GeminiVideoOmniV2 - ComfyUI Built-in Node Documentation

Google Gemini Omni (Video) generates a video with audio from a text prompt using Google’s Gemini Omni Flash models. You can optionally attach reference images and/or videos to guide the result or to edit existing footage. Describe the desired length (3-10 seconds) directly in the prompt.

Inputs

Common Inputs

Parameter	Description	Data Type	Required	Range
`model`	The Gemini video model used to generate the video.	DYNAMIC_COMBO	Yes	”Omni Flash 1.1" "Omni Flash”

Omni Flash 1.1 Inputs

Parameter	Description	Data Type	Required	Range
`prompt`	Describe the video to generate, or the edit to apply to an attached video. Specify the length directly in the prompt, e.g. “a 6-second clip” or, for the ‘extend’ task, “extend by 5 seconds”; the generated length may be 3-10 seconds and defaults to 10. The output has audio. (default: "")	STRING	Yes	-
`resolution`	Output resolution. (default: “720p”)	COMBO	Yes	”360p" "720p" "1080p" "4k”
`aspect_ratio`	Output aspect ratio: 16:9 (landscape) or 9:16 (portrait). The ‘edit’ and ‘extend’ tasks keep the aspect ratio of the input video instead. (default: “16:9”)	COMBO	Yes	”16:9" "9:16”
`task_type`	What to do with the prompt and the attached media. With ‘auto’ the model decides. ‘text_to_video’ generates from the prompt alone and rejects attached media. ‘image_to_video’ animates one image, or interpolates from a starting frame to an ending frame when two are attached. ‘reference_to_video’ treats the attached media as subject references. ‘edit’ rewrites exactly one attached video, and ‘extend’ appends new footage to it, so the output starts with the input video. (default: “auto”)	COMBO	Yes	”auto" "text_to_video" "image_to_video" "reference_to_video" "edit" "extend”
`seed`	Seed controls whether the node should re-run; results are non-deterministic regardless of seed. (default: 42)	INT	Yes	0 to 2147483647

Omni Flash Inputs

Parameter	Description	Data Type	Required	Range
`prompt`	Describe the video to generate, or the edit to apply to an attached video. Specify the length directly in the prompt, e.g. “a 6-second clip”; length may be 3-10 seconds. The output is 720p, 24 FPS, with audio. (default: "")	STRING	Yes	-
`aspect_ratio`	Output aspect ratio: 16:9 (landscape) or 9:16 (portrait). The ‘edit’ task keeps the aspect ratio of the input video instead. (default: “16:9”)	COMBO	Yes	”16:9" "9:16”
`task_type`	What to do with the prompt and the attached media. With ‘auto’ the model decides. ‘text_to_video’ generates from the prompt alone and rejects attached media. ‘image_to_video’ animates one image, or interpolates from a starting frame to an ending frame when two are attached. ‘reference_to_video’ treats the attached media as subject references. ‘edit’ rewrites exactly one attached video. (default: “auto”)	COMBO	Yes	”auto" "text_to_video" "image_to_video" "reference_to_video" "edit”
`temperature`	Controls randomness. Lower is more focused/deterministic, higher is more varied. (default: 1.0)	FLOAT	Yes	0.0 to 2.0 (step 0.01)
`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 (step 0.01)
`seed`	Seed controls whether the node should re-run; results are non-deterministic regardless of seed. (default: 42)	INT	Yes	0 to 2147483647

Reference Inputs

Parameter	Description	Data Type	Required	Range
`images`	Growable slot: connect up to 14 images (`image_1`…`image_14`). Optional reference image(s) to guide or animate the video. With the ‘image_to_video’ task the first one is the starting frame and an optional second one is the ending frame.	IMAGE	No	0 to 14 images
`videos`	Growable slot: connect up to 3 videos (`video_1`…`video_3`). Optional reference video(s) to guide or edit. Each up to 10 seconds long.	VIDEO	No	0 to 3 videos

Notes:

The prompt must not be empty; the node raises an error if it is.
The text_to_video task generates from the prompt alone — attaching images or videos raises an error.
The image_to_video task accepts images only (no videos) and requires exactly 1 or 2 images: the first is the starting frame and the optional second is the ending frame.
The edit task (both models) and the extend task (Omni Flash 1.1 only) require exactly one input video and keep the aspect ratio of that input video, overriding aspect_ratio.
At most 14 images and 3 videos can be attached, and each attached video must be 10 seconds or shorter.
Omni Flash always outputs 720p 24 FPS video with audio; resolution selection is only available with Omni Flash 1.1.
temperature and top_p controls are only available with the Omni Flash model; Omni Flash 1.1 uses fixed generation settings.

Outputs

Output Name	Description	Data Type
`video` (first output)	The generated video with audio. For Omni Flash: 720p, 24 FPS. For Omni Flash 1.1: the resolution selected in the `resolution` input.	VIDEO
`text` (second output)	The text content generated by the model along with the video (may be empty).	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): 7a0dda4bcd662c9df3c680297ec9de7886d35e618de8b3ce0cd95b9afd13a892

​Inputs

​Common Inputs

​Omni Flash 1.1 Inputs

​Omni Flash Inputs

​Reference Inputs

​Outputs