WanAnimatePlus Encode
The encode node for v2v and inpainting
- vae
- image
- mask
- samples
Most Wan Animate workflows start from a reference image, not existing footage. But the moment you want video-to-video - restyle your footage, extend a clip, or inpaint - you need to get a starting video into latent space. That's this node: WanAnimatePlus Encode takes an IMAGE (your existing video frames, or a still) plus your vae, and hands the sampler a LATENT to start from instead of pure noise. It's the front half of the v2v story, Decode being the back half.
The settings are mostly VRAM. enable_vae_tiling and the four tile numbers (tile_x, tile_y, tile_stride_x, tile_stride_y) mirror the Decode node exactly - tiling saves memory, smaller tiles introduce more seams. Defaults (272/272 tiles, 144/128 strides) are a sane starting point. Leave tiling off if your card handles the encode comfortably; a 5-second clip at 480p usually does.
Two inputs that aren't about memory. noise_aug_strength (default 0) adds noise to the encoded latents. The tooltip names the use case: it's "helpful for leapfusion I2V where some noise can add motion and give sharper results" - because a perfectly clean latent start leaves the model no room to invent motion. Crank it a little for v2v where you want the footage to move, keep it at 0 for faithful reconstruction. latent_strength (default 1) is a multiplier on the encoded latents; lower values weaken the starting anchor and allow more deviation - the same idea, on a different axis. The mask input is how you do masked inpainting: encode the video, mask the region, and let the sampler regenerate only what's inside.
How it fits the chain. Encode's samples output feeds the samples input on the WanAnimatePlus Sampler (or SamplerFromSettings), and you'll usually want add_noise_to_samples on in the sampler when you're starting from clean video - that's the pair that actually enables v2v. It also plays with the fork's signature features: transition_video gets encoded through this same VAE path in AnimateEmbeds.
Install via ComfyUI Manager (search "WanAnimatePlus") or git clone https://github.com/wuwukaka/ComfyUI-WanAnimatePlus into custom_nodes, restart. The fork-wide rule applies as ever: don't mix this encode with original WanVideoWrapper sampler nodes, since the latent layouts disagree. And if your v2v output looks frozen, the fix is almost always noise_aug_strength or add_noise_to_samples, not more steps.
Inputs (10)
| Name | Type | Default | Description |
|---|---|---|---|
| vae | WANVAE | — | |
| image | IMAGE | — | |
| enable_vae_tiling | BOOLEAN | false | Drastically reduces memory use but may introduce seams |
| tile_x | INT | 27264–2048 | Tile size in pixels, smaller values use less VRAM, may introduce more seams |
| tile_y | INT | 27264–2048 | Tile size in pixels, smaller values use less VRAM, may introduce more seams |
| tile_stride_x | INT | 14432–2048 | Tile stride in pixels, smaller values use less VRAM, may introduce more seams |
| tile_stride_y | INT | 12832–2048 | Tile stride in pixels, smaller values use less VRAM, may introduce more seams |
| noise_aug_strengthopt | FLOAT | 0.0000–10 | Strength of noise augmentation, helpful for leapfusion I2V where some noise can add motion and give sharper results |
| latent_strengthopt | FLOAT | 1.0000–10 | Additional latent multiplier, helpful for leapfusion I2V where lower values allow for more motion |
| maskopt | MASK | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| samples | LATENT | — |