← Back to Dashboard
Module 05 · Days 29–35
Generative AI & Multimodal Systems

Images, audio, video, and models that reason across modalities. Understand diffusion models at an intuitive level, work with vision-language models, and ship a real multimodal product.

D29
Diffusion models: the theory behind Stable Diffusion
Latent spaces, denoising diffusion, and why DALL-E works. Explained as learned image compression plus controlled noise.
D30
Image generation in practice: SDXL, ControlNet & inpainting
Text-to-image, img2img, inpainting, ControlNet conditioning — practical image generation engineering.
D31
Vision-language models: CLIP, LLaVA & GPT-4o Vision
Models that see and speak. Zero-shot classification with CLIP, visual Q&A with LLaVA, structured extraction.
D32
Audio AI: Whisper, TTS & audio classification
Transcribe locally with Whisper, synthesise speech with Bark, classify audio events with Wav2Vec.
D33
Multimodal application architecture & latency budgets
Designing systems with mixed inputs. Latency budgets, streaming responses, cost control.
D34
Evaluating generated content: CLIP score, FID & safety
Measuring quality of generated outputs. FID, CLIP score, human preference scoring, and content safety.
🏗
🏗️ PROJECT: Multimodal AI App Combining 2+ Modalities
Build a deployed app combining at least 2 modalities. Meeting recorder + summariser or receipt scanner + tracker.