GMO's Blog

October 28, 2023 AnimateDiff Introduction AnimateDiff enhances ComfyUI, providing a stable and modular interface for diffusion tasks. It has evolved from its original version, sd-webui-animatediff, now offering smoother video clips with a similar output to tools such as zeroscrope and runaway . Methodology I evaluated AnimateDiff based on its ability to produce smooth video clips. Tests were conducted using ComfyUI and Warpfusion to assess usability, performance, and video quality. The tool was used with two interfaces: ComfyUI (node-based) and Warpfusion (Google Colab). Installation and Setup Integrating AnimateDiff with ComfyUI was complex, requiring many attempts. Workflows were found on GitHub , YouTube , and Reddit .  Features and Functionality AnimateDiff stands out for its smooth video generation, especially for longer clips, though it can be slower than alternatives. Usability Warpfusion was easier to use and allowed frame previews during generation. ComfyUI was more chall...

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 FYI some info like the "how to" is outdated since this presentation is from October 2023.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 http://discodiffusion.com/   Optional to support developer on Patreon Provides updates to warpfusion to include new features such as controlnets, loras, stable diffus...

  Development of a Customized Lora: A Deep Dive into Personalized AI-Generated Imagery October 29th, 2023 Objective: The primary objective of this project was to create a professional-grade photograph using a custom AI model, termed "Lora." This initiative was inspired by Aragon (https://www.aragon.ai/), a commercial AI tool available at Aragon AI, which offers similar capabilities but at a cost. Methodology: The project began with the compilation of two image datasets. The initial dataset lacked diversity, consisting of overly similar images, which necessitated the creation of a second, more varied dataset. Both the standard definition (SD) and extra-large (XL) versions of Lora were trained using Kohya's framework, renowned for its robustness and versatility in managing advanced AI models. Challenges and Results: The anticipated high-quality output from the XL Lora did not meet expectations initially, as the images suffered from pixelation and blurriness. Adjusting the s...

 This is a music video I did for Entheos. I used SDXL. This was so much fun to work on because of the music! I did this months ago so I can't quite remember the details or process I did to make this but it involved a lot of experimentation. I spent a lot of time fine tuning the prompt with many trial and error attempts.  This is one of the tests I did to find the most smooth AI output: 

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...

 Check out the Google Colab Notebook for a word frequency project I worked on for my class, "Python for AI".  You can try it out here ! You can download and upload any text to try out the notebook. I used The Republic by Plato: https://www.sciencetheearth.com/uploads/2/4/6/5/24658156/plato_-_the_republic.pdf