AIGC Stylized Video

#AIGC #Shanghai in 30's #Stable Diffusion #LoRA

Project Overview

In this work, I designed a workflow for generating video in 1930’s Shanghai city style based on comfyUI. The models include SDXL, AnimateDiff, SVD and my custom LoRA.
To generate video in specific artstyle, I trained a stylized LoRA using 30+ pictures of Shanghai city and corresponding objects at that time.

Workflow

The overall generation process includes two steps. First is Text to Image, where primarily use diffusion model and my custom stylized LoRA.
The images would then go through an adjustment process, where upscale model will be utilized.
The second step is Image to Video. The diffusion model is applied again generate video outputs.

LoRA Output

Here are some picture examples generated using custom LoRA.

Model Comparison

Here shows the generated images using different versions of LoRA.

Video Showcase

Here are corresponding video outputs.

Conclusion

This work designed a complete workflow including image and video generation. It also applied custom stylized LoRA model to control the art style of outputs.