daBongo LMS AI Training Courses

Lesson Content

What ControlNet does.

ControlNet is a conditioning technique that provides structural guidance to the generation – the model follows the structural information extracted from a reference image (pose, depth map, edge structure) while generating new visual content from the prompt. This separates the visual content (from the prompt) from the structural guidance (from the ControlNet reference), enabling precise control over composition and pose without constraining the visual style.

ControlNet Pose.

ControlNet Pose extracts the human pose skeleton from a reference image and uses it to guide the body positioning of generated human subjects. The generated person will adopt the same pose as the reference – different appearance, clothing, and style from the prompt, but the same body positioning and stance. Useful for: consistent character poses across a series, matching generated characters to specific reference poses, and action and dynamic pose generation.

ControlNet Depth and Structure.

ControlNet Depth extracts the depth map from a reference image – the spatial depth relationships between foreground, midground, and background. Using this depth map as a ControlNet guide produces generations that maintain the same spatial depth structure as the reference. ControlNet Canny (edge detection) extracts compositional outlines for structural guidance. These are useful for: maintaining scene architecture and spatial relationships across variations, using architectural or environment reference for consistent spatial layout.

ControlNet strength and balance.

Like image guidance, ControlNet has a strength control – balancing how strictly the generation follows the structural guidance vs. how much creative freedom the prompt has. For precise pose or structure matching: higher strength. For loose structural inspiration: lower strength.