Latent space

Latent space is the compact, internal representation a diffusion model works in while generating an image. Instead of building the picture pixel by pixel from the start, the model first shapes a much smaller, compressed pattern of numbers, then decodes that pattern into the full-resolution image you see at the end.

Working in this compressed space is what makes modern image models fast enough to run on consumer hardware, since there is far less data to process at each generation step.

In tendre.AI, this is the internal mechanism behind every generation, handled automatically so you only ever interact with the prompt and the final image.