ComfyUI Node
Wan VAE Decode
A ComfyUI node in latent with 11 inputs and 1 output.
Wan VAE Decode
- vae
- latents
- image
◄horizontal_tiles1►
◄vertical_tiles1►
◄overlap1►
◄last_frame_fixfalse►
◄frames_per_pass0►
◄batch_per_pass0►
◄max_decode_megapixels0.0►
◄accumulate_dtypefp32►
◄output_dtypeauto►
Categorylatent
Inputs (11)
| Name | Type | Default | Description |
|---|---|---|---|
| vae | VAE | VAE used to decode the latents. Must correspond to the model that produced the latents. Respects the VAE's device/dtype and internal time/space scaling. | |
| latents | LATENT | Latent dict with key 'samples' shaped [B,C,F,H,W]. F=frames (F=1 for images). H/W are latent-space, not pixel-space. | |
| horizontal_tiles | INT | 11–12 | Tiles across WIDTH in latent space. |
| vertical_tiles | INT | 11–12 | Tiles across HEIGHT in latent space. |
| overlap | INT | 10–16 | Overlap in LATENT pixels. Output overlap = overlap * VAE scale. |
| last_frame_fix | BOOLEAN | false | Append last latent frame before decode, trim after. For video VAEs. |
| frames_per_passopt | INT | 00–64 | Limit frames per pass (0 = all). |
| batch_per_passopt | INT | 00–64 | Limit batch per pass (0 = all). |
| max_decode_megapixelsopt | FLOAT | 0.00–400 | VRAM budget ≈ decH * decW * out_frames * batch / 1e6. 0 = off. |
| accumulate_dtypeopt | COMBO | fp32 | Accumulator dtype for blending. |
| output_dtypeopt | COMBO | auto | Final NHWC dtype ('auto' = fp16 on CUDA, else fp32). |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | — |