_______FUSHCA DESIGN_______

In this personal project I wanted to create a digital experience that covers not only Unreal Engine environments, but wanted to discover what all I could make using AI assistance. This took in total about 12 weeks of my time.
I created a world with a vision, that would test AIs ability to use an odd mix of ques: Japanese and Mexican influence.
You’ll see it in the cars, the characters, and the graphics.
The video I created using: Blender, ComfyUI, and Seedance 2.0. Image and theme world creation using: LORA learning via Koyha_SS, ComfyUI, Unreal, Photoshop. Image Plate creation using: LORA learning via Koha_SS Firefly, Nano Banana, and Flux.
The video explores character narrative of driver Frankie and her dog Ollie.

This was my first session of images after I spent a few weeks setting up my internal database. My LoRAs has my internal database of images to learn from and I had them working through Comfy UI. You can see that while they are in an old area, these cars are very generic and not japanese despite my mentions in the prompts for JDM rally cars with meso american accents. 

After doing some R&D I discovered that I didnt have enough images in my database for my internal AI to learn from. I added more. While the spatial quality and livery proportions wer emore what i wanted, it was still ignoring my meso american asks.

Here I am using very desciptive prompts about meso american eagles on doors. You can see the art is nice, it is thinking. However it is missing a few things. I went back to the beginning and decided to look over all the text base learning prompts given to my images by the system. 

Here is where I had my first breakthrough. Its learning!!! I had to go back throgh my internal database and create a language for livery spatial awareness and educate it on meso american art. 

The next step is to have images that look like real images and not AI slop. Something I can use. I have started looking at 3rd step workflows to polish blockout liveries. This is still a WIP

Image content creation using traditional workflows for concepting based on references. For this task I used myself and Noodle. The style of my vision is realistc with some minor fantasy illustrative touches. In the plates you can see the evolution of the character to the final and some supporting plates for emotion and visul qeus.

Same method for the vehicle, I started by using 3d models I made using my LORA system above to get the final car. I created many interiors o describe what I needed for the inside of the car.

For location I used Google earth, and my plate images for the charatcer to derive a setting that is both real and illustrative.