ComfyUI Node
Video Tile Reference Color Match
Post-merge: match large-scale color to a reference clip (LR/pre-upscale) while preserving merged detail. Uses Gaussian low/high split; long batches are chunked for PyTorch limits.
Video Tile Reference Color Match
- merged
- reference
- IMAGE
◄low_frequency_sigma14.0►
◄color_pull0.58►
◄detail_mix1.00►
◄preserve_merged_luminancetrue►
◄luma_scale_clamp4.00►
◄reference_resizebicubic►
CategoryVideo Tiler
Inputs (8)
| Name | Type | Default | Description |
|---|---|---|---|
| merged | IMAGE | Video Tile Merge output [B,H,W,3]. | |
| reference | IMAGE | Same content at LR or pre-upscale resolution; resized internally. Batch 1 broadcasts to merged batch. | |
| low_frequency_sigmaopt | FLOAT | 14.00–256 | Gaussian σ (px) for LF/HF split. Higher = only broader color moves. ~0 skips blur (strong reference imprint). |
| color_pullopt | FLOAT | 0.580–1 | How much low-frequency RGB follows the reference vs merged LF. |
| detail_mixopt | FLOAT | 1.000–1 | Gain on high-frequency residual from merged (1 = keep full upscale detail). |
| preserve_merged_luminanceopt | BOOLEAN | true | If true, rescale RGB after pull so Rec.709 luma matches merged (reference brightness won’t flatten scene). |
| luma_scale_clampopt | FLOAT | 4.001.05–10 | Max/min multiplier when locking luma (prevents extreme corrections). |
| reference_resizeopt | COMBO | bicubic | Interpolation when upsampling reference to merged width/height. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| IMAGE | IMAGE | Color-matched IMAGE; same shape/dtype as merged input. |