Nodes/DDRK-Omega-Sampler/DDRK Omega Sampler
ComfyUI Node

DDRK Omega Sampler

A ComfyUI node in sampling/custom_samplers with 23 inputs and 1 output.

By HVOSTOVSKY·Created 22 days ago·Updated 12 days ago· 0
DDRK Omega Sampler
    • SAMPLER
    ◄integratorauto►
    ◄sde_strength0.08►
    ◄sharpness0.30►
    ◄saber_fusion0.30►
    ◄saber_modeauto►
    ◄use_ema_sabertrue►
    ◄ema_decay0.70►
    ◄dyn_thresh_percentile0.995►
    ◄latent_rescale0.00►
    ◄hc2_corrector0.00►
    ◄sigma_adapt0.00►
    ◄hc2_max_order2►
    ◄limiter_kappa1.00►
    ◄momentum_beta0.25►
    ◄sde_seed-1►
    ◄s_churn0►
    ◄s_tmin0.0►
    ◄s_tmax999999►
    ◄s_noise1.00►
    ◄content_awaretrue►
    ◄auto_optimizetrue►
    ◄debug_modefalse►
    ◄debug_tag►
    Categorysampling/custom_samplers

    Inputs (23)

    NameTypeDefaultDescription
    integratorCOMBOauto5 options: auto, hc2, rk4, heun, euler
    sde_strengthFLOAT0.080–0.5Stochastic noise. 0 = deterministic. FM only.
    sharpnessFLOAT0.300–1.5Final sharpening. Auto-reduced for few-step.
    saber_fusionFLOAT0.300–1Stabilization. Auto-capped to 0.15 for FM photo models.
    saber_modeCOMBOautoAuto detects video by 5D latent with >1 frame.
    use_ema_saberBOOLEANtrue—
    ema_decayFLOAT0.700–0.99—
    dyn_thresh_percentileFLOAT0.9950.9–1Dynamic thresholding percentile. 1.0 = disabled. Universal (FM + EDM).
    latent_rescaleFLOAT0.000–1Attenuates latent values far from the per-image mean (beyond ~2 std). 0 = disabled. NOTE: this is NOT classical CFG-rescale (Lin et al.), which compares conditional vs unconditional predictions — those are already combined before this sampler is called, so that method cannot be implemented here. Renamed from 'cfg_rescale' to stop implying otherwise.
    hc2_correctorFLOAT0.000–1HC2 selective corrector. 0 = off. Otherwise, on any step where the high-order correction exceeds this fraction of the first-order step, HC2 spends a SECOND model call to re-evaluate the denoiser at the step's endpoint and redo the step with an interpolated slope instead of an extrapolated one. Costs one extra call per firing. Small values (0.01-0.05) fire almost every step; larger values fire only on difficult steps. EXPERIMENTAL: strong on synthetic tests, unvalidated on images.
    sigma_adaptFLOAT0.000–0.5HC2 adaptive step placement. 0 = off. Otherwise the sampler may move the intermediate sigma values by up to this fraction to equalise estimated error across steps. Step COUNT, start and terminal zero are unchanged, so timing and the progress bar are unaffected. EXPERIMENTAL and the most speculative option here: schedules are already heavily tuned per model family, and nudging them may fight that tuning. Try 0.10-0.20.
    hc2_max_orderINT21–3HC2 maximum order. 2 = second order, one model call per step (default). 3 = allow third order: uses two past denoiser evaluations, spends ONE extra model call on the first step to bootstrap (a multistep method's first step is otherwise first-order and caps the whole run's accuracy), and falls back to second order on any step where the expansion stops converging. Ignored by every other integrator.
    limiter_kappaFLOAT1.000.1–3HC2 slope limiter. Caps the 2nd-order correction at this multiple of the 1st-order step, per element. 1.0 = the correction may at most double or cancel the step, never reverse it. Lower it (0.5-0.7) if high CFG still blows out highlights; raise it toward 2-3 to let HC2 run closer to unlimited 2nd order on smooth content. Ignored by every other integrator.
    momentum_betaFLOAT0.250–0.8Velocity EMA. Reduces oscillation. 0 = disabled.
    sde_seedINT-1-1–18446744073709550000SDE noise seed. -1 = random (non-deterministic).
    s_churnFLOAT00–100EDM churn (Karras Alg 2). 0 = off. Try 5-15 for SDXL. EDM models only; silently ignored for Flow Matching.
    s_tminFLOAT0.00–999999EDM churn lower sigma bound (Karras Alg 2). Churn only fires while s_tmin <= sigma <= s_tmax. 0.0 = no lower bound.
    s_tmaxFLOAT9999990–999999EDM churn upper sigma bound. Default (max) means unbounded, matching Karras Alg 2's s_tmax=inf. Lower it to restrict churn to high-sigma steps only.
    s_noiseFLOAT1.000–2EDM churn noise multiplier. 1.0 = standard.
    content_awareBOOLEANtrueContent-aware SABER — edge-gated fusion. Disable for pixel-art/flat styles.
    auto_optimizeBOOLEANtrueAuto-disable SABER/SDE/momentum and force euler for FM few-step. Does NOT set steps/cfg.
    debug_modeBOOLEANfalseWrite a per-step diagnostics log (JSON+CSV+summary) to your ComfyUI output folder. Off by default; adds overhead only when on.
    debug_tagSTRINGOptional label included in the debug log filename, e.g. 'test1'.

    Outputs (1)

    NameTypeDescription
    SAMPLERSAMPLER—