Atlas · GenAI 2026

Diffusion Models

Diffusion models & Flow Matching

conceptPeak: 2023Generative ArchitecturesAI consensus: 2/3

Prerequisites

hardDeep Learning
Diffusion models use U-Nets or DiTs with noise schedules, score matching, and denoising — all advanced DL concepts
mediumProbability Theory
Diffusion theory involves ELBO, KL divergence between forward/reverse processes, and variational bounds

Recommended reference

Prince, S. (2023) Understanding Deep Learning, Ch.18 — free at udlbook.com; covers diffusion from first principles. Paper: Ho et al. (2020) 'Denoising Diffusion Probabilistic Models'

Notes from AI deep research

Anthropic Opus

SD, FLUX, Sora. DiT = kluczowa architektura. Flow Matching upraszcza trening

Google Deep Think

Generowanie obrazu, 3D, wideo [G#29]

Related skills

→ is subcategory of: Deep Learning(3/3)
→ is subcategory of: GenAI(3/3)
← is an instance of: Hugging Face Diffusers(0/3)
← is an instance of: Stable Diffusion(0/3)
← is subcategory of: Image Generation(0/3)