HunyuanVideo15LatentUpscaleWithModel - ComfyUI Built-in Node Documentation

The Hunyuan Video 15 Latent Upscale With Model node increases the resolution of a latent image representation.

The Hunyuan Video 15 Latent Upscale With Model node increases the resolution of a latent image representation. It first upscales the latent samples to a specified size using a chosen interpolation method, then refines the upscaled result using a specialized Hunyuan Video 1.5 upscale model to improve quality.

Inputs

Parameter	Description	Data Type	Required	Range
model	The Hunyuan Video 1.5 latent upscale model used to refine the upscaled samples.	LATENT_UPSCALE_MODEL	Yes	N/A
samples	The latent image representation to be upscaled.	LATENT	Yes	N/A
upscale_method	The interpolation algorithm used for the initial upscaling step (default: "bilinear").	COMBO	Yes	"nearest-exact" "bilinear" "area" "bicubic" "bislerp"
width	The target width for the upscaled latent, in pixels. A value of 0 will calculate the width automatically based on the target height and the original aspect ratio. The final output width will be a multiple of 16 (default: 1280).	INT	Yes	0 to 16384 (step: 8)
height	The target height for the upscaled latent, in pixels. A value of 0 will calculate the height automatically based on the target width and the original aspect ratio. The final output height will be a multiple of 16 (default: 720).	INT	Yes	0 to 16384 (step: 8)
crop	Determines how the upscaled latent is cropped to fit the target dimensions.	COMBO	Yes	"disabled" "center"

Note on Dimensions: If both width and height are set to 0, the node returns the input samples unchanged. If only one dimension is set to 0, the other dimension is calculated to preserve the original aspect ratio. Both values are clamped to a minimum of 64, and the upscale target passed to the interpolation step is width // 16 by height // 16, so the requested dimensions are effectively rounded down to multiples of 16.

Outputs

Output Name	Description	Data Type
LATENT	The upscaled and model-refined latent image representation, returned as a float tensor on the CPU.	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): 71af49eefb010aadd30f8699b751ad901b1ee385d6cbeedd3a83995a1a623516