ComfyUI Node
⚗ Alchemica KSampler (Krea 2, all-in-one)
AlchemicaSampler in a single node: the knobs that matter, the rest from the calibration profile.
⚗ Alchemica KSampler (Krea 2, all-in-one)
- model
- positive
- latent_image
- negative
- vae
- refiner_model
- opts
- latent
- sigma_plot
- report
- image
◄seed0►
◄profile▾►
◄modebalanced►
◄two_stagetrue►
◄redraw-1.00►
◄detail0.12►
◄variety0.60►
◄denoise1.00►
◄contraction0.85►
◄steps-1►
◄cfg-1.00►
◄preview_methodlatent2rgb (fast)►
◄degrid0.0►
Category⚗ AlchemicaMente/sampling
Inputs (20)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| positive | CONDITIONING | — | |
| latent_image | LATENT | At the final resolution (⚗ Resolution node) | |
| seed | INT | 00–18446744073709550000 | — |
| profile | COMBO | (auto) recognises the model and uses its calibration profile | |
| mode | COMBO | balanced | How many steps: veloce (fast) ~30 s, equilibrato (balanced) ~45 s, massimo dettaglio (max detail) ~70 s at 3 MP |
| two_stage | BOOLEAN | true | Composes at half resolution and refines at full resolution. Required above ~1.5 MP |
| redraw | FLOAT | -1.00-1–0.95 | How freely the second stage may redraw. -1 = from the profile (recommended). Higher = more new detail, but the composition may change |
| detail | FLOAT | 0.120–0.25 | Micro-relief of skin and fabric. 0.12 is the validated value; above 0.2 coloured speckles come back |
| variety | FLOAT | 0.600–1.2 | Ancestral noise: texture and seed-to-seed variety. 0 = deterministic. 2026-10-05: on checkpoint+LoRA combos with a strong LoRA, above ~0.3 this can produce wet-skin/droplet artifacts ('gocce') regardless of where it's gated — if you see that, lower this to 0.15-0.2 first (confirmed clean at 0.15 on a fine-tuned checkpoint with a strong LoRA at x2) |
| denoise | FLOAT | 1.000–1 | 1.0 = from scratch. <1 = img2img (disables two-stage mode) |
| contraction | FLOAT | 0.850.3–1.2 | Scale of the initial noise. 1 = standard ComfyUI noise. 2026-10-05: NOT a cause of 'gocce' by itself (tested at 1.0 with eta 0.6: gocce unchanged), but at LOW varieta (~0.3) the default 0.85 is measurably cleaner than 1.0 — leave at 0.85 unless you have a specific reason to change it |
| steps | INT | -1-1–50 | Step override. -1 = from 'modo'. Single-stage: total steps. Two-stage: overrides only stage 2 (refinement); stage 1 stays tied to the calibration profile |
| cfg | FLOAT | -1.00-1–20 | CFG override, flat (no high/low schedule). -1 = 1.0 (Turbo default: negative skipped, 0 extra cost) |
| negativeopt | CONDITIONING | Used only if you set cfg > 1 | |
| vaeopt | VAE | Optional: the final image is decoded and SHOWN in this node, and returned on the 'image' output (like the Efficiency KSampler) | |
| refiner_modelopt | MODEL | Optional: composes with the main model and REFINES with this one (e.g. anatomy from a fine-tune, skin from base Krea 2). Requires two-stage mode | |
| optsopt | ALCHEMICA_OPTS | For tinkerers: overrides everything with the option nodes (including contraction/steps/cfg set here) | |
| preview_methodopt | COMBO | latent2rgb (fast) | Live preview while sampling. latent2rgb = instant, a bit soft; taesd = sharper (needs the TAE model in models/vae_approx, otherwise latent2rgb); 'ComfyUI setting' = whatever ComfyUI is set to (often none) |
| degridopt | FLOAT | 0.00–10 | Only with a VAE connected: removes the faint 2-pixel line grid of Krea 2's VAE from the image shown and returned here (the latent is untouched). Max correction in grey levels: 2 = recommended, 0 = off |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| latent | LATENT | — |
| sigma_plot | IMAGE | — |
| report | STRING | — |
| image | IMAGE | — |