Generative AI with Diffusion Models
NVIDIA DLI
October 30, 2026
The program
NVIDIA Deep Learning Institute's "Generative AI with Diffusion Models" course covers building an image generation model end to end. Through hands-on PyTorch labs on NVIDIA GPUs, I implemented a diffusion model step by step, starting with simple image denoising and ending with images generated from text descriptions.
- Issuer
- NVIDIA · Deep Learning Institute (DLI)
- Issued on
- October 30, 2026
- Format
- Online course with hands-on GPU labs (Jupyter, PyTorch, NVIDIA A10G)
- Validation
- Automatically graded hands-on assessment
- Signed by
- Howard Wright, Vice President, NVIDIA
- Certificate ID
The certificate
Curriculum
- 01From U-Net to diffusionDenoising FashionMNIST images with a U-Net
- 02Diffusion modelsStep-by-step noising over multiple timesteps and timestep embedding
- 03OptimizationsGroupNorm, GELU, Rearrange Pooling, sinusoidal embeddings, residual connections
- 04Classifier-Free Diffusion GuidanceClass-controlled generation, then color images (TF Flowers dataset)
- 05CLIPText-to-image generation of flowers
Final assessment
Train a conditional diffusion model on MNIST whose generated digits must be recognized with over 95% accuracy by a pre-trained classifier.
Skills gained
U-Net architecture
Downsampling and upsampling blocks, transposed convolutions, skip connections
Diffusion models (DDPM)
Variance (beta) schedule, forward diffusion, reverse diffusion and noise prediction
Model optimization
Group Normalization, GELU, Rearrange Pooling (einops), sinusoidal time embeddings, residual connections, checkerboard artifact fix
Classifier-Free Diffusion Guidance
Class conditioning, Bernoulli context masking, weighted reverse diffusion
CLIP
Text and image encodings, cosine similarity, text-to-image pipeline
GPU training
PyTorch, torch.compile, custom datasets preloaded on GPU
Skills covered
Technologies
Generative AI with Diffusion Models · NVIDIA · Deep Learning Institute (DLI) · ID VQrENWFlSCqh3FXs4hD3_w