GeminiNodeV2 - ComfyUI Built-in Node Documentation

Generate text responses with Google’s Gemini models.

Generate text responses with Google’s Gemini models. Provide a text prompt and, optionally, one or more images, audio clips, videos, or files as multimodal context.

Inputs

Common Inputs

Parameter	Description	Data Type	Required	Range
model	The Gemini model used to generate the response.	DYNAMIC_COMBO	Yes	"Gemini 3.7 Flash" "Gemini 3.5 Flash" "Gemini 3.1 Pro" "Gemini 3.1 Flash-Lite"
prompt	Text input to the model. Include detailed instructions, questions, or context. Must contain at least one non-whitespace character. (default: "")	STRING	Yes
seed	Seed for sampling. Set to 0 for a random seed. Deterministic output isn’t guaranteed. (default: 42)	INT	Yes	0 to 2147483647
system_prompt	Foundational instructions that dictate the model’s behavior. (default: "")	STRING	No

Gemini 3.7 Flash Inputs

These inputs appear when model is set to "Gemini 3.7 Flash".

Parameter	Description	Data Type	Required	Range
thinking_level	How hard the model reasons internally before answering. HIGH improves quality on difficult tasks but costs more (thinking) tokens and is slower. (default: “MEDIUM”)	COMBO	Yes	"LOW" "MEDIUM" "HIGH"
temperature	Controls randomness. Lower is more focused/deterministic, higher is more creative. (default: 1.0)	FLOAT	Yes	0.0 to 2.0
top_p	Nucleus sampling: sample from the smallest token set whose cumulative probability reaches top_p. (default: 0.95)	FLOAT	Yes	0.0 to 1.0
max_output_tokens	Maximum tokens to generate, including the model’s internal thinking. With thinking_level HIGH, a low value can leave no room for the answer; raise this if responses come back empty or truncated. The model stops early when finished, so a higher cap costs nothing extra for short replies. (default: 32768)	INT	Yes	16 to 65536

Gemini 3.5 Flash Inputs

These inputs appear when model is set to "Gemini 3.5 Flash".

Parameter	Description	Data Type	Required	Range
thinking_level	How hard the model reasons internally before answering. HIGH improves quality on difficult tasks but costs more (thinking) tokens and is slower. (default: “MEDIUM”)	COMBO	Yes	"MINIMAL" "LOW" "MEDIUM" "HIGH"
temperature	Controls randomness. Lower is more focused/deterministic, higher is more creative. (default: 1.0)	FLOAT	Yes	0.0 to 2.0
top_p	Nucleus sampling: sample from the smallest token set whose cumulative probability reaches top_p. (default: 0.95)	FLOAT	Yes	0.0 to 1.0
max_output_tokens	Maximum tokens to generate, including the model’s internal thinking. With thinking_level HIGH, a low value can leave no room for the answer; raise this if responses come back empty or truncated. The model stops early when finished, so a higher cap costs nothing extra for short replies. (default: 32768)	INT	Yes	16 to 65536

Gemini 3.1 Pro Inputs

These inputs appear when model is set to "Gemini 3.1 Pro".

Parameter	Description	Data Type	Required	Range
thinking_level	How hard the model reasons internally before answering. HIGH improves quality on difficult tasks but costs more (thinking) tokens and is slower. (default: “HIGH”)	COMBO	Yes	"LOW" "HIGH"
temperature	Controls randomness. Lower is more focused/deterministic, higher is more creative. (default: 1.0)	FLOAT	Yes	0.0 to 2.0
top_p	Nucleus sampling: sample from the smallest token set whose cumulative probability reaches top_p. (default: 0.95)	FLOAT	Yes	0.0 to 1.0
max_output_tokens	Maximum tokens to generate, including the model’s internal thinking. With thinking_level HIGH, a low value can leave no room for the answer; raise this if responses come back empty or truncated. The model stops early when finished, so a higher cap costs nothing extra for short replies. (default: 32768)	INT	Yes	16 to 65536

Gemini 3.1 Flash-Lite Inputs

These inputs appear when model is set to "Gemini 3.1 Flash-Lite".

Parameter	Description	Data Type	Required	Range
thinking_level	How hard the model reasons internally before answering. HIGH improves quality on difficult tasks but costs more (thinking) tokens and is slower. (default: “LOW”)	COMBO	Yes	"LOW" "HIGH"
temperature	Controls randomness. Lower is more focused/deterministic, higher is more creative. (default: 1.0)	FLOAT	Yes	0.0 to 2.0
top_p	Nucleus sampling: sample from the smallest token set whose cumulative probability reaches top_p. (default: 0.95)	FLOAT	Yes	0.0 to 1.0
max_output_tokens	Maximum tokens to generate, including the model’s internal thinking. With thinking_level HIGH, a low value can leave no room for the answer; raise this if responses come back empty or truncated. The model stops early when finished, so a higher cap costs nothing extra for short replies. (default: 32768)	INT	Yes	16 to 65536

Media and File Inputs

The following inputs are shared by all four models and appear alongside the model-specific inputs.

Parameter	Description	Data Type	Required	Range
images	Growable slot: connect 1 to 16 images (image_1 … image_16). Optional image(s) to use as context for the model. Up to 16 images.	IMAGE	No	0 to 16 images
audio	Growable slot: connect one audio clip (audio_1). Optional audio clip to use as context for the model.	AUDIO	No	0 to 1 clip
video	Growable slot: connect one video clip (video_1). Optional video clip to use as context for the model.	VIDEO	No	0 to 1 clip
files	Optional file(s) to use as context for the model. Accepts inputs from the Gemini Input Files node.	GEMINI_INPUT_FILES	No

Note: When media (images, audio, or video) is attached, the node uploads the first 10 media items to ComfyAPI storage and passes them as URLs; this URL budget is shared across all media types and is consumed in order (video first, then audio, then images). Any remaining media is encoded inline as base64 data, with a maximum combined inline payload of 18 MB. If the inline payload would exceed 18 MB, the node raises an error. The prompt parameter must contain at least one non-whitespace character. Setting seed to 0 requests a random seed.

Outputs

Output Name	Description	Data Type
output	The generated text response from the Gemini model. If the model produces no text, the string “Empty response from Gemini model…” is returned.	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): 00e0f614303fa723eb787ad763e0b0c6322f89abf43d93b697357527b2fae49c