WanFirstLastFrameToVideo - ComfyUI Built-in Node Documentation

The WanFirstLastFrameToVideo node prepares conditioning for video generation by combining a start frame and an end frame with text prompts.

The WanFirstLastFrameToVideo node prepares conditioning for video generation by combining a start frame and an end frame with text prompts. It encodes the frame images into latent space, creates a mask that tells the video model which frames are already known, and attaches CLIP vision features when they are provided. The node outputs updated positive and negative conditioning plus an empty latent that defines the size and length of the video to generate.

Inputs

Parameter	Description	Data Type	Required	Range
positive	Positive text conditioning used to guide the video generation.	CONDITIONING	Yes	-
negative	Negative text conditioning used to guide the video generation.	CONDITIONING	Yes	-
vae	VAE model used to encode the combined frame images into latent space.	VAE	Yes	-
width	Width of the generated video in pixels (default: 832, step: 16).	INT	Yes	16 to MAX_RESOLUTION
height	Height of the generated video in pixels (default: 480, step: 16).	INT	Yes	16 to MAX_RESOLUTION
length	Number of frames in the video sequence (default: 81, step: 4).	INT	Yes	1 to MAX_RESOLUTION
batch_size	Number of videos to generate at once (default: 1).	INT	Yes	1 to 4096
clip_vision_start_image	CLIP vision features extracted from the start image. If both start and end CLIP vision inputs are provided, their features are combined.	CLIP_VISION_OUTPUT	No	-
clip_vision_end_image	CLIP vision features extracted from the end image. If both start and end CLIP vision inputs are provided, their features are combined.	CLIP_VISION_OUTPUT	No	-
start_image	Starting frame image for the video sequence. Its first length frames are used and resized to width × height.	IMAGE	No	-
end_image	Ending frame image for the video sequence. Its last length frames are used and resized to width × height.	IMAGE	No	-

Note: When at least one of start_image or end_image is provided, the node builds a combined frame sequence where the start and end frames are filled in and the remaining frames use a neutral gray placeholder (0.5). A mask marks the filled regions as known and the placeholder regions as unknown, allowing the video model to generate the in-between frames. When a start image is provided, the known region also extends 3 extra frames beyond the image. The same encoded frame image and mask are attached to both positive and negative conditioning. If both CLIP vision inputs are provided, their hidden states are concatenated; if only one is provided, it is used on its own. The latent video length is derived from length after temporal compression: ((length - 1) // 4) + 1.

Outputs

Output Name	Description	Data Type
positive	Positive conditioning with the encoded frame image, mask, and (if provided) CLIP vision features attached.	CONDITIONING
negative	Negative conditioning with the encoded frame image, mask, and (if provided) CLIP vision features attached.	CONDITIONING
latent	Empty latent tensor (all zeros) shaped for the given batch size, video length, and resolution.	LATENT

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): 0072e441cb80334c3c961d1bbf2d081c78bc38ed1eacca840c577a2d01b36f05