SV3D_Conditioning - ComfyUI Built-in Node Documentation

SV3DConditioning prepares conditioning data for 3D video generation using the SV3D model.

SV3D_Conditioning prepares conditioning data for 3D video generation using the SV3D model. It takes an initial image and processes it through CLIP vision and VAE encoders to create positive and negative conditioning, along with a latent representation. The node generates camera elevation and azimuth sequences for multi-frame video generation based on the specified number of video frames.

Inputs

Parameter	Description	Data Type	Required	Range
clip_vision	The CLIP vision model used for encoding the input image	CLIP_VISION	Yes	-
init_image	The initial image that serves as the starting point for 3D video generation	IMAGE	Yes	-
vae	The VAE model used for encoding the image into latent space	VAE	Yes	-
width	The output width for the generated video frames (default: 576, step of 8)	INT	Yes	16 to MAX_RESOLUTION
height	The output height for the generated video frames (default: 576, step of 8)	INT	Yes	16 to MAX_RESOLUTION
video_frames	The number of frames to generate for the video sequence (default: 21)	INT	Yes	1 to 4096
elevation	The camera elevation angle in degrees for the 3D view (default: 0.0, step of 0.1)	FLOAT	Yes	-90.0 to 90.0

Note: The camera azimuth starts at 0 degrees and increases by a constant amount each frame so that the camera completes a full 360-degree orbit around the object across the generated frames. The per-frame increment is calculated as 360 divided by (video_frames - 1), using a minimum divisor of 2 when only one frame is requested. The elevation value stays constant for every frame.

The init_image is scaled to the specified width and height before VAE encoding, and the returned latent uses dimensions of video_frames x 4 x (height // 8) x (width // 8).

Outputs

Output Name	Description	Data Type
positive	The positive conditioning data containing image embeddings and camera parameters for generation	CONDITIONING
negative	The negative conditioning data with zeroed embeddings and latents for contrastive generation	CONDITIONING
latent	An empty latent tensor with dimensions matching the specified video frames and resolution	LATENT

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): e28173cfa560290e66b032687088cf0b981256ca5c21f6aa608e0fdaec886665