Details about VisionForge (Stable Diffusion 1.5)

Developed Stable Diffusion 1.5 from scratch to gain a comprehensive understanding of its architecture, with a focus on the impact of activation functions on image generation. Implemented a custom weight conversion system to map pre-trained Stable Diffusion weights to the new model, allowing for detailed experimentation. The project demonstrated how even minor changes to activation functions can significantly alter the generated images, highlighting their critical role in generative AI.