Testing Seedance 2.0 with Pets and People — Reference Images vs Text-to-Video

Recently, FourFeetz Studios tested Seedance 2.0 through Protoface Studio to explore how naturally AI video can portray interactions between people and pets.

One of the most interesting findings was the practical difference between using reference images and generating a scene entirely from text.

We first created a photorealistic image of a woman with a Golden Retriever in a café and attempted to use it as the visual starting point.

In the production environment we tested, however, photorealistic human references could trigger a realistic-person reference restriction, preventing the render from proceeding.

We also tested the Reference Image workflow and encountered a similar limitation.

Instead, we removed the image and generated the scene using only a text prompt.

The result was surprisingly convincing in several areas.

Physical contact between the woman's hand and the dog looked natural, the dog's movement around the person was believable, and the overall photorealistic appearance was strong.

Prompt adherence, however, was not perfect.

For example, the dog was intended to approach from the floor, but the generated scene placed the dog on the café table.

This test gave us an important production lesson.

For natural emotional interaction between a person and a pet, Seedance 2.0's Text-to-Video workflow can be a useful option.

However, when exact blocking, position, or action order matters, shorter and more clearly defined scenes are generally easier to control than trying to describe everything in one generation.

It is also important to distinguish between a model's general capabilities and the input workflows currently available through the platform being used.

At FourFeetz Studios, we will continue documenting both successful results and practical limitations discovered during real AI pet-video production.