BerniniConditioning - ComfyUI Built-in Node Documentation - ComfyUI

The BerniniConditioning node prepares video and image conditioning data for the Wan2.2-A14B model. It encodes source videos, reference videos, and reference images using the provided VAE, then attaches them to the conditioning data as in-context tokens. The task is automatically inferred from which inputs are connected.

Inputs

Parameter	Description	Data Type	Required	Range
`positive`	Positive conditioning data	CONDITIONING	Yes	-
`negative`	Negative conditioning data	CONDITIONING	Yes	-
`vae`	VAE model used to encode video and image inputs	VAE	Yes	-
`width`	Width of the output latent (default: 832)	INT	Yes	16 to 8192 (step: 16)
`height`	Height of the output latent (default: 480)	INT	Yes	16 to 8192 (step: 16)
`length`	Number of frames in the output latent (default: 81)	INT	Yes	1 to 8192 (step: 4)
`batch_size`	Number of videos to generate in a single batch (default: 1)	INT	Yes	1 to 4096
`source_video`	Source video to edit or restyle (v2v, rv2v). Resized to width/height and trimmed to length.	IMAGE	No	-
`reference_video`	Video to insert into the source video (ads2v). Trimmed to length and resized with preserved aspect ratio (long edge capped at ref_max_size).	IMAGE	No	-
`reference_images`	Reference images injected as in-context tokens (r2v, rv2v). Each image is encoded independently at its own native aspect ratio (long edge capped at ref_max_size). Up to 8 images can be provided.	IMAGE	No	0 to 8 images
`ref_max_size`	Max size for the long edge of reference_video and reference_images. Resized with preserved aspect ratio and snapped to 16px (default: 848).	INT	No	16 to 8192 (step: 16)

Note: The task is inferred from which inputs are connected:

No inputs connected → text-to-video (t2v)
source_video only → video-to-video (v2v)
source_video + reference_images → reference-guided video editing (rv2v)
reference_images only → reference-to-video (r2v)
source_video + reference_video → insert image/video into video (ads2v)

Context streams are attached in a fixed order: source_video (source ID 1), then reference_video (source ID 2), then each reference image (source IDs 3, 4, …). A reference image input containing multiple images contributes one stream per image.

Outputs

Output Name	Description	Data Type
`positive`	Positive conditioning with context latents attached	CONDITIONING
`negative`	Negative conditioning with context latents attached	CONDITIONING
`latent`	Empty latent tensor with dimensions matching the specified width, height, length, and batch size	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): 70b5026f9479249bc5d1fe4695583193216e34e1b092764ac30e0ec3b7eba670