Diffusion model

A diffusion model generates images or video by starting from random noise and removing the noise step by step until a picture emerges.

  • Models and architecture
  • Updated

Most open image and video generators, including FLUX, Qwen-Image, Wan and LTX, are diffusion models, often built on the transformer architecture. A text encoder turns your prompt into numbers that guide each denoising step.

Distilled versions are trained to need fewer steps. FLUX 3 Image Klein, for example, runs in 16 steps, while its Base version takes 50 and is meant for fine-tuning. Fewer steps means faster pictures, usually at a small cost in flexibility.

Related terms

  • Multimodal model A multimodal model can handle more than one type of data, such as text, images, audio or video, as input, as output, or both.
  • Distillation Distillation trains a smaller student model to imitate a larger teacher model, keeping much of its ability at a fraction of the size.
  • LoRA LoRA, short for low-rank adaptation, fine-tunes a model by training a small set of added weights instead of changing all of its parameters.

Keep reading

All terms