Generate Text - ComfyUI Built-in Node Documentation

The TextGenerate node uses a CLIP model to create text based on a user’s prompt. It can optionally use images, video, or audio as additional context to guide the text generation. You can control the length of the output, enable a thinking mode for supported models, and choose whether to use random sampling with various settings or to generate text without sampling.

Inputs

Common Inputs

Parameter	Description	Data Type	Required	Range
`sampling_mode`	Controls whether random sampling is used during text generation. When set to “on”, additional sampling parameters become available. When set to “off”, the node generates text without random sampling.	DYNAMIC_COMBO	Yes	`"on"` `"off"`
`clip`	The CLIP model used for tokenizing the prompt and generating text.	CLIP	Yes	N/A
`prompt`	The text prompt that guides the generation. This field supports multiple lines and dynamic prompts. The default value is an empty string.	STRING	Yes	N/A
`image`	An optional image that can be used alongside the text prompt to influence the generated text.	IMAGE	No	N/A
`video`	Video frames as an image batch. Assumed to be 24 FPS; subsampled to 1 FPS internally.	IMAGE	No	N/A
`audio`	An optional audio input that can be used alongside the text prompt to influence the generated text.	AUDIO	No	N/A
`max_length`	The maximum number of tokens the model will generate. The default value is 512.	INT	Yes	1 to 32768
`thinking`	Operate in thinking mode if the model supports it. The default value is False.	BOOLEAN	No	True or False
`use_default_template`	Use the built in system prompt/template if the model has one. The default value is True. This is an advanced parameter.	BOOLEAN	No	True or False
`mtp`	Speculative decoding with the checkpoint’s multi-token-prediction head. Has no effect without MTP weights. `"auto"` adapts the draft depth and breadth to the observed token matches and execution cost, `"2"` to `"5"` pin a constant draft depth, and the `tree` modes pin both breadth and depth. Sampled output stays correctly distributed but differs from non-MTP output for the same seed (default: `"auto"`).	COMBO	No	`"auto"` `"off"` `"2"` `"3"` `"4"` `"5"` `"tree2s1"` `"tree2s2"` `"tree3s1"` `"tree4s1m8"`
`system_prompt`	Replaces the system prompt in the model’s chat template. Ignored when the default template is not used. Connect a STRING input instead of typing in the node (default: empty).	STRING	No	N/A

Sampling Parameters (when `sampling_mode` is “on”)

Parameter	Description	Data Type	Required	Range
`temperature`	Controls the randomness of the output. Lower values make the output more predictable, higher values make it more creative. The default value is 0.7.	FLOAT	Yes	0.01 to 2.0
`top_k`	Limits the sampling pool to the top K most likely next tokens. A value of 0 disables this filter. The default value is 64.	INT	Yes	0 to 1000
`top_p`	Uses nucleus sampling: it keeps the smallest set of most likely tokens whose cumulative probability reaches this value. The default value is 0.95.	FLOAT	Yes	0.0 to 1.0
`min_p`	Sets a minimum probability threshold for tokens to be considered. The default value is 0.05.	FLOAT	Yes	0.0 to 1.0
`repetition_penalty`	Penalizes tokens that have already been generated to reduce repetition. A value of 1.0 applies no penalty. The default value is 1.05.	FLOAT	Yes	0.0 to 5.0
`seed`	A number used to initialize the random number generator for reproducible results. The default value is 0.	INT	Yes	0 to 18446744073709551615
`presence_penalty`	Penalizes new tokens based on whether they have appeared in the text so far, encouraging the model to talk about new topics. The default value is 0.0.	FLOAT	No	0.0 to 5.0

Note: The sampling parameters above are only active and visible in the node interface when sampling_mode is set to “on”. When sampling_mode is set to “off”, no sampling parameters are available and the node generates text without random sampling. Note: When the generated text contains a reasoning block (it opens with <think> or the prompt ends with it), the reasoning is returned separately: generated_text carries the answer and thinking carries the reasoning with the opening tag removed. Otherwise generated_text carries everything the model produced and thinking is empty.

Outputs

Output Name	Description	Data Type
`generated_text`	The text generated by the model based on the input prompt and optional image, video, or audio, with any reasoning block separated out into the `thinking` output.	STRING
`thinking`	The reasoning block the model produced, with the opening `<think>` tag removed. Empty when the model produced no reasoning block.	STRING

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): e0854b0ce5a41e195a2634f4739c22c7dd0df1f2309cb8a02b506b8791bf3bee

​Inputs

​Common Inputs

​Sampling Parameters (when sampling_mode is “on”)

​Outputs

Inputs

Common Inputs

Sampling Parameters (when `sampling_mode` is “on”)

Outputs