All glossary terms
AI & Generation

What is ControlNet?

A diffusion model add-on that conditions generation on a structural input — pose, depth map, edge map, or layout — alongside the text prompt. Enables precise control over AI output.

In detail

ControlNet adds a second input channel to a diffusion model. Alongside the text prompt, the model receives a structural reference (pose skeleton, depth map, Canny edges, segmentation mask, scribble) that constrains the spatial layout of the output. Common ControlNet types: Canny (edge-following), Depth (3D structure), OpenPose (human pose), Tile (preserve fine detail), Scribble (rough sketch guidance). Textile applications include using ControlNet Tile to preserve user-uploaded sketch structure while applying AI style, ControlNet Canny to keep motif outlines stable across colorway variations, and ControlNet Depth for 3D fabric drape rendering.

Example

A designer sketches a rough floral on paper, scans it, and runs ControlNet Canny + SDXL with prompt 'watercolor peony, dusty pink and sage'. The output preserves the exact composition of the sketch (Canny edges) but renders it in detailed watercolor — much more controllable than text-only generation.

Related terms

Diffusion model
A class of generative AI models that produce images by iteratively denoising random Gaussian noise into coherent imagery. The dominant architecture for AI image generation in 2026, including textile pattern AI.
Stable Diffusion XL
Stability AI's open-source 2.6 billion parameter diffusion model released July 2023. The dominant model for textile pattern AI as of 2026, used by most production textile-AI platforms.
FLUX
Black Forest Labs' 12 billion parameter diffusion model released August 2024. Produces higher-fidelity generations than SDXL at higher GPU cost. Used by textile AI platforms for high-detail generation.
LoRA
Low-Rank Adaptation — a fine-tuning technique that adds small trainable matrices to a frozen base diffusion model. Lets users customize Stable Diffusion or FLUX for specific aesthetics with minimal compute.
Image-to-image
A diffusion model mode that takes an input image and a prompt, producing an output that combines the input's structure with the prompt's style. The opposite of text-to-image.

Go deeper