Home › Courses › Generative AI Fundamentals
From GANs to diffusion to flow matching. Build and customize generative AI systems
6 phases. 15 lessons. 15 labs. From GANs and VAEs through diffusion models, Stable Diffusion, and FLUX. Master ControlNet, LoRA, and DreamBooth for customization. Build video, audio, and 3D generation pipelines. Explore frontier techniques — flow matching, rectified flows, and visual autoregressive modeling (VAR). Evaluate with FID, CLIP score, and human preference. You build real generative pipelines and graduate with a portfolio that proves you can ship generative AI in production.
6 phases · 15 lessons · 15 labs · 1 capstone pipeline
Take Deep Learning from Scratch first — this course builds on it.
GANs from Scratch · VAEs from Scratch · DDPM (Denoising Diffusion Probabilistic Models) · Latent Diffusion & Stable Diffusion / FLUX · Text-to-Image Generation · ControlNet for Conditional Generation · LoRA Fine-Tuning · DreamBooth & Textual Inversion · Video, Audio & 3D Generation · Flow Matching & Rectified Flows · Visual Autoregressive Modeling (VAR) · Generative Model Evaluation (FID, CLIP Score)
PyTorch · Diffusers · FLUX · ComfyUI · ControlNet · Weights & Biases
It's building models that create — images, audio, video — from text prompts. From GANs that learn to generate through adversarial training to diffusion models that learn to denoise, you implement the full spectrum of generative AI techniques and build pipelines that create content.
You build generative AI systems — training models, fine-tuning for specific styles or subjects, and building production pipelines. When generation quality is poor, you diagnose whether it's the model architecture, the conditioning, the sampling strategy, or the training data.
Generative AI is transforming content creation — marketing, design, entertainment, e-commerce. Companies need engineers who can build and customize generative models, not just use Midjourney. The engineer who implemented DDPM from scratch is the one who can fine-tune models for specific brands and build production generation pipelines.
It is not a Midjourney tutorial. It is not about prompt engineering for DALL-E. It is about building the generative models themselves — GANs, VAEs, diffusion — from scratch, and customizing them for specific use cases. If you want to understand how AI creates, this is the course.
Yes — this course uses PyTorch, CNNs, and training loops throughout. Deep Learning from Scratch builds those foundations; this course applies them to generative modeling.
Yes — training diffusion models and GANs requires a GPU. The course includes guidance on using free GPU resources (Google Colab, Kaggle) and cloud GPU rental (RunPod, Lambda Labs).
ML & AI Engineering covers generative AI in 4 lessons — enough to use pre-trained models. This course spends 15 lessons implementing GANs, VAEs, DDPM, and flow matching from scratch, fine-tuning with LoRA and ControlNet, and building video, audio, and 3D generation pipelines. It is the difference between using generative AI and building it.
Yes. The course covers flow matching and rectified flows as the modern alternative to diffusion sampling, visual autoregressive modeling (VAR) as a new generation paradigm, and references current models including FLUX, Sora, Veo, Kling, Suno, and 3D Gaussian Splatting. The field moves fast — this course is updated to reflect the state of the art.
A generative AI pipeline that creates images, audio, and video from text — with fine-tuned models, custom styles, and production-ready quality. Plus the ability to customize any generative model for specific use cases.
LoRA (Low-Rank Adaptation) · Fine-Tuning
Deep Learning from Scratch · Transformers & LLMs from Scratch · Computer Vision Engineering · Multimodal AI Systems · ML & AI Engineering · Generative Media Engineering
Create a free account — the opening phases of 24 of 30 courses are free, no credit card. Or see Pro pricing.
All courses · Pricing · About · FAQ · Glossary