ComfyUI Workflow: ControlNet Depth (FLUX)

This is the official Depth ControlNet example from the ComfyUI FLUX examples. The workflow takes a depth map of a source image (or generates one from a regular photo using the included preprocessor) and uses it as a structural guide for FLUX.1. The depth relationships stay locked: foreground stays foreground, midground stays midground, background stays background, but everything else is prompt-driven.

The FLUX.1 Depth LoRA is loaded as a controlnet rather than a full model. No separate depth ControlNet checkpoint needed. The LoRA provides the depth conditioning and FLUX.1 provides the base generation.

Key benefits

• Depth LoRA replaces the need for a separate ControlNet checkpoint, simpler model setup.
• Locked depth relationships, free style in the prompt.
• Useful for portrait compositing, scene restyling, and product photography variations.
• Official ComfyUI example, maintained alongside the core.

How it works

1. Download the workflow PNG from the linked GitHub URL and drag it into ComfyUI.
2. Get the model files: FLUX.1 fp8 base + FLUX.1 Depth LoRA. The LoRA is the depth controlnet in this case.
3. Provide a depth map: either load a pre-computed depth map image, or run the included preprocessor on a regular photo to extract one.
4. Write the prompt describing the style, lighting, and content. The depth map handles the spatial structure.
5. Queue the prompt. Generation runs at FLUX.1 base speed.

Depth is the right tool when you have a scene with strong foreground/background separation and you want to restyle it without losing the spatial layout. The depth LoRA approach is more efficient than running a full ControlNet model on top of FLUX.