CLIPTextEncodeSD3 - ComfyUI Built-in Node Documentation

CLIPTextEncodeSD3 processes text inputs for Stable Diffusion 3 models by encoding multiple text prompts using different CLIP models. It handles three separate text inputs (clip_g, clip_l, and t5xxl) and provides options for managing empty text padding. The node ensures proper token alignment between different text inputs and returns conditioning data suitable for SD3 generation pipelines.

Inputs

Parameter	Description	Data Type	Required	Range
`clip`	The CLIP model used for text encoding	CLIP	Yes	-
`clip_l`	Text input for the local CLIP model. Supports multiline text and dynamic prompts.	STRING	Yes	-
`clip_g`	Text input for the global CLIP model. Supports multiline text and dynamic prompts.	STRING	Yes	-
`t5xxl`	Text input for the T5-XXL model. Supports multiline text and dynamic prompts.	STRING	Yes	-
`empty_padding`	Controls how empty text inputs are handled. When set to “none”, empty text inputs for `clip_g`, `clip_l`, or `t5xxl` result in empty token lists instead of padding. When set to “empty_prompt”, empty inputs are tokenized as empty prompts (standard padding behavior). This is an advanced parameter (default: “none”).	COMBO	Yes	`"none"` `"empty_prompt"`

Parameter Constraints:

When empty_padding is set to “none”, empty text inputs for clip_g, clip_l, or t5xxl result in empty token lists instead of padding
The node automatically balances token lengths between clip_l and clip_g inputs by padding the shorter one with empty tokens when lengths differ
All text inputs support dynamic prompts and multiline text entry

Outputs

Output Name	Description	Data Type
`CONDITIONING`	The encoded text conditioning data ready for use in SD3 generation pipelines	CONDITIONING

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): 874869bac024e6b5ac6b4bf4f79c31bb750e54f7096f6638647aac6b95bb202f

​Inputs

​Outputs

Inputs

Outputs