COMFY PAL embeds the full LensCowboy PAL viewport inside ComfyUI as a custom node. Use it for spatial layout (where things go in 3D), camera setup (position, lens, DOF, animation), and multipass render output that feeds directly into your AI graph.

The node exposes five image passes ready to drop into KSampler, ControlNet (depth/normal), Video Combine, compositing nodes, and more:

• Beauty — rendered colour pass (available on all tiers, including Free)
• Depth — z-depth, perfect for depth ControlNet (COMFY PAL or any SaaS tier)
• Normal — surface normals, for normal ControlNet (COMFY PAL or any SaaS tier)
• Alpha — matte for compositing (COMFY PAL or any SaaS tier)
• ID matte — per-object masks (COMFY PAL or any SaaS tier)

Free tier: beauty pass only, capped at 512px. Depth / normal / alpha / ID matte return blank tensors — unlock the full set with COMFY PAL ($7/mo) or any LensCowboy SaaS subscription.

Rendering happens locally on your machine via wgpu (Metal / Vulkan / DirectX 12). No cloud infrastructure, no 15MB base64 caps, no LensCowboy compute charges. Your GPU, your render.

Via git clone

cd ~/ComfyUI/custom_nodes
git clone https://github.com/Lenscowboy/comfyui-lenscowboy-pal.git
pip install -r comfyui-lenscowboy-pal/requirements.txt
# restart ComfyUI, then search "PAL" in the node picker

Via ComfyUI Manager

Once published to the Comfy Registry: search comfy_pal in Manager, click install, restart. (Registry publish pending — use git clone for now.)

Local renderer (optional but recommended)

If you want direct Python rendering (bypasses the iframe round-trip, no 15MB cap, runs on your GPU), install the renderer deps:

pip install -r comfyui-lenscowboy-pal/pal_renderer/requirements.txt

Dependencies: pygfx, rendercanvas, gltflib, trimesh, assimp-py (for FBX), imageio, numpy.

Free tier (no account needed)

1. Place the PAL Layout node in your graph
2. Click Open Viewport — a full-screen modal with the PAL scene editor opens
3. Drag a model in, or connect Load3D / Hunyuan3D / Tripo to the model_3d input
4. Position the camera, hit Render
5. Queue Prompt — the beauty_pass output flows to your next node

Subscriber tier

After subscribing, paste your API key into the node's lc_api_key input. All passes unlock; resolution goes up to 2048px; local renderer works end-to-end.

Inputs

Outputs

Beauty

Standard lit render. Feeds cleanly into KSampler as an img2img starting point, or into edit-style image models as a structural reference. Default materials are neutral grey — AI restyling downstream replaces the look anyway, which is the point.

Depth

Z-depth from the camera. Drop straight into a Depth ControlNet conditioning chain for shot-accurate spatial control. Normalised [0, 1]; near objects darker, far objects lighter by default.

Normal

RGB-encoded surface normals. Perfect for Normal ControlNet — preserves surface detail and lighting direction across AI restyles.

Alpha

Single-channel matte of rendered geometry vs background. Use it to composite the AI-restyled beauty pass over a different background, or to mask out specific regions before passing to downstream nodes.

ID matte

Per-object colour-coded mask. Each object in the scene gets a unique flat colour; use Image Mask by Color or similar to isolate individual objects for per-element processing.

Set use_local_renderer=True on the node to bypass the iframe and render directly in Python via pygfx on your GPU.

Benefits

• No iframe round-trip — skip the postMessage dance; queue and go
• No 15MB base64 cap — direct numpy tensor handoff
• Runs on your GPU — Metal (macOS), Vulkan (Linux), DirectX 12 (Windows)
• Zero cloud cost — nothing hits LensCowboy infra

Tradeoffs

• Requires the optional pal_renderer/requirements.txt deps installed
• Beauty, alpha, and normal passes are available now. Depth and ID matte are on the active roadmap — use the iframe render path for those in the meantime
• Uses the camera from your PAL viewport session (open viewport once to set camera, then local renderer uses it)

Upstream ComfyUI 3D generators connect cleanly: Hunyuan3D, Tripo, Meshy, Rodin, Load3D, Load3DAnimation, TripoSR. The node walks the graph on viewport open to auto-load any upstream model, so you don't need to queue once before seeing your geometry.

Depth-guided img2img

Drop the beauty into KSampler as init image, run depth pass into Depth ControlNet for spatial faithfulness. Result: AI-generated image that respects your layout.

PAL → beauty_pass → KSampler (init)
PAL → depth_pass  → ControlNet (depth_strict)
       → KSampler (conditioning)

Multi-angle generation

Same scene, multiple camera positions, batched through your image model. Each camera preset renders a fresh beauty + depth, AI samples stay consistent via seed and pose conditioning.

Animated video via Video Combine

Render a frame sequence, feed into Animatediff or your video restyle chain, terminate in Video Combine for an mp4.

Per-object control

Use id_matte_pass to isolate an object, process differently (inpaint, colour-swap, relight), then recombine with the original beauty.

COMFY PAL at $7/mo or $49/yr unlocks everything you need for solo ComfyUI work: all passes, full resolution, no caps.

SaaS Creator+ (see tiers) adds the Daily pipeline, cloud renders, project/shot libraries, and Drive output.

Docs for the SaaS side:

• Full PAL documentation — the SaaS-side PAL viewport in detail
• Pipeline documentation — how Daily / Weekly / PAL runs work
• LCBE documentation — script → breakdown → bid workflow
• Platform overview — how all the modules fit together

"Upload failed: 401" in the render dialog

Your current plan doesn't include the feature you just tried to use (typically Save to Drive or Export to Pipeline). The dialog should now lock these options up front and show an upgrade modal. If you're seeing this error, hard-refresh the browser (⌘⇧R / Ctrl+Shift+R) to pick up the latest iframe build.

Model doesn't appear in the viewport object list

The node auto-loads 3D models from upstream nodes (Load3D / Hunyuan3D / Tripo etc.) when the viewport opens. If nothing appears, check:

• Is the model connected to model_3d, GLB, or OBJ?
• Hard refresh the browser so the latest JS loads.

FBX loads but renders black

Embedded FBX textures sometimes fail to decode (common with proprietary image types). The viewport auto-swaps broken materials for neutral gray so geometry stays visible. For textured FBX, re-export as GLB from your DCC — much more portable.

Local renderer: "ImportError: gltflib required"

Install the renderer deps: pip install -r pal_renderer/requirements.txt inside the node directory.

Local renderer depth + ID matte show as black

Those two passes are on the active roadmap for the local renderer. Use the iframe render path for depth and ID matte for now — those work end-to-end today.