Depth Map

Depth Map extracts a 3D distance map from a reference image and uses it to constrain the spatial layout of a generation. Foreground stays foreground, background stays background, and the rough shapes of objects stay where they were — but materials, color, lighting, and style are wide open to the prompt.

It is the right tool when you care about where things sit in space more than what they look like up close. Architecture, environment design, product photography, and room redesigns all live here.

Watch the tutorial: Depth ControlNet — Style Transfers with Spatial Precision

#What it does

Pocket gives you three depth-style preprocessors, each with a slightly different read on the scene:

Depth — a straight depth map. Preserves the spatial layout directly from the reference. Strong on foreground, looser on background detail.
Normal Bae — uses surface-normal maps to infer structure. Best for architecture and product design where surfaces and orientations matter. Tends to preserve detail across both foreground and background more evenly than Depth.
Segmentation — identifies and isolates distinct regions (sky, wall, furniture, person) and uses them as a layout mask. Use when you want clean object boundaries reusable in your composition.

Pick the variant that matches what you're trying to keep: shapes-in-space (Depth), surface-and-volume (Normal Bae), or region boundaries (Segmentation).

#How to start

Open Advanced Guidance and tap Depth Map.
Pick a reference image — a room, a building, a still life, a landscape.
Choose the depth variant (Depth, Normal Bae, or Segmentation) for what you need to preserve.
Write a prompt for the new look — different materials, different era, different lighting.
Generate.

#Prompt tips

Prompt for materials and atmosphere. "Marble walls, late-afternoon sun through windows, photoreal interior" — the structure already comes from the reference.
Specify era or genre when redecorating. "Mid-century modern," "brutalist concrete," "Art Deco gold and black" all work as clean style anchors.
Match the camera language — wide angle, fisheye, tilt-shift, top-down. The depth map carries layout but not lens character.
Strength tuning. Higher strength locks the layout strictly; lower strength lets the model reshape objects while keeping the general scene composition.

#Workflows

Room redesign. Photograph a room you want to redecorate, run Depth, and prompt for the style you're considering — Scandinavian, industrial, cottagecore. Furniture stays where it is; finishes change. Good for previewing renovations.

Architecture style transfer. Use Normal Bae on a photo of a building, then prompt for a different architectural era. Volumes are preserved; ornamentation, materials, and weathering shift.

Product photography variants. Shoot a product once on a plain background. Use Depth to reshoot it in twenty different scenes without moving the product.

Environment concept art. Start from a real landscape photo. Prompt for "alien planet," "post-apocalyptic," or "fantasy realm" and watch the same valley turn into a different world.

Same scene, different time. Take one reference, generate "morning," "noon," "sunset," "stormy night." The layout is constant; the mood transforms.

#Tips

Cleaner references give cleaner depth maps. Cluttered scenes can confuse the preprocessor.
Segmentation is the strongest choice when you want to reuse the outline of objects rather than their volumes.
Depth Map pairs nicely with strong style prompts because the structural fidelity gives the style somewhere to live.
Push steps up to 30+ for architectural scenes — extra steps help the model resolve clean straight lines.