Image and Video Generation
For engineers who want to understand how modern image and video generators work, from the maths of diffusion to the tools that control and fine-tune them. You will be able to run, steer and adapt Stable Diffusion pipelines in code, judge video generation platforms on your own shots, and build a small tool that produces a consistent image series.
Course Content
How latent diffusion turns noise into an image, worked through in real numbers, and how the denoising network is built and trained. 2 lessons, about 40 minutes.
Take control of where and what diffusion paints with ControlNet, inpainting, outpainting and image-to-image, then teach a model new subjects and styles with LoRA, DreamBooth and textual inversion. 2 lessons, about 30 minutes.
Fill in missing frames with optical flow and learned interpolation, and choose and use commercial video generators by the trade-offs that actually matter for your work. 2 lessons, about 35 minutes.
Build a tool that generates a coherent image series, with a pinned style, one named axis of variation and a number that measures consistency. 1 lesson, about 20 minutes.