| images | IMAGE | | Input video frames (B,H,W,3) float32 [0,1]. |
| width | INT | 83264–4096 | Output canvas width. |
| height | INT | 48064–4096 | Output canvas height. |
| enable_depth | BOOLEAN | true | Run the depth pass. Requires at least ONE depth source wired. |
| enable_pose | BOOLEAN | true | Run the pose pass. |
| enable_canny | BOOLEAN | true | Run the canny pass. |
| canny_threshold1 | INT | 1000–500 | Canny lower hysteresis threshold. |
| canny_threshold2 | INT | 2000–500 | Canny upper hysteresis threshold. |
| canny_aperture | COMBO | 3 | Sobel aperture for Canny (odd: 3/5/7). |
| depth_colorize | BOOLEAN | false | If true, colorize grayscale depth with INFERNO colormap. Skipped when external_depth_map is already RGB. |
| depth_invert | BOOLEAN | false | Invert depth (1 - depth). Use when source produces 'far = bright' but you want 'near = bright' (typical ControlNet expectation). |
| pose_detection_threshold | FLOAT | 0.050–1 | YOLO confidence threshold (only used when posemodel is wired). |
| pose_draw_threshold | FLOAT | 0.300–1 | Per-keypoint score threshold for drawing the skeleton. |
| combined_layout | COMBO | horizontal_3 | Layout for the combined output. grid_2x2 = depth | pose // canny | original. |
| depth_backend | COMBO | auto | Which depth backend to use. 'auto' tries: external_depth_map -> any wired loader -> built_in_midas. 'built_in_midas' makes the node fully self-contained (downloads MiDaS small via torch.hub on first use, ~80MB). |
| enable_normal | BOOLEAN | true | Compute Sobel-from-depth NORMAL map. No model required (uses depth pass output). |
| normal_strength | FLOAT | 1.00.1–10 | Scales the Sobel gradients before normalisation. Higher = stronger normal contrast. |
| blend_mode | COMBO | weighted_avg | How to combine depth+pose+canny+normal into blended_map. linear_dodge=additive (sharp), screen=highlight-safe, channel_split=Fun-Control (R=depth/G=canny/B=pose). |
| depth_weight | FLOAT | 1.000–4 | Weight of depth in blended_map. |
| pose_weight | FLOAT | 1.000–4 | Weight of pose in blended_map. |
| canny_weight | FLOAT | 1.000–4 | Weight of canny in blended_map. |
| normal_weight | FLOAT | 0.500–4 | Weight of normal map in blended_map. |
| external_depth_mapopt | IMAGE | | Pre-computed depth IMAGE batch from ANY upstream node. Highest priority. |
| damodel_v2opt | DAMODEL | | DepthAnything V2 model bundle from kijai/ComfyUI-DepthAnythingV2 (DownloadAndLoadDepthAnythingV2Model). Models: ComfyUI/models/depthanything/. |
| da3_modelopt | DA3MODEL | | DepthAnything V3 config bundle from PozzettiAndrea/ComfyUI-DepthAnythingV3. Use the V3 pack's Inference node and feed its IMAGE output into external_depth_map. Models: ComfyUI/models/depthanything3/. |
| depthcrafter_modelopt | DEPTHCRAFTER_MODEL | | DepthCrafter bundle from akatz-ai/ComfyUI-DepthCrafter-Nodes. Temporally consistent video depth. Models: ComfyUI/models/depthcrafter/. |
| depth_pro_modelopt | DEPTH_PRO_MODEL | | Depth-Pro bundle from spacepxl/ComfyUI-Depth-Pro. Metric depth. Models: ComfyUI/models/depth/ml-depth-pro/. |
| posemodelopt | POSEMODEL | | From ONNX Detection Model Loader (V2) or animal-pose loader. Used if enable_pose=True AND no external_pose_map wired. |
| external_pose_mapopt | IMAGE | | Pre-rendered pose map from any upstream node (e.g. Fannovel16/comfyui_controlnet_aux DWPose / OpenPose / AnimalPose). Highest priority for pose. |
| depthcrafter_stepsopt | INT | 51–100 | DepthCrafter only: diffusion inference steps. |
| depthcrafter_guidanceopt | FLOAT | 1.00.1–10 | DepthCrafter only: classifier-free guidance. |
| depthcrafter_windowopt | INT | 1101–200 | DepthCrafter only: temporal window size. |
| depthcrafter_overlapopt | INT | 250–100 | DepthCrafter only: window overlap. |
