LC Lighting Control V2 🔦

Most "relighting" in ComfyUI means pushing your picture back through a diffusion model - IC-Light and friends - and hoping the identity, the logo on the shirt and the stitching survive. LC Lighting Control V2 is narrower and much safer: give it your image plus a normal map and a depth map and it repaints the light mathematically. No sampler, no second pass, no re-generation. The pixels you spent an hour on stay the pixels you spent an hour on.

The tradeoff is honest: it can't invent light it can't compute, and it has no idea a surface is brushed aluminium. But for "this render is lit from the wrong side", "this product shot needs a window light", or "I need a rim light behind the subject" - it's instant, deterministic, and the face can't drift.

V1 (LC Lighting Control 🔦) still ships and still works - V2 is a separate node, not a replacement.

How it works

The normal map is decoded from RGB to unit vectors, and depth is normalised to 0 = near, 1 = far with a percentile stretch (auto-flipped if the map runs the other way). Each light adds a diffuse term from the normals: surface facing the light goes bright, surface turned away doesn't.

Cast shadows are where V2 earns its name. Depth becomes a height field and rays are marched along the light direction with elevation included, so shadow length follows the light's height and shadows soften with distance from the occluder. Depth edges get snapped toward the image's real edges first, and if you feed the optional subject mask, the subject lifts off the background to throw a clean shadow onto it.

V1 was shading. V2 is shading plus geometry.

Inputs and outputs

Required: image, normal_map, depth_map. Optional: mask (subject matte - LC's own background-removal or person-mask nodes, or SAM).

Start with preset (Rembrandt, Soft window left/right, Split (hard side), Top light, Under light, Rim, Golden hour, Campfire, Cyberpunk, Key + fill, Flat front). It sets the lights and shadows; touch anything and it flips to custom. The three you'll actually turn:

• blend - the safety valve. 0 is the original, 1 is fully relit, and 0.7–0.9 is where this looks like a photograph rather than a filter.
• light1_x / light1_y / light1_z - where the light is. X left/right, Y up/down, Z how frontal. Or just drag the handle on the node's light stage (Shift+drag or wheel for Z).
• shadows plus shadow_amount / shadow_blur - soft, hard or off. Hard is a sun or a bare bulb; soft is a window.

An advanced toggle reveals the fine controls (relief, depth_falloff, shadow_height, shadow_length, shadow_softness, subject_distance, mask_dome). The one to know is relief - how deep the face is in 3D, and what decides whether a nose casts a nose-shaped shadow or nothing. shadow_height sets shadow length independently of the light's Z.

Outputs: image (relit), shadow_mask (white = lit, black = in shadow - genuinely useful as a matte), and light_layer (the normalised illumination buffer, handy for compositing).

Install

Manager → ComfyUI LC123 Nodes, or:

cd ComfyUI/custom_nodes
git clone https://github.com/lonecatone23/ComfyUI_LC123_nodes.git

Restart. No pip install - it's NumPy and PyTorch, no SciPy or OpenCV.

The maps are on you. Depth Anything V2 (Kijai's or comfyui_controlnet_aux) is still the default depth preprocessor in 2026, V3 notwithstanding, and BAE/DSINE normal maps come from ControlNet Auxiliary Preprocessors. The README points at both, plus remBG for a subject mask. The example workflow is workflows/Lonecats Lighting Control Module V2.json.

Common issues

The lighting looks inside-out - the light is dark and the shadow side is bright. Your depth map is inverted. Bright-equals-near instead of near-equals-dark is the usual culprit; flip it. The node tries to auto-detect this, but it can be fooled by a subject filling the frame.

Grey fringe or halo around the subject when a mask is connected. The README's own advice: turn the mask off first to confirm, then lower the blend or tighten the matte. High blend with a soft mask fringes.

Faces go plasticky and flat. relief is too low for the face - you get shading with no self-shadowing. Raise it. Conversely, if you're getting long cobweb shadows off every strand of hair, it's too high.

Shadows crawl into the background or vanish. background_shadow is the control: 1 assumes a wall right behind the subject, 0 assumes a distant background. With no mask connected the node guesses this from the far part of the depth map, and it can guess wrong on a busy scene.

Nothing looks real. Check fill: at its default 0.18 shadows stay off black, at 0 everything unlit goes pitch dark, and past 0.5 the frame flattens out. And if you want a material-aware relight - real specular behaviour, at the cost of identity drift - this isn't the tool. That's an IC-Light pass.