ControlNet Depth Maps Best Practices

last edited

ControlNet depth maps are grayscale images encoding camera distance: brighter areas are closer, darker areas are farther away. These best practices use depth guidance to preserve the 3D structure of a reference image during generation. Strength advice below comes from community practice, not official documentation.

ComfyUI setup

A typical workflow has four steps. Load an image, create its depth map, pass the map to a ControlNet, and generate the result. Testing at low resolution before the final render reduces processing time and compute use.

The official ComfyUI documentation covers node setup for Flux ControlNet workflows. See ComfyUI core for the base node set.

Community-tested parameters

Community practice favors strengths between 0.5 and 0.8. A common starting point is 0.6. Stability recommends a narrower range for its large SD3.5 models. That range runs from 0.7 to 0.8.

Depth and lineart form a common two-ControlNet pairing. Depth preserves spatial relationships, while lineart reinforces outlines. The pairing suits architecture, products, and character design. Multiple ControlNets need balanced strength settings. IP-Adapter + ControlNet combined workflows describes a production-oriented setup.

Model support

The model lllyasviel/sd-controlnet-depth was trained on Stable Diffusion v1-5 with depth maps from an estimator. It is meant for SD1.5 only.

Stable Diffusion received a large-model depth ControlNet on Nov 26, 2024. The release covers version 3.5. It uses depth maps generated by a depth estimation model.

Flux received a tools release on Nov 21, 2024. It includes depth and edge models. Adoption patterns are summarized in what the SD community still uses.

In Feb 2026, alibaba-pai published a FLUX.2-dev Fun-Controlnet-Union model that lists depth among its supported conditions.

Sources