ComfyUI Depth ControlNet: Control Spatial Layout and Perspective

What is Depth ControlNet?

Depth ControlNet analyzes the distance between objects and the camera in your reference image, creating a depth map — a grayscale image where white areas are close and black areas are far. The AI uses this depth map to maintain the same spatial layout when generating a new image.

This is especially powerful for:

• Interior design — restyle a room while keeping the exact furniture layout
• Architecture — transform a building's style while preserving its structure
• Scene composition — maintain foreground/background relationships across style changes
• Product photography — keep depth-of-field relationships consistent

Prerequisites

Plugin Installation

ComfyUI doesn't include a depth preprocessor by default. You need the ComfyUI ControlNet Auxiliary Preprocessors plugin.

Install it via ComfyUI Manager (search for "ControlNet Auxiliary Preprocessors") or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/Fannovel16/comfyui_controlnet_aux
cd comfyui_controlnet_aux
pip install -r requirements.txt

Restart ComfyUI after installation.

Models

File Placement

ComfyUI/
├── models/
│   ├── checkpoints/
│   │   └── dreamshaper_8.safetensors
│   └── controlnet/
│       └── control_v11f1p_sd15_depth.pth

Building the Workflow

1. Load Image — load your reference photo
2. Zoe-DepthMapPreprocessor — generates a depth map from the image (from the Auxiliary Preprocessors plugin)
3. Preview Image — preview the depth map to verify quality
4. Load Checkpoint — loads the SD1.5 model
5. Load ControlNet Model — loads control_v11f1p_sd15_depth.pth
6. Apply ControlNet (Advanced) — injects depth information into conditioning
7. CLIP Text Encode (x2) — positive and negative prompts
8. KSampler → VAE Decode → Save Image

Key Parameters

Depth Preprocessor

Apply ControlNet (Advanced)

Prompt Tips for Depth Control

Include spatial and quality keywords for best results:

Spatial terms: depth of field, perspective, spatial layout, foreground, background

Quality terms: professional, high quality, detailed, realistic

Style terms: describe your target style clearly — modern minimalist interior, cyberpunk cityscape, watercolor landscape

Use Case: Interior Restyling

One of the most popular Depth ControlNet applications is restyling a room:

1. Take a photo of any room
2. Extract its depth map with the Zoe preprocessor
3. Write a prompt describing the new style: modern scandinavian living room, white walls, wooden furniture, natural light, professional interior photography
4. Generate — the new image keeps the exact room layout while applying the target style

Common Issues and Fixes

Generated image has weak spatial sense

• Verify the depth map preview — it should clearly show near (white) and far (black) areas
• Increase strength to 0.9–1.0
• Increase sampling steps to 25–30

Depth map looks flat or inaccurate

• Use a higher resolution setting on the preprocessor (768 or 1024)
• Ensure the input image has clear depth — flat product shots won't produce good depth maps
• Try a different preprocessor (MiDaS instead of Zoe, or vice versa)

Details are getting lost

• Lower the cfg value to 5–6 (too high cfg crushes subtle detail)
• Add more specific detail keywords to your prompt

Wrong style but correct layout

• The depth map is working — refine your text prompts
• Check that negative prompts aren't conflicting with your desired style

Related Guides

• ControlNet Overview — All ControlNet types explained
• Canny ControlNet — Edge-based structure control
• OpenPose ControlNet — Human pose control