TextGenerateV2 - ComfyUI Built-in Node Documentation

The Generate Text node uses a CLIP model to produce text from a prompt.

The Generate Text node uses a CLIP model to produce text from a prompt. Unlike the older single-attachment node, it accepts growable stacks of images, videos and audio files as extra context, so you can feed a whole reference set in one node. You control the output length, whether sampling is used, and whether the model’s own chat template and thinking mode apply.

​

Inputs

Parameter	Description	Data Type	Required	Range
clip	The CLIP model used to tokenize the prompt and generate the text.	CLIP	Yes	-
prompt	The prompt that guides the generation. Supports multiple lines and dynamic prompts (default: empty string).	STRING	Yes	-
images	Growable group of up to 50 images, image1 through image50, used as visual context for the model.	AUTOGROW	No	0 to 50 images
videos	Growable group of up to 100 video inputs, video1 through video100. Each video is a batch of frames assumed to be 24 FPS and subsampled to 1 FPS internally.	AUTOGROW	No	0 to 100 videos
audios	Growable group of up to 100 audio inputs, audio1 through audio100, used as audio context for the model.	AUTOGROW	No	0 to 100 audio files
max_length	Maximum number of tokens to generate (default: 512).	INT	Yes	1 to 32768
sampling_mode	on samples the output and reveals the sampling parameters below. off generates greedily without random sampling.	DYNAMIC_COMBO	Yes	"on" "off"
thinking	Operate in thinking mode if the model supports it (default: False).	BOOLEAN	No	True or False
use_default_template	Use the built in system prompt or chat template if the model has one (default: True). 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	-

​

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 (default: 0.7).	FLOAT	Yes	0.01 to 2.0
top_k	Limits the sampling pool to the K most likely next tokens. A value of 0 disables the filter (default: 64).	INT	Yes	0 to 1000
top_p	Nucleus sampling: keeps the smallest set of most likely tokens whose cumulative probability reaches this value (default: 0.95).	FLOAT	Yes	0.0 to 1.0
min_p	Minimum probability a token must reach to stay in the pool (default: 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 (default: 1.05).	FLOAT	Yes	0.0 to 5.0
presence_penalty	Penalizes new tokens based on whether they have appeared in the text so far, encouraging the model to move to new topics (default: 0.0).	FLOAT	Yes	0.0 to 5.0
seed	Number used to initialize the random number generator for reproducible results (default: 0).	INT	Yes	0 to 18446744073709551615

Note: The sampling parameters above only exist when sampling_mode is "on". With "off" the node generates text without random sampling and the parameters are hidden.

Note: When the model’s output opens a reasoning block with <think>, the node splits it: 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. Reasoning that was cut off by max_length stays in generated_text.

​

Outputs

Output Name	Description	Data Type
generated_text	The text the model generated, 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): b61210726faa495a342a5eaaebe64f800ec1d2af4e2fc1321d8c3d7c594bd9ca