SAM3DBody_Render - ComfyUI Built-in Node Documentation

Renders 3D body pose data into an image using a selectable style. The node accepts pose data from the SAM3D body tracker (MHR) or from an external Y-up rig such as Kimodo, and can composite the result over an optional background image (or a black canvas when none is provided). Available render styles include a shaded 3D mesh, a binary silhouette, 2D and 3D OpenPose-style skeletons, and SCAIL-style body capsules.

Inputs

Common Inputs

Parameter	Description	Data Type	Required	Range
`render_style`	Render mode. ‘mesh’ = 3D MHR mesh rasterized through the camera. ‘silhouette’ = binary mask of the mesh. ‘openpose_2d’ = flat 2D skeleton. ‘openpose_3d’ = OpenPose skeleton as flat-shaded 3D model. ‘scail’ = SCAIL 3D capsules. (default: “mesh”)	DYNAMIC_COMBO	Yes	”mesh" "silhouette" "openpose_2d" "openpose_3d" "scail”
`pose_data`	MHR pose data, or external Y-up rig pose data (KimodoSample). All render styles work for external rigs that carry OpenPose joint maps in their `_skeleton_override` (KimodoSample does).	MHR_POSE_DATA or KIMODO_POSE_DATA	Yes	—
`background`	Per-frame background. Omitted = black canvas.	IMAGE	No	—
`width`	Output width in pixels. 0 = use pose data’s native image_size. If only one of width/height is set, the other is derived preserving the original aspect. (default: 0)	INT	No	0 to 16384, step 8
`height`	Output height in pixels. 0 = use pose data’s native image_size. If only one of width/height is set, the other is derived preserving the original aspect. (default: 0)	INT	No	0 to 16384, step 8
`camera_info`	Free 6DOF camera override. When wired, the pose is re-projected through this camera (position/target/zoom/rotation/FoV) instead of the predicted one.	LOAD_3D_CAMERA	No	—

Mesh Inputs

These parameters appear when render_style is “mesh”.

Parameter	Description	Data Type	Required	Range
`shader`	Preset shader. ‘normals’ = current surface normal in camera space (OpenGL Y+ normal-map convention: +X→R, +Y→G, +Z→B). ‘rainbow’ = RealisDance style body-Y jet; the ‘rainbow_face_*’ variants override face verts with normal/per-region colors; ‘depth’ = linear gray. (default: “default”)	DYNAMIC_COMBO	No	”default" "normals" "rainbow" "rainbow_face_normal" "rainbow_face_semantic" "depth”
`rainbow_tilt_z`	Rotate rainbow jet axis around Z (forward). Differentiates left/right. Only available when `shader` is “rainbow”, “rainbow_face_normal”, or “rainbow_face_semantic”. (default: -35.0)	FLOAT	No	-90.0 to 90.0, step 0.5
`rainbow_tilt_x`	Rotate rainbow jet axis around X (right). Differentiates front/back. Only available when `shader` is “rainbow”, “rainbow_face_normal”, or “rainbow_face_semantic”. (default: 0.0)	FLOAT	No	-90.0 to 90.0, step 0.5
`opacity`	Mesh alpha over the background image, or over black when none is connected. (default: 1.0)	FLOAT	No	0.0 to 1.0, step 0.01
`person_palette_falloff`	Per-person desaturation toward white: track k gets a (1 - falloff^k) pastel mix (SCAIL ‘softer second person’). 1.0 = off. (default: 0.6)	FLOAT	No	0.1 to 1.0, step 0.05
`region`	’hands_only’ filters faces via the precomputed `hand_vert_mask` (LBS weights against canonical hand KPs) — isolates the hand mesh for debugging. Falls back to full mesh if the mask is missing. (default: “full_body”)	COMBO	No	”full_body" "hands_only”

Silhouette Inputs

When render_style is “silhouette”, the node renders a binary mask of the 3D mesh. This mode has no additional parameters.

OpenPose 2D Inputs

These parameters appear when render_style is “openpose_2d”.

Parameter	Description	Data Type	Required	Range
`marker_radius_px`	Body keypoint dot radius (px). (default: 4)	INT	No	1 to 32, step 1
`stick_width_px`	Body limb ellipse half-width (px). DWPose default = 4. (default: 4)	INT	No	1 to 32, step 1
`limb_alpha`	Per-limb alpha. DWPose default = 0.6. (default: 0.6)	FLOAT	No	0.0 to 1.0, step 0.05
`face_style`	’full’ = all face landmarks (sapiens-238 if present, else rig-fallback ~30). ‘eyes_mouth’ = rig-fallback subset (~12 dots: eyes + outer lips only). ‘disabled’ = no face dots. (default: “disabled”)	COMBO	No	”disabled" "full" "eyes_mouth”
`hand_style`	Draw 21+21 hand keypoints + sticks. ‘disabled’ = no hands. ‘dwpose’ = solid-blue dots; ‘openpose’ = rainbow dots. (default: “disabled”)	COMBO	No	”disabled" "dwpose" "openpose”
`person_palette_falloff`	Per-person desaturation: track k blends toward white by 1 - falloff^k. Track 0 stays vivid; 1.0 disables falloff. (default: 0.6)	FLOAT	No	0.1 to 1.0, step 0.05

OpenPose 3D Inputs

These parameters appear when render_style is “openpose_3d”.

Parameter	Description	Data Type	Required	Range
`radius_m`	Limb capsule radius in meters (thin = stick-like). (default: 0.015)	FLOAT	No	0.004 to 0.1, step 0.001
`include_hands`	Draw 21+21 hand keypoints as 3D capsules. (default: True)	BOOLEAN	No	True or False
`person_palette_falloff`	Per-person desaturation: track k blends toward white by 1 - falloff^k. Track 0 stays vivid; 1.0 disables falloff. (default: 0.6)	FLOAT	No	0.1 to 1.0, step 0.05

SCAIL Inputs

These parameters appear when render_style is “scail”.

Parameter	Description	Data Type	Required	Range
`radius_m`	Capsule radius in meters (SCAIL reference: ~0.022 m). (default: 0.022)	FLOAT	No	0.005 to 0.2, step 0.001
`hand_style`	Composite 2D OpenPose hands on top of the 3D capsule body (matches SCAIL — no 3D hand capsules). ‘disabled’ = no hands. ‘dwpose’ = solid-blue hand dots; ‘openpose’ = rainbow dots. Sticks stay rainbow per-finger either way. (default: “dwpose”)	COMBO	No	”disabled" "dwpose" "openpose”
`face_style`	’full’ = all face landmarks (sapiens-238 if present, else rig-fallback ~30). ‘eyes_mouth’ = rig-fallback subset (~12 dots: eyes + outer lips only). ‘disabled’ = no face dots. (default: “disabled”)	COMBO	No	”disabled" "full" "eyes_mouth”
`person_palette_falloff`	Per-person desaturation: track k blends toward white by 1 - falloff^k. Track 0 stays vivid; 1.0 disables falloff. (default: 0.6)	FLOAT	No	0.1 to 1.0, step 0.05

Notes

If both width and height are 0, the output uses the pose data’s native image size. If only one of them is set, the other is derived while preserving the original aspect ratio. A connected background is resized to match the render resolution.
When camera_info is connected, the pose is re-projected through that camera instead of the predicted one.
In mesh mode, rainbow_tilt_z and rainbow_tilt_x are only available when shader is set to “rainbow”, “rainbow_face_normal”, or “rainbow_face_semantic”.
In mesh mode, when region is “hands_only”, the hand region filter requires the pose data to contain a hand vertex mask; if the mask is missing, the full mesh is rendered instead.
In scail mode, hands are always drawn as 2D overlays; there are no 3D hand capsules.
When the output resolution differs from the pose data’s native resolution, the openpose_2d marker and stick sizes are scaled proportionally.
If the background has fewer frames than the pose data, the last background frame is reused for the remaining frames.
The output contains one frame per input pose frame. If the pose data contains no frames, a single black image is returned.

Outputs

Output Name	Description	Data Type
`image`	The rendered frames: the pose data drawn in the selected render style, composited over the background when one is connected, or over black otherwise. One frame per input pose frame, returned as a single batched image.	IMAGE

This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub

Source fingerprint (SHA-256): 96556283cf07727e6b4bb3549537bf925ed771bab8607f65c93ab54a5f0e9ba5

​Inputs

​Common Inputs

​Mesh Inputs

​Silhouette Inputs

​OpenPose 2D Inputs

​OpenPose 3D Inputs

​SCAIL Inputs

​Notes

​Outputs