**Title: DDP Seoul Light Media Art Facade Proposal
Author: Nsyme
Date posted: 2022/12/07
Summary
Seoul Light Festival is one of the biggest projection mapping media art events in Korea. In this year, I’m part of the main artist of this show and I’m gonna make 3 mins of video art,. fully made out of generative AI models like Disco diffusion, and Stable Diffusion.
Background
The resolution of the Video is 6000x1600 and I need to make 5400 frames of images to make 3 mins of video. For the size of the video and the speed of the diffusion models, I desperately need A100 and to test and experiment with all the visuals before actual render, I need several A100 GPUs.
Timeline
1 month (Sep 5th, 2022 ~ ): Research various type of diffusion model notebook (Disco Diffusion, Stable diffusion, Stable Warpfusion, Stable diffusion Deforum) and experiment
1 month (Oct 3rd, 2022 ~): Revise & develop models and experiment visual aesthetics & experiment ai-upscaler(Real-Esrgan) and frame interpolation(FILM, RIFE)
1 month (Nov 7th, 2022 ~): Fine-tune Stable Diffusion model with Dream Booth. Render a bunch of different kinds of animation with Deforum Stable diffusion model.
Target
AI image(video) generation (Stable diffusion)
AI upscale (Real-ESRgan)
AI frame interpolation for video (FILM)
Fine-tune Stable diffusion mode
AI model training
Publish trained model on Kaggle
Required Hardware Specifications
A100@40GB * 4