Challenge
Can a fictional character be generated consistently across various angles, scenes, and lighting conditions using generative AI?
Approach
Created a base image of “Horseman” in MidJourney, animated it in Kling, extracted 16 frames, and used them to train a custom FLUX LoRA model for consistent identity.
My Role
I designed the character, built the training dataset, trained the LoRA, created scene prompts, and tested animations across different tools.
Result
Achieved strong visual consistency of the character across multiple scenes. The study shows potential for scalable character pipelines using LoRA fine-tuning.