ControlNet

Install and use pipeline-specific ControlNet guidance in 360 Hextile

Last updated: 2026-08-24

ControlNet guides diffusion with structure derived from an image, such as edges, depth, a pose, or a sketch. Support is pipeline-specific: selecting another pipeline can change the available types and model downloads.

Supported pipelines

Pipeline Types available
SDXL Canny, Canny Small, Depth, Soft Edge
SD 2.1 Canny, Depth, OpenPose, Scribble
SD 1.5 Canny, Depth, OpenPose, Scribble, MLSD, Segmentation, Normal, Lineart
Qwen-Image (Alpha) Union model: Depth, Canny, OpenPose, Soft Edge, Scribble

SD 3.5, FLUX.1, FLUX.2, TeleStyle, Chroma1-HD, and Z-Image Turbo do not expose ControlNet in the current app.

Install a ControlNet model

ControlNet models do not download silently during a render.

  1. Select a supported diffusion pipeline.
  2. Open ControlNet from the module stack or Add Module.
  3. Select a type.
  4. If it is not installed, use the install action. The model enters the same Model Manager download queue used by pipelines and LoRAs.
  5. Wait for Installed, then enable ControlNet.

Qwen-Image uses one Union download for all five of its modes. SDXL, SD 2.1, and SD 1.5 use a separate managed model for each type. If a selected model is missing, the render fails with an install instruction; it does not fetch hidden files in the background.

Choose a guidance type

  • Canny preserves strong edges. Canny Small is SDXL's smaller, lower-VRAM Canny model.
  • Depth guides spatial layout from an estimated or attached depth map.
  • OpenPose extracts body and hand pose where the selected pipeline supports it.
  • Scribble follows loose sketch-like edges.
  • Soft Edge follows broader organic contours. It is available on SDXL and Qwen-Image Alpha.
  • MLSD extracts straight lines for architecture. It is available on SD 1.5.
  • Segmentation, Normal, and Lineart are SD 1.5-only modes.

Depth source

Depth guidance has two sources:

Mode What it does
Auto Estimates depth for this pass. The estimate is transient and is not saved as a graph channel.
Attached Channel Uses a reusable depth channel on the render.

Use an attached channel when the depth must also drive fog, export, or later passes. If the chosen channel is missing or stale, the app asks you to regenerate or attach depth instead of silently falling back to Auto.

Settings

  • Strength controls how strongly the structural input affects diffusion. Start near the pipeline's displayed default, then lower it for more prompt freedom or raise it for tighter control.
  • Guidance Start is the fraction of diffusion steps before ControlNet starts.
  • Guidance End is the fraction after which ControlNet stops.
  • Canny Low / High Threshold choose which edges survive preprocessing.
  • Depth Boost increases or reduces contrast in depth guidance.

The current configuration stores selected models in diffusion.controlnet.models[], with shared guidance_start, guidance_end, and preprocessor settings. Saved Workflows keep this configuration, but a Workflow still needs a compatible active pipeline and installed ControlNet model.

Good first checks

  • If the result is too rigid, lower Strength or end guidance earlier.
  • If structure is weak, raise Strength or lower the Canny thresholds.
  • If a Workflow changes pipeline, reopen ControlNet and confirm that its selected type exists for the new pipeline.
  • If Depth does nothing, verify the attached channel or switch deliberately to Auto.

See Channels, Modules & Compatibility, and the Pipeline Guide.

Esc
Searching...
No results found.
Type to search the documentation
Navigate Select Esc Close

We use cookies for analytics to understand how visitors use our site. Privacy Policy