded hoodie. Because all the angles are generated in the same diffusion pass by Nano Banana Pro, it locks in the character's identity, the fabric texture of the hoodie, and the lighting across all views. There is no drift between the front view and the side view because they are effectively the same image.

Step 2: Animating with Kling 2.5

Here is where the magic happens. We took that generated contact sheet and fed it directly into Kling 2.5.

Kling’s latest model is surprisingly adept at understanding 3D geometry from these multi-angle layouts. By providing it with the contact sheet as the reference frame, the model "knows" exactly what the character looks like from the side and back before it even begins generating the movement.

The Result

We typed a simple prompt instructing the character to turn and look at the camera. The result was a fully animated video of our guy in the Clicky hoodie performing a smooth movement with perfect 3D consistency.

The logo on the hoodie didn't warp, the facial features remained sharp, and the lighting stayed true to the original Nano Banana Pro generation. It feels less like "generating" a video and more like directing a virtual shoot.

Strap in, creatives. These are wild times.

A couple of other cool examples I found…