CyberKrea Sampler

ComfyUI-CyberKrea-Sampler

<p align="center"> <img src="assets/icon.png" alt="CyberKrea Sampler icon" width="200"> </p>

A compact Krea 2 Turbo sampler and resolution-aware empty latent node for workflows that already patch the model with LoRAs and/or NegPiP.

One model connection keeps the model supplied by your LoRA/NegPiP chain active throughout sampling. Choose a preset, then adjust any control independently. CyberKrea uses its own node identifiers and category so it can coexist with other sampler packs.

The sampling engine is derived from ComfyUI-KreaPhoton by Kostiantyn Hrytsuk. CyberKrea adds a compact interface, a single model path, resolution selection and its own maintenance fixes.

Node

CyberKrea Sampler is under the CyberKrea category.

Connections

| Option | Description | |---|---| | model | The final Krea 2 model from your model chain. Connect it after LoRAs and/or NegPiP. | | positive | Positive conditioning from your prompt or NegPiP chain. | | latent_image | A 16-channel Krea 2 / Wan21 latent, such as the output from CyberKrea Empty Latent. | | seed | Controls the initial noise and all seeded sampling-noise streams for reproducible runs. | | negative | Optional negative conditioning. In cyberkrea schedule mode it enables the sigma-dependent CFG ramp described below. Ignored in reference mode, which uses CFG 1. Leave it disconnected when NegPiP already handles negative concepts. | | vae | Optional. Fully decodes the first image in the batch and displays a thumbnail on the node. This costs a full VAE decode for that image; the complete LATENT output is unchanged. |

Sampler controls

Selecting a preset immediately fills in all visible control values. You can then change any individual value without losing the rest of that preset.

| Option | Description | |---|---| | preset | Loads fast, balanced, quality, or reference values, including the schedule mode. Preview and denoise are preserved. | | steps | Number of denoising evaluations in the sampler loop, from 1 to 64. CFG and model patches can require additional work within each evaluation. | | sampler | euler is the first-order method used by all presets. euler_2m uses variable-step Adams–Bashforth history on consecutive deterministic steps; set eta0=0 to explore this mode. Restarts and stochastic steps reset its history. | | restart_frac | Fraction of the existing step budget reserved for the restart pass. Active only with schedule_mode=cyberkrea and denoise=1. Skipped if the budget cannot fit both descents. | | sigma_r | Noise level to jump back to at the restart. Higher values make the restart stronger. 0 disables restart and keeps the entire step budget for the first descent. | | plunge | Takes a direct end step before the lower-sigma restart refines the image. Active only with schedule_mode=cyberkrea and denoise=1. | | detail | Strength of the detail sigma adjustment. Higher values emphasize fine detail; excessive values can look harsh. | | eta0 | Maximum gated ancestral-noise strength during the middle of sampling. 0 disables this extra noise injection. | | sigma_gate | Eta noise is disabled at or below this sigma. Below 0.35, strength ramps up to eta0 at sigma 0.35. A gate of 0.35 or higher uses a hard cutoff. | | contraction | Multiplies the initial noise: 1.0 leaves it unchanged, 0.70 is the preset value, and 0 removes it. Restart and ancestral noise are separate. Very low values are experimental. | | preview_method | default follows ComfyUI's global preview setting. Explicit overrides are auto, latent2rgb, taesd, and none. TAESD requires lighttaew2_1 in models/vae_approx. The previous global setting is restored after sampling. | | schedule_mode | cyberkrea retains the existing schedule. reference uses the official Turbo uniform timestep grid with fixed mu=1.15, CFG 1 and no plunge/restart. Other manual controls still apply; select the reference preset to reset them to neutral values. | | denoise | 1.0 generates from noise. Values below 1 refine an existing or upscaled latent using the final requested steps of a longer, plain schedule, without plunge/restart. 0 returns the latent unchanged (an optional VAE thumbnail can still be decoded). |

Preset defaults:

| Preset | Steps | Detail | Sampler | |---|---:|---:|---| | fast | 8 | 0.50 | euler | | balanced | 12 | 0.60 | euler | | quality | 16 | 0.70 | euler | | reference | 8 | 0.00 | euler |

Fast, balanced and quality use restart_frac=0.25, sigma_r=0.65, plunge=true, eta0=1.0, sigma_gate=0.10, and contraction=0.70. Quality previously displayed euler_2m, but these stochastic defaults never used its second-order step; the updated sampler selection describes the same effective sampling behavior.

The reference preset selects schedule_mode=reference, restart_frac=0, plunge=false, detail=0, eta0=0, and contraction=1. With its default Euler sampler this provides a neutral baseline based on the official Krea 2 Turbo sampling code. The fixed shift is independent of image resolution. LoRAs, NegPiP and other patches on the supplied MODEL remain active; selecting this preset does not remove them. Changing its manual controls can move away from that baseline.

Refinement and latent upscaling

For a second pass, connect an already generated, VAE-encoded or enlarged latent:

first-pass latent -> SesquiLSR (Wan 2.1) -> CyberKrea Sampler -> VAE Decode

Start with preset=reference, steps=3 and denoise=0.30, using the same Krea 2 model and matching positive conditioning. This is a starting point for testing, not a guarantee of image quality. Lower denoise generally preserves more of the source; higher denoise allows more changes. An empty latent has no image to refine.

For 0 < denoise < 1, the schedule uses the final steps transitions from a plain schedule with int(steps / denoise) total steps. This happens in both schedule modes; plunge and restart are bypassed, while detail, eta and contraction remain manually controllable. Denoise is a schedule fraction, not a direct sigma value. Three steps at 0.30 still execute three denoising evaluations.

Preset changes preserve denoise. Set it back to 1.0 when returning to full text-to-image generation. No extra enhancer package is required.

At full denoise, one step always uses the full [1, 0] descent. A restart needs at least two steps without plunge, or three with plunge. Smaller budgets fall back to a single descent while preserving the requested evaluation count.

In cyberkrea schedule mode with a negative input connected, CFG is exactly 1.0 at sigma ≤ 0.7, smoothly rises to 2.25 between 0.7 and 0.9, and stays 2.25 above 0.9. This is a ramp with a plateau, not a bounded window. ComfyUI's CFG=1 optimization can apply at low sigma unless a model patch disables it. Without negative, the sampler uses CFG 1.0 throughout. The reference schedule always uses CFG 1.0 and ignores the negative input.

API and saved workflows

The frontend obtains preset and resolution tables from Python's node metadata; there are no separate JavaScript copies. Existing workflows keep their saved control values and explicit preview choices. To follow the global preview setting in an older workflow, choose default yourself.

ComfyUI API prompts must include the visible required controls. The supplied values take precedence: changing only the preset label does not overwrite them. Headless clients can obtain the templates from GET /object_info/CyberKreaSampler, at CyberKreaSampler.input.required.preset[1].cyberkrea_presets, and merge their overrides before submitting the prompt. Direct Python callers can use resolve_settings("quality", steps=20) to fill omitted values from a preset.

The new schedule_mode and denoise widgets are appended after the existing widgets. Older workflows and API prompts can omit them: existing presets retain the CyberKrea schedule and full denoise. A direct call with preset="reference" and no schedule override resolves to the reference schedule. Frontend preset selection fills schedule_mode visibly; it never silently rewrites a saved choice.

CyberKrea Empty Latent

CyberKrea Empty Latent creates the 16-channel latent expected by Krea 2 and outputs the selected width and height as integers. Choose a size tier first; the resolution dropdown then shows only the matching dimensions.

| Option | Description | |---|---| | size | Selects the S, M, L, or XL megapixel tier, filters the resolution dropdown, and preserves the selected aspect ratio. | | resolution | Selects the concrete width, height, and aspect ratio within the chosen tier. | | batch_size | Number of empty latents generated in one batch. Higher values require more VRAM. |

| Aspect | S (~1.0 MP) | M (~1.4 MP) | L (~1.7 MP) | XL (~2.1 MP) | |---|---|---|---|---| | 1:1 | 1024x1024 | 1184x1184 | 1312x1312 | 1440x1440 | | 4:3 | 1152x864 | 1344x1008 | 1504x1120 | 1664x1248 | | 3:4 | 864x1152 | 1008x1344 | 1120x1504 | 1248x1664 | | 3:2 | 1344x896 | 1568x1040 | 1600x1088 | 1776x1184 | | 2:3 | 896x1344 | 1040x1568 | 1088x1600 | 1184x1776 | | 16:9 | 1344x768 | 1568x880 | 1728x960 | 1920x1088 | | 9:16 | 768x1344 | 880x1568 | 960x1728 | 1088x1920 | | 21:9 | 1568x672 | 1792x768 | 2016x864 | 2240x960 | | 9:21 | 672x1568 | 768x1792 | 864x2016 | 960x2240 | | 5:4 | 1120x896 | 1360x1088 | 1440x1152 | 1600x1280 | | 4:5 | 896x1120 | 1088x1360 | 1152x1440 | 1280x1600 |

Every dimension is divisible by 16 for the Wan21 VAE and Krea 2 patch layout. Some original buckets approximate their displayed aspect ratio; the added 21:9, 9:21, 5:4 and 4:5 buckets are exact. These additions extend the available canvas sizes; their image quality has not been validated through generation tests.

NegPiP wiring

base model -> LoRA(s) -> NegPiP model output -> CyberKrea Sampler (model)
prompt/NegPiP positive conditioning ----------> CyberKrea Sampler (positive)
CyberKrea Empty Latent -----------------------> CyberKrea Sampler (latent_image)

Leave the sampler's negative input empty when NegPiP is already handling negative concepts.

Installation

Copy the complete ComfyUI-CyberKrea-Sampler folder to:

ComfyUI/custom_nodes/ComfyUI-CyberKrea-Sampler

Restart ComfyUI. No extra Python packages are required beyond ComfyUI's own PyTorch/Comfy modules.

After updating, also refresh the browser so it loads the matching frontend extensions and node metadata. See CHANGELOG.md for changes.

Development checks

With Python 3.11+, PyTorch, NumPy and Node.js 22+ installed:

python -B -m unittest discover -s tests -v

Tests cover schedule budgets, tensor-level sampling, seed reproducibility, guidance, previews, validation and frontend widget behavior. ComfyUI boundaries use test doubles; image quality and third-party patch compatibility still need generation tests with a real Krea 2 checkpoint.

License

Released under the MIT license. The original Kostiantyn Hrytsuk copyright notice is retained alongside Cyberdelia's notice. See LICENSE.