CivArchive
    Flux 2 Dev (32B) Unquantized - Dariusz Zawadzki-style Biomechanical Surrealism - v1.0
    Preview 147225821
    Preview 147225908
    Preview 147225919
    Preview 147225933
    Preview 147225954
    Preview 147225959
    Preview 147225981
    Preview 147225953
    Preview 147225980
    Preview 147225962
    Preview 147225982

    LoRA Architecture: Flux 2 Dev (32B) | Precision: Pure bf16 (unquantized / no compromises) | Rank: 128

    This LoRA was trained on a pure, unquantized Flux 2 Dev (32B) stack, with no reduction in tensor precision.

    Rather than acting as a simple, flat stylizer, this model encapsulates the full aesthetic syntax of Dariusz Zawadzki’s dark biomechanical surrealism: dry, cracked bone, forged metal, patinated brass, organic tubing, and deep, layered oil glazes.

    Thanks to its high rank (128) and native bf16 precision, the model offers absolute flexibility depending on the applied LoRA weight.

    Weights

    Recommended LoRA weights and modes: 1.00

    "Pure Fine Art" Mode (Easel Painting)

    Use this for generating standalone, artistic canvases from scratch.

    Produces a dense, oily texture, featuring craquelure (cracked paint), sfumato, and a deep, noir-inspired aesthetic all without any artificial gloss.

    1.00 – 0.40

    VFX and Photographic Integration Mode (Blending with photo references)

    An ideal mode for working with portraits or faces via conditioning. At a weight of 0.20, the base 32B photorealism model retains 100% of the subject's anatomical and facial recognizability, while the LoRA applies effects reminiscent of high-budget American prosthetic makeup and practical VFX (such as fused metal, exoskeletal masks, or crumbling bone).

    Generation Settings

    Parameter

    Value

    Note

    Sampler: UniPC

    Ensures stable convergence for the 32B model.

    Scheduler: sg_uniform

    Provides a clean distribution of steps without aggressive denoising.

    Steps: 55

    Optimal rendering time for maximum detail.

    CFG / Guidance: 3.5 (Gold standard)

    3.5 preserves geometry; use 2.2–2.5 for more painterly/artistic experiments.

    Latent Resolution: 4MP (e.g., 2048x2048)

    Utilize the resolution conditioning from Flux 3.

    ComfyUI Workflow:

    All images in the gallery contain embedded metadata. Simply drag and drop any PNG/JPG file into ComfyUI to automatically load the complete workflow.

    Custom Detailer Node (included)

    Our custom node for micro-contrast and detail enhancement (designed for textures like bone and metal) is included as part of the workflow.

    Soft Clamp Toggle: OFF

    Soft Clamp is an experimental feature; in the current version, enabling it may limit the model's full potential, so it is recommended to keep it disabled.

    Description

    LORA
    Flux.2 D

    Details

    Downloads
    14
    Platform
    CivitAI
    Platform Status
    Available
    Created
    10/29/2026
    Updated
    10/31/2026
    Deleted
    -
    Trigger Words:
    r128f2dariuszv

    Files

    r128f2dariuszv.safetensors

    Mirrors

    CivitAI (1 mirrors)

    FLUX2D+REFERENS+LayerSteering+aspectRatio.json

    NODA_Orakul_Viking_Async.zip

    Mirrors

    CivitAI (1 mirrors)