Hy3D Render MultiView Depth

Hy3DRenderMultiViewDepth is Hy3DRenderMultiView's quieter sibling. Same six-view ring, same camera defaults, same renderer philosophy - but instead of the normal and position maps that feed the paint stage, it renders depth maps. Same renderer, different output, different job.

Where the normal/position renderer is the control-image factory for texturing, this one is a tool for understanding and inspecting your mesh - and for building your own conditioning if you're doing something clever that Hunyuan's default pipeline doesn't cover. It's the node you reach for when you want to look at the geometry, not when you want to paint it.

What it does

Inputs: a TRIMESH, render_size and texture_size (both default 1024), and an optional camera_config to override the six views. Outputs are just:

• depth_maps (IMAGE) - six grayscale depth images, one per view (replicated to RGB so standard image nodes will touch them)
• masks (MASK) - the per-view alpha masks

The six cameras are the same azimuth 0/90/180/270 + top/bottom set the pack uses everywhere, and it even calls unload_all_models() before rendering, so it plays nice in a graph that's been holding the shape model in VRAM.

When you'd actually use it

Honestly, not often, and the article can be short because the node is too. The stock Hunyuan3D texturing path doesn't consume depth maps - Hy3DSampleMultiView wants normals and positions. Depth-from-mesh is useful for:

• Debugging - a quick depth render tells you at a glance whether the topology is watertight or full of holes.
• Custom conditioning - if you're building your own multiview diffusion setup that wants depth control (very much the advanced end of this pack).
• 2D downstream work - depth renders can feed a depth-aware upscaler or img2img pass elsewhere in your graph.

Install

Pack-level: Manager search "Hunyuan3DWrapper", or git clone https://github.com/kijai/ComfyUI-Hunyuan3DWrapper into custom_nodes, then pip install -r requirements.txt. Like its sibling it's part of the texture-gen renderer, so on Windows it wants the matching prebuilt custom_rasterizer wheel (wheels/custom_rasterizer-0.1.0+torch260.cuda126-cp312-cp312-win_amd64.whl for the current portable, or whatever matches your build).

Common issues

• The depth maps look washed out - depth is encoded per-view, so a small object floating in a big frame reads as mostly white. Not a bug, just a convention thing.
• Cameras don't match what you sampled with - same trap as the normal renderer: if you override cameras in Hy3DCameraConfig, keep every multiview node on the same config.