Nodes/ComfyUI-LZNodes/LZ Anima Artist Options (Advanced)
ComfyUI Node

LZ Anima Artist Options (Advanced)

A ComfyUI node in LZNodes/Anima with 14 inputs and 1 output.

By liz-ils·Created 6 months ago·Updated 3 months ago· 1
LZ Anima Artist Options (Advanced)
    • advanced_options
    start_block0
    end_block-1
    start_percent0.000
    end_percent1.000
    normalize_weightstrue
    artist_ema_alpha0.00
    lowrank_k1
    artist_static_capturefalse
    static_capture_k6
    artist_anchor_qfalse
    anchor_seeds_count1
    anchor_user_blend0.00
    anchor_deep_layer_threshold-1
    layer_filter
    CategoryLZNodes/Anima

    Inputs (14)

    NameTypeDefaultDescription
    start_blockINT00–63Starting block (inclusive). 0 = first layer
    end_blockINT-1-1–63Ending block (inclusive). -1 = last layer
    start_percentFLOAT0.0000–1Sampling progress start. 0.0 = sampling start
    end_percentFLOAT1.0000–1Sampling progress end. 1.0 = sampling end
    normalize_weightsBOOLEANtrueTrue: weights normalized to relative proportions. False: weights used as independent strengths. If ::weight syntax is used in artist_chain, this switch is automatically disabled.
    artist_ema_alphaFLOAT0.000–0.95Cross-step EMA smoothing coefficient (only effective for fusion=interpolate). 0.0: off (default) 0.3-0.5: light smoothing 0.5-0.8: medium-heavy smoothing >0.8: strong smoothing Auto-reset on new sampling (sigma rising).
    lowrank_kINT11–32LoRA-style low-rank injection dimension (only effective for combine_mode=lowrank_avg). k=1: all artists along a single consensus direction k=2-3: retain main style directions (recommended) k>=N: equivalent to output_avg (no projection)
    artist_static_captureBOOLEANfalseH' cross-step temporal averaging: first K steps accumulate artist attention, then freeze from step K onwards.
    static_capture_kINT61–12H' accumulation steps (only when artist_static_capture=True). K=1: v18 single-point cache K=6: recommended default
    artist_anchor_qBOOLEANfalseUse fixed-seed anchor hidden state for artist cross-attn Q, fully decoupling from user seed. Pre-runs one model forward on first use (~1s overhead).
    anchor_seeds_countINT11–4Number of fixed seeds for anchor pre-run (only when anchor_q=True). More seeds = less systematic bias, more overhead.
    anchor_user_blendFLOAT0.000–1Anchor/user x blend ratio (only when anchor_q=True). Q = blend * user_x + (1-blend) * anchor_x 0.0: pure anchor; 1.0: pure user x.
    anchor_deep_layer_thresholdINT-1-1–64Only use anchor in shallow layers (only when anchor_q=True). -1: all layers use anchor (default) N>=0: layer idx < N uses anchor, idx >= N uses user x
    layer_filteroptSTRINGAdvanced layer selection (optional). Comma-separated block indices, supports ranges and negative indices. Example: '0,3,5-10,-1' Overrides start_block/end_block when set.

    Outputs (1)

    NameTypeDescription
    advanced_optionsANIMA_OPTS