🔹 Estimated Duration: 6-8 hours
🔹 Level: Intermediate (requires basic Python and ML knowledge; completion of Course 0 and Course 2 recommended)
🔹 Focus: Theoretical-practical, emphasizing deep conceptual understanding and step-by-step guided application
🔹 Tools: Hugging Face Diffusers, 🤗 Transformers, Google Colab (GPU), Automatic1111 (optional local), Replicate API (optional cloud)
Generative artificial intelligence has burst onto the global cultural scene as a transformative force, redefining the boundaries of creativity, design, visual storytelling, and content production. Unlike traditional predictive models that classify, cluster, or forecast, generative models aim fundamentally to create — to generate new data that did not previously exist, yet appears realistic, coherent, and aesthetically valid within a given context. This capacity for synthesis, stylistic imitation, and extrapolation has unlocked unprecedented possibilities across diverse fields such as digital art, marketing, architecture, fashion, entertainment, and education.
Among the multiple architectures developed to tackle this task — from Autoregressive Models to Generative Adversarial Networks (GANs) — Diffusion Models have emerged as the dominant paradigm for high-quality image generation, thanks to their training stability, ability to produce fine details, and robustness against mode collapse, a frequent issue with GANs. Models such as Stable Diffusion, DALL·E 2, Imagen, and Midjourney — all based on diffusion variants — have demonstrated an astonishing ability to translate complex textual descriptions into coherent, stylized, and often surprisingly original visual images.
This course is designed to guide students from the conceptual foundations of diffusion-based generation to the practical implementation of models capable of generating customized images, including fine-tuning techniques to adapt these models to specific styles, objects, or concepts. This is not merely about learning to use a tool, but about understanding the internal mechanism that allows a machine to “imagine” from words.