ControlNet AI Models

ControlNet models guide image generation with an extra structural signal such as pose, depth, edges, scribbles, segmentation, or tiles. They are useful when plain text prompts do not provide enough control over composition, layout, or subject placement.

Choose the ControlNet type based on the constraint you need. Pose models help with body position, depth models preserve scene geometry, edge models follow outlines, and tile models are often used for detail enhancement or upscaling workflows.

ControlNet types at a glance

• Pose (OpenPose) — control body and hand position from a skeleton image
• Depth — preserve scene geometry and spatial layout
• Canny / Edge — follow outlines and contours
• Scribble — generate from rough sketches
• Tile — enhance detail or upscale existing images
• Segmentation — guide generation with semantic regions

ControlNet quality depends heavily on the conditioning image. Before production use, test whether the model follows the guide too weakly, too strongly, or introduces artifacts around edges and fine details. The best choice is usually the model that preserves the structure you care about without making the output look rigid.