Skip to main content
ControlNets are integrated guidance networks that help StreamDiffusion understand the structure of your input. You can think of them as glasses for your AI model’s conceptual eyes.

By using preprocessors, Daydream can take your input before it’s fed into the AI model, and extract key characteristics, such as:
  • Edges
  • 3D Depth Position
  • Color Palette
  • Details and Patterning
These characteristics act as the model’s “reference map,” guiding StreamDiffusion toward outputs that stay aligned with your original image. Each ControlNet also includes a strength parameter, which determines how strictly the model follows the extracted characteristics. A higher strength forces the model to stay very close to your input’s structure, while a lower strength gives the model more freedom to reinterpret, stylize, or transform the image. Just like how people need the right prescription to see clearly, an AI model needs the right ControlNet settings to understand what matters in your input. By choosing the appropriate preprocessor, ControlNet, and strength you ensure the model has the structural “vision” it needs to produce a strong, reliable output.

​
Types of ControlNets

SDXL, SD15, SDTurboCanny Pninput, preprocessor, output; with prompt: blueberriesCanny extracts clean edge outlines from your input, helping the model accurately follow its shapes and contours.
SDXL, SD15, SDTurboDepth Pninput, preprocessor, output; with prompt: blueberriesDepth analyzes the distance and structure within your scene, helping the model maintain a believable sense of 3D space.
SDXL, SD15Tile Pninput, preprocessor, output; with prompt: blueberries
Tile captures fine textures and high-frequency details, helping the model preserve crispness during generation.
SDTurboColor Pninput, preprocessor, output; with prompt: blueberries
Color maps broad color regions from your input, helping the model keep the overall palette and tone aligned.
SDTurboPose Pninput, preprocessor, output; with prompt: sun wukongOpenPose detects body joints and limb positions, helping the model stay consistent with the character’s pose.
SDTurboHED extracts soft, stylized linework, helping the model follow your input’s gentle contours and artistic outlines. The edge extraction is considerably less than Canny.Multi-ControlNets

​
Multi ControlNets

Multi-ControlNets let you use several ControlNets at once, giving the model multiple types of structure to follow. Each ControlNet captures a different characteristic from your input, so combining them creates a more complete guide. For example, Canny + Depth works great because Canny gets the edges, while Depth handles 3D placement. This feature is exclusive to Daydream’s backend, since their server pipeline supports running multiple ControlNets together in real time. Multi Control Net Ezgif Com Optimize Gi low value single controlnet vs multi controlnet using depth and canny; prompt: blueberries Using Multi-ControlNets helps StreamDiffusion produce outputs that stay accurate, detailed, and true to your input.