Nodes/ComfyUI-WanAnimatePlus/WanAnimatePlus SamplerSettings
ComfyUI Node

WanAnimatePlus SamplerSettings

All the APG and chain guidance, none of the preview

By wuwukaka·Created 4 months ago·Updated about a month ago· 409
WanAnimatePlus SamplerSettings
  • model
  • image_embeds
  • text_embeds
  • samples
  • feta_args
  • context_options
  • cache_args
  • flowedit_args
  • slg_args
  • loop_args
  • experimental_args
  • sigmas
  • unianimate_poses
  • fantasytalking_embeds
  • uni3c_embeds
  • multitalk_embeds
  • freeinit_args
  • sampler_inputs
steps30
cfg6.00
shift5.00
seed0
force_offloadtrue
schedulerunipc
riflex_freq_index0
denoise_strength1.00
batched_cfgfalse
rope_functioncomfy
start_step0
end_step-1
add_noise_to_samplesfalse
guidance_modecfg
apg_eta0.50
apg_momentum0.00
apg_norm_threshold50
apg_omega4.00
apg_omega_I4.50
apg_omega_TI4.00
chain_omega_V1.25
chain_omega_I4.50
chain_omega_TI4.00

Easy SamplerSettings is the tidied-down settings node for people who want the seven common controls. SamplerSettings is its fully-loaded sibling: every input the big Sampler has, but instead of sampling, it packages them into a SAMPLER_ARGS output for WanAnimatePlus SamplerFromSettings to execute. One node for tuning, another for running, and you never lose access to the advanced guidance modes.

The advanced controls carry over wholesale. guidance_mode gives you the four-way choice - cfg, apg, apg_chain, cfg_chain - with the full APG parameter set (apg_eta, apg_momentum, apg_norm_threshold, apg_omega, apg_omega_I, apg_omega_TI) and the Bernini chain omegas (chain_omega_V, chain_omega_I, chain_omega_TI). If you're running Bernini, this is where you set the values its node's recommended_guidance output spells out: r2v gets apg_chain with apg_omega_I=4.5 and apg_omega_TI=4.0, rv2v gets cfg_chain. riflex_freq_index (0 off, 6 reference) for non-looping frame extension, start_step/end_step for schedule slicing, denoise_strength and add_noise_to_samples for v2v, batched_cfg for the speed/memory trade, rope_function for compilable attention, and the usual optional bundles (context_options, cache_args, slg_args, loop_args, sigmas, and the model embeds like uni3c_embeds or unianimate_poses). flowedit_args is present but deprecated per its tooltip - ignore it.

The honest use case. This is the node for people who want the full sampler's power and a readable graph. If you only ever touch steps, cfg, shift, seed and scheduler, the Easy version is the right call - fewer places to fat-finger. If you're juggling Bernini guidance modes across experiments, SamplerSettings + SamplerFromSettings means your guidance config lives in one panel you can bookmark, while the preview lives somewhere it can breathe.

The traps. Same as its siblings: sampler_inputs must actually reach a SamplerFromSettings, or the run executes with defaults you never set. And don't run this path alongside a direct Sampler in the same generation - pick one sampling path. Finally, the Easy SamplerSettings and this node both emit SAMPLER_ARGS, but only the full version carries guidance mode and the APG/chain fields, so a Bernini workflow wired through the Easy node quietly loses its cfg_chain. That's the case for using this one.

Install via ComfyUI Manager (search "WanAnimatePlus") or git clone https://github.com/wuwukaka/ComfyUI-WanAnimatePlus into custom_nodes, restart, and keep the chain on WanAnimatePlus nodes so the SAMPLER_ARGS layout stays consistent.

CategoryWanAnimatePlus

Inputs (40)

NameTypeDefaultDescription
modelWANVIDEOMODEL
image_embedsWANVIDIMAGE_EMBEDS
stepsINT30
cfgFLOAT6.000–30
shiftFLOAT5.000–1000
seedINT00–18446744073709550000
force_offloadBOOLEANtrueMoves the model to the offload device after sampling
schedulerCOMBOunipc21 options: unipc, unipc/beta, dpm++, dpm++/beta, dpm++_sde, dpm++_sde/beta, +15
riflex_freq_indexINT00–1000Frequency index for RIFLEX, disabled when 0, default 6. Allows for new frames to be generated after without looping
text_embedsoptWANVIDEOTEXTEMBEDS
samplesoptLATENTinit Latents to use for video2video process
denoise_strengthoptFLOAT1.000–1
feta_argsoptFETAARGS
context_optionsoptWANVIDCONTEXT
cache_argsoptCACHEARGS
flowedit_argsoptFLOWEDITARGSFlowEdit support has been deprecated
batched_cfgoptBOOLEANfalseBatch cond and uncond for faster sampling, possibly faster on some hardware, uses more memory
slg_argsoptSLGARGS
rope_functionoptCOMBOcomfyComfy's RoPE implementation doesn't use complex numbers and can thus be compiled, that should be a lot faster when using torch.compile. Chunked version has reduced peak VRAM usage when not using torch.compile
loop_argsoptLOOPARGS
experimental_argsoptEXPERIMENTALARGS
sigmasoptSIGMAS
unianimate_posesoptUNIANIMATE_POSE
fantasytalking_embedsoptFANTASYTALKING_EMBEDS
uni3c_embedsoptUNI3C_EMBEDS
multitalk_embedsoptMULTITALK_EMBEDS
freeinit_argsoptFREEINITARGS
start_stepoptINT00–10000Start step for the sampling, 0 means full sampling, otherwise samples only from this step
end_stepoptINT-1-1–10000End step for the sampling, -1 means full sampling, otherwise samples only until this step
add_noise_to_samplesoptBOOLEANfalseAdd noise to the samples before sampling, needed for video2video sampling when starting from clean video
guidance_modeoptCOMBOcfgGuidance mode: cfg (standard CFG), apg (single-condition APG), apg_chain (image-reference APG chain), or cfg_chain (Bernini chained CFG)
apg_etaoptFLOAT0.500–1APG: parallel/orthogonal balance (0=orthogonal only, 1=full)
apg_momentumoptFLOAT0.00-1–1APG: EMA momentum for smoothing guidance differences
apg_norm_thresholdoptFLOAT500–1000APG: L2 norm clipping threshold (0=disabled)
apg_omegaoptFLOAT4.000–30APG: guidance strength (equivalent to CFG scale, replaces cfg when APG is active)
apg_omega_IoptFLOAT4.500–30APG chain: image-only guidance strength
apg_omega_TIoptFLOAT4.000–30APG chain: text+image guidance strength
chain_omega_VoptFLOAT1.250–30Chain CFG: video-only guidance strength
chain_omega_IoptFLOAT4.500–30Chain CFG: extra reference context strength (VI - V; reference video/images)
chain_omega_TIoptFLOAT4.000–30Chain CFG: text+image guidance strength

Outputs (1)

NameTypeDescription
sampler_inputsSAMPLER_ARGS