ComfyUI Node
GPU LayerStyles ImageScaleDown
Downscale to exact dimensions with antialiased bicubic or Lanczos-3 in float32. Either dimension exceeding the input is rejected. RGBA uses premultiplied alpha while filtering. Resized values are clamped to [0, 1]; unchanged sizes are copied without filtering.
GPU LayerStyles ImageScaleDown
- image
- image
◄width512►
◄height512►
◄methodbicubic►
◄output_devicecpu►
◄batch_size0►
CategoryGPU LayerStyles/Image
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | — | |
| width | INT | 5121–16384 | — |
| height | INT | 5121–16384 | — |
| method | COMBO | bicubic | 2 options: bicubic, lanczos |
| output_device | COMBO | cpu | Return output on CPU (default) after float32 processing on ComfyUI's selected compute device. GPU keeps output on that device to avoid transfers between compatible nodes. ComfyUI CPU mode returns CPU output. |
| batch_size | INT | 00–2147483647 | Frames per processing chunk. 0 starts at up to 64 frames. Positive values can exceed 64, capped by the output length. Allocation failures halve the chunk size for this run; each run starts fresh. Complete inputs and outputs still need memory. Pass smaller batches through the entire workflow to reduce complete-batch memory. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | — |