ComfyUI Node
Bernini-R VAE Decode
Tiled VAE decode with optional colour matching. Matches decoded video colour distribution to a reference image for improved temporal consistency.
Bernini-R VAE Decode
- vae
- samples
- ref_image
- images
◄colormatchdisabled►
◄enable_tilingfalse►
◄tile_x272►
◄tile_y272►
◄tile_stride_x144►
◄tile_stride_y128►
◄blend_strength1.00►
CategoryBernini-R/VAE
Inputs (10)
| Name | Type | Default | Description |
|---|---|---|---|
| vae | VAE | Bernini-R VAE (Wan 16-channel, 4x temporal downscale) | |
| samples | LATENT | Latent to decode (from sampler output) | |
| colormatchopt | COMBO | disabled | Color transfer algorithm. 'hm-mvgd-hm' is a good default for temporal consistency |
| enable_tilingopt | BOOLEAN | false | Spatial tiling for VRAM-limited GPUs. Enables decoding very large frames |
| tile_xopt | INT | 27264–4096 | Tile width in pixels |
| tile_yopt | INT | 27264–4096 | Tile height in pixels |
| tile_stride_xopt | INT | 14416–4096 | Horizontal stride (must be < tile_x). Larger = fewer tiles, more artifacts |
| tile_stride_yopt | INT | 12816–4096 | Vertical stride (must be < tile_y). Larger = fewer tiles, more artifacts |
| ref_imageopt | IMAGE | Reference image for color matching. If not provided, uses first decoded frame per batch | |
| blend_strengthopt | FLOAT | 1.000–1 | Color match blend. 0.0 = original, 1.0 = full transfer |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| images | IMAGE | — |