Text encoder
In an image or video generator, the text encoder is a separate language model that turns your prompt into numbers the generator can use. It adds to the download and to memory use.
An image generator is usually a pipeline of parts. A text encoder reads the prompt, a diffusion transformer creates the picture in a compressed space, and an autoencoder (VAE) turns the result into pixels. The text encoder is often a full language model, so a model's parameter count can understate what you have to download.
MOSS-Image has a 5B generator, but its repository is 27.2 GB, of which 16.8 GB is the Qwen3.5-9B text encoder. Z-Image's 6B generator comes with an 8 GB text encoder, Qwen-Image-2.1's 8B encoder makes its full pipeline about 33 GB, and MiniMax H3-Image's 66.7 GB encoder is nearly as large as the rest of its files. Anima is the opposite case, with a 1.19 GB encoder in a 5.6 GB pipeline. When you plan memory, add the encoder to the generator, and remember that many tools can offload or unload it once it has read the prompt.
Related terms
- 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.
- Parameters Parameters are the numbers inside a model that are adjusted during training; their count, such as 7B or 70B, is the usual measure of a model's size.
- VRAM VRAM is the memory on a graphics card; it decides which models a GPU can hold, because the weights must fit in memory for the model to run quickly.