ComfyUI Node
Helm Apply (strength x axis)
A ComfyUI node in Helm with 14 inputs and 1 output.
Helm Apply (strength x axis)
- model
- model
◄axisaxis►
◄block20►
◄strength0.10►
◄dial0.00►
◄cond_onlyfalse►
◄start_percent0.00►
◄end_percent1.00►
◄modepush►
◄mask►
◄layoutnone►
◄grid►
◄clamp_k1.0►
◄mask_offset0►
CategoryHelm
Inputs (14)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| axis | STRING | axis | — |
| block | INT | 20-1–512 | -1 = adopt the block recorded inside the axis file (the block it was built at); otherwise this explicit index |
| strength | FLOAT | 0.10-10–10 | scale of the injected axis. Model-specific; see calibration.json |
| dial | FLOAT | 0.000–1 | optional normalized 0-1 dial. If >0 and the axis recorded a safe_max, strength = dial * safe_max (keeps you inside the verified band). |
| cond_only | BOOLEAN | false | Default False preserves the published result. True (recommended for CFG models) adds the axis only to cond batch items; current axes were NOT measured with cond_only. |
| start_percent | FLOAT | 0.000–1 | sampling-fraction window start. Default 0.0 preserves the published result. |
| end_percent | FLOAT | 1.000–1 | sampling-fraction window end. Default 1.0 preserves the published result. |
| modeopt | COMBO | push | push = published x+strength*axis. norm = same, rescaled to the token's original norm. clamp = move the axis projection, clamped to +/- clamp_k*||x||. |
| maskopt | STRING | path to a black/white PNG. White = apply the axis there (requires layout + grid). Empty = global (published behavior). | |
| layoutopt | COMBO | none | token order of the block output. h_w = image; t_h_w = frame-major video; h_w_t = frame-inner. Never guessed: the mask is refused if the grid does not multiply to the token count. |
| gridopt | STRING | explicit H,W (image) or H,W,T (video). Required whenever a mask is given; the grid is never inferred. | |
| clamp_kopt | FLOAT | 1.00.1–10 | clamp mode only: projection limit as a multiple of each token's norm. |
| mask_offsetopt | INT | 00–1000000 | leading packed tokens to skip before the mask. Z-Image packs 32 text/cond rows before the 4096 spatial tokens; H3 packs [text|cond|audio|video], so pass the row count before the video slice. Requires mask. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| model | MODEL | — |