WanAnimatePlus Uni3C Embeds
Point the Uni3C controlnet at your latent — and say when it applies
- controlnet
- render_latent
- render_mask
- uni3c_embeds
Once you've loaded a Uni3C controlnet, something has to actually apply it. That's this node. It takes the loaded controlnet, the latent you're sampling, and a strength, and produces uni3c_embeds - the conditioning bundle the WanAnimatePlus sampler reads during denoising. It's the "where and how hard" half of the Uni3C pair, and it's where you'll spend your tuning time.
ControlNet thinking transfers directly here, and the KB framing is the useful one: the control weight decides how much structure wins, and the start/end percent decide when in the denoise it's allowed to win. In 2026 ControlNet practice, gating the condition to the composition phase (say start 0.0, end 0.5) lets the model add its own detail late in the run instead of being clamped the whole way. That's exactly what these knobs are for.
The knobs
strength(default 1.0) - how hard the controlnet pulls. Start at 1, nudge down if the structure dominates into artifacts, up if the model is ignoring it.start_percent(0) andend_percent(1) - the step window (as a fraction of the run) where the condition applies. The tooltip is unambiguous: start percent of the steps to apply, end percent to apply. 0→1 means always.offload(default true) - unloads the controlnet before the main model's block processing to save VRAM. Leave it on unless you're chasing every last bit of speed.render_latent(optional) - pass the latent you're sampling; this is how the node knows the geometry it's conditioning.render_mask(optional) - here's the trap: the tooltip says "NOT IMPLEMENTED!" The code literally raisesNotImplementedErrorif you connect a mask. Don't wire it, and if a workflow you downloaded has it connected, that's why it fails.
Output is a single uni3c_embeds (UNI3C_EMBEDS) that feeds the WanAnimatePlus sampler's control input. Loader → Embeds → sampler is the shape of it:
WanAnimatePlus Uni3C ControlnetLoader → WanAnimatePlus Uni3C Embeds → WanAnimatePlus Sampler
Install
Standard for the pack:
cd ComfyUI/custom_nodes
git clone https://github.com/wuwukaka/ComfyUI-WanAnimatePlus.git
Restart ComfyUI, or install via Manager (search "ComfyUI-WanAnimatePlus"). Keep the original ComfyUI-WanVideoWrapper installed, and remember the pack rule: don't mix WanAnimatePlus nodes with original WanVideoWrapper nodes in one workflow.
The one thing that trips people: that tempting render_mask port is a placeholder with a "NOT IMPLEMENTED!" sticker on it. Every other input here is honest, well-tooltipped, and behaves like the ControlNet dials you already know.
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| controlnet | WANVIDEOCONTROLNET | — | |
| strength | FLOAT | 1.0000–10 | — |
| start_percent | FLOAT | 0.000–1 | Start percent of the steps to apply the controlnet |
| end_percent | FLOAT | 1.000–1 | End percent of the steps to apply the controlnet |
| render_latentopt | LATENT | — | |
| render_maskopt | MASK | NOT IMPLEMENTED! | |
| offloadopt | BOOLEAN | true | If enabled, the controlnet model will be offloaded before main model block processing to save VRAM. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| uni3c_embeds | UNI3C_EMBEDS | — |