WanAnimatePlus Samplerv2
Scheduler from a node, extras from a bundle, guidance on the panel
- model
- image_embeds
- scheduler
- text_embeds
- samples
- extra_args
- samples
- denoised_samples
The v1 Sampler tried to own everything - steps, shift, scheduler, twenty optional bundles - and the panel got unmanageable. Samplerv2 is the refactor: it hands the schedule to a dedicated scheduler node and the extras to a bundle node, and keeps only the guidance controls you actually tune on its own face.
What changed. The big one: steps, shift, and the scheduler enum are gone. Instead, scheduler is a WANVIDEOSCHEDULER object - produced by the pack's WanAnimatePlus Scheduler or Schedulerv2 nodes, where steps, shift and the scheduling curve actually live. Build your schedule there, feed it in here. And instead of a dozen optional embed inputs, there's a single extra_args input accepting the WANVIDSAMPLEREXTRAARGS bundle from WanAnimatePlus SamplerExtraArgs - context options, RIFLEX, cache, loop args and model embeds all ride in that one wire.
What stayed. model, image_embeds, cfg (default 6), seed, force_offload, plus text_embeds, samples (init latents for v2v) and add_noise_to_samples (needed when starting from clean video). And critically, the full guidance block: guidance_mode (cfg/apg/apg_chain/cfg_chain) with apg_eta, apg_momentum, apg_norm_threshold, apg_omega, apg_omega_I, apg_omega_TI, and chain_omega_V, chain_omega_I, chain_omega_TI. That's the whole Bernini story - rv2v editing wants cfg_chain with the chain omegas, r2v wants apg_chain - and it's all here on the panel, not buried in a bundle. Note there's no steps here because the scheduler node owns it; if you're moving from v1, re-wire your steps/shift into the scheduler or nothing changes when you drag the sliders.
When to use it. If you're building a graph with Scheduler/Schedulerv2 + SamplerExtraArgs anyway - which is the pattern this node is designed around - Samplerv2 is the clean middle of that chain: schedule in, extras in, guidance tuned, samples + denoised_samples out for WanAnimatePlus Decode. It's also the sampler that SamplerExtraArgs is explicitly built to feed. If you'd rather have steps on the same node as everything else, the v1 Sampler or Easy Sampler still do that.
Install via ComfyUI Manager (search "WanAnimatePlus") or git clone https://github.com/wuwukaka/ComfyUI-WanAnimatePlus into custom_nodes, restart. Fork-wide rules apply: WanAnimatePlus chain only, and don't mix this with the original wrapper's scheduler - the WANVIDEOSCHEDULER object is the fork's own type.
Inputs (20)
| Name | Type | Default | Description |
|---|---|---|---|
| model | WANVIDEOMODEL | — | |
| image_embeds | WANVIDIMAGE_EMBEDS | — | |
| cfg | FLOAT | 6.000–30 | — |
| seed | INT | 00–18446744073709550000 | — |
| force_offload | BOOLEAN | true | Moves the model to the offload device after sampling |
| scheduler | WANVIDEOSCHEDULER | — | |
| text_embedsopt | WANVIDEOTEXTEMBEDS | — | |
| samplesopt | LATENT | init Latents to use for video2video process | |
| add_noise_to_samplesopt | BOOLEAN | false | Add noise to the samples before sampling, needed for video2video sampling when starting from clean video |
| guidance_modeopt | COMBO | cfg | Guidance mode: cfg (standard CFG), apg (single-condition APG), apg_chain (image-reference APG chain), or cfg_chain (Bernini chained CFG) |
| apg_etaopt | FLOAT | 0.500–1 | APG: parallel/orthogonal balance (0=orthogonal only, 1=full) |
| apg_momentumopt | FLOAT | 0.00-1–1 | APG: EMA momentum for smoothing guidance differences |
| apg_norm_thresholdopt | FLOAT | 500–1000 | APG: L2 norm clipping threshold (0=disabled) |
| apg_omegaopt | FLOAT | 4.000–30 | APG: guidance strength |
| apg_omega_Iopt | FLOAT | 4.500–30 | APG chain: image-only guidance strength |
| apg_omega_TIopt | FLOAT | 4.000–30 | APG chain: text+image guidance strength |
| chain_omega_Vopt | FLOAT | 1.250–30 | Chain CFG: video-only guidance strength |
| chain_omega_Iopt | FLOAT | 4.500–30 | Chain CFG: extra reference context strength (VI - V; reference video/images) |
| chain_omega_TIopt | FLOAT | 4.000–30 | Chain CFG: text+image guidance strength |
| extra_argsopt | WANVIDSAMPLEREXTRAARGS | — |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| samples | LATENT | — |
| denoised_samples | LATENT | — |