Use Case - AI Image Generation

Below is my presentation. I copy and pasted the text from my slides, so the formatting is kinda funky! FYI some info like the "how to" is outdated since this presentation is from October 2023.


Here are the links to the result:



Look What I Did - Jekyll Island Fiat Scratch




Frost Children - HI 5 




MIZMOR - No Place To Arrive




Gromo - Prologue/The Vortex (Short Film)


AI Image Generation

in the Real World


Goal: I wanted to make digital/animated music videos for my music without spending money. 


What I needed in an image generation tool: FREE! Video to video User friendly Smooth/not flickery

Warpfusion

https://github.com/Sxela/WarpFusion 

Stable_warpfusion.ipynb

https://www.patreon.com/sxela 


Free ❌✅ (mostly free)

Video to video ✅

User friendly  ✅

Visually Smooth vid2vid ✅


  • Based on disco diffusion

  • Optional to support developer on Patreon

  • Provides updates to warpfusion to include new features

  • such as controlnets, loras, stable diffusion XL

  • Tech support

  • Discord community


Problems I ran into: - Google Colab became expensive - Current hardware was too slow or could not handle image generation I wanted to do - Initial set up can be difficult - I had trouble installing CUDAs/ setting up paths for cuda and python
BASIC OVERVIEW!!

Style strength:


The higher the number, the more it follows the prompt


The lower the number, the more it resembles the original video

It takes ~30 minutes to start generating images. It depends how long the video is.




  • I practiced by making fan art 

  • Eventually led to some paid work



Revolver Magazine




Comments

Post a Comment

Popular posts from this blog

Big Data - Regression & Classification algorithms

Image
Regression Analysis of Boston Housing Data Google Colab Link In my exploration of regression algorithms using the Boston Housing dataset, I tried several models to predict housing prices. Gradient Boost Tree Regressor t his model stands out due to its impressive performance metrics. It achieves a Root Mean Square Error (RMSE) of 3.67409 on the test data, coupled with an R-squared value of 0.84279, indicating a strong fit compared to the basic linear regression models. Random Forest Regressor This model displays robustness similar to the Gradient Boost Tree, with a comparable RMSE and an R-squared value of 0.786508. It offers a reliable alternative with its ensemble approach. Decision Tree Regressor Although slightly less effective than the Random Forest, the Decision Tree Regressor still shows substantial predictive capability with an R-squared value of 0.727668, illustrating its utility in scenarios where model interpretability is key. Classification for SMS Spam Detection Google Cola...
Read more