Image and video diffusion: how generation works and what it costs
Diffusion as iterative denoising, latent diffusion and why it is 48x cheaper per step, samplers and step counts, classifier-free guidance, text conditioning, the cost model (steps x CFG x per-step time), speed levers such as distillation and caching, and video extensions.
16 min·Verified 2026-11Pro+50 XP
Where LLMs generate one token at a time, diffusion models generate an entire image by refining noise over many steps. The cost structure is different: many forward passes of a big network, each fairly heavy and compute-bound.
Pro content
Keep reading with Inference Academy Pro
Unlock the rest of this lesson, plus every advanced lesson, hands-on lab, the larger playground models and completion certificates.