Nodes/ComfyUI-HSWQ-Loader-and-Tools/HSWQ Ultimate SD Upscale
ComfyUI Node

HSWQ Ultimate SD Upscale

Upscales an image and runs image-to-image on tiles from the input image.

By ussoewwin·Created 8 months ago·Updated about 15 hours ago· 16
HSWQ Ultimate SD Upscale
  • image
  • model
  • positive
  • negative
  • vae
  • upscale_model
  • IMAGE
upscale_byAuto
target_height4320
seed0
steps20
cfg8.00
sampler_name
scheduler
denoise0.20
mode_type
tile_width512
tile_height512
mask_blur8
tile_padding32
seam_fix_mode
seam_fix_denoise1.00
seam_fix_width64
seam_fix_mask_blur8
seam_fix_padding16
force_uniform_tilestrue
tiled_decodefalse
batch_size1
tensor_boostfalse
Categoryimage/upscaling

Inputs (28)

NameTypeDefaultDescription
imageIMAGEThe image to upscale.
modelMODELThe model to use for image-to-image.
positiveCONDITIONINGThe positive conditioning for each tile.
negativeCONDITIONINGThe negative conditioning for each tile.
vaeVAEThe VAE model to use for tiles.
upscale_byCOMBOAutoChoose 'Auto' to calculate the scale from target vertical pixels, or select a fixed magnification.
target_heightINT432064–8192Target output height in pixels. Used only when upscale_by is 'Auto'.
seedINT00–18446744073709550000The seed to use for image-to-image.
stepsINT201–10000The number of steps to use for each tile.
cfgFLOAT8.000–100The CFG scale to use for each tile.
sampler_nameCOMBOThe sampler to use for each tile.
schedulerCOMBOThe scheduler to use for each tile.
denoiseFLOAT0.200–1The denoising strength to use for each tile.
upscale_modelUPSCALE_MODELThe upscaler model for upscaling the image.
mode_typeCOMBOThe tiling order to use for the redraw step.
tile_widthINT51264–8192The width of each tile.
tile_heightINT51264–8192The height of each tile.
mask_blurINT80–64The blur radius for the mask.
tile_paddingINT320–8192The padding to apply between tiles.
seam_fix_modeCOMBOThe seam fix mode to use.
seam_fix_denoiseFLOAT1.000–1The denoising strength to use for the seam fix.
seam_fix_widthINT640–8192The width of the bands used for the Band Pass seam fix mode.
seam_fix_mask_blurINT80–64The blur radius for the seam fix mask.
seam_fix_paddingINT160–8192The padding to apply for the seam fix tiles.
force_uniform_tilesBOOLEANtrueForce all tiles to be the same as the set tile size, even when tiles could be smaller. This can help prevent the model from working with irregular tile sizes.
tiled_decodeBOOLEANfalseWhether to use tiled decoding when decoding tiles.
batch_sizeINT11–4096The number of tiles to process in a batch. Higher values can reduce processing time but use more VRAM. If you get tensor size mismatch with FP8/FP4 (quantized) models, set this to 1.
tensor_boostBOOLEANfalseEnable Blackwell Per-Weight CUDA Graph Tensor Boost during USDU tile upscaling. ON raises VRAM by several GB; RTX 5090 32GB+ recommended. Keep OFF for tiled upscale.

Outputs (1)

NameTypeDescription
IMAGEIMAGEThe final upscaled image.