ComfyUI Node
VAE Decode Tiled (safe, optional cleanup)
A ComfyUI node in IAMCCS/HW with 11 inputs and 1 output.
VAE Decode Tiled (safe, optional cleanup)
- samples
- vae
- IMAGE
◄tiletrue►
◄tiling_modeauto►
◄tile_size512►
◄overlap64►
◄temporal_size256►
◄temporal_overlap32►
◄cleanup_before_decodefalse►
◄last_frame_fixfalse►
◄duration_hint_s0►
CategoryIAMCCS/HW
Inputs (11)
| Name | Type | Default | Description |
|---|---|---|---|
| samples | LATENT | — | |
| vae | VAE | — | |
| tile | BOOLEAN | true | — |
| tiling_mode | COMBO | auto | 2 options: auto, manual |
| tile_size | INT | 51264–4096 | — |
| overlap | INT | 640–4096 | — |
| temporal_size | INT | 2568–4096 | Only for video VAEs: widget-scale frames per temporal chunk (divided by VAE temporal compression at runtime). Keep >= 256 to avoid seam artifacts on LTX-2 (256 / 8 = 32 latent frames). |
| temporal_overlap | INT | 324–4096 | Only for video VAEs: widget-scale overlap frames between chunks (divided by VAE temporal compression at runtime). |
| cleanup_before_decode | BOOLEAN | false | — |
| last_frame_fix | BOOLEAN | false | LTX VAE workaround: duplicates the last latent frame before decoding to prevent last-frame artifacts. |
| duration_hint_s | INT | 00–3600 | Video duration in seconds — used only by 'Probe HW & Apply' to compute the correct tile size recommendation. 0 = auto-detect from workflow. Does NOT affect the actual decode. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| IMAGE | IMAGE | — |