CivArchive
    Full-body photoreal generation on Flux.2 Klein 9B Distilled, conditioned on 4 reference photos. No LoRAs. - v1.0
    NSFW
    Preview 144959124

    This workflow generates full-body, extremely detailed male nudes (210 cm bodybuilder) in only 4 steps using a distilled Klein 9B / Flux2 setup.

    Core feature: the ImpactWildcardProcessor randomly selects a complete scenario each run — body pose and angle, facial expression, location (gym, rooftop, pool deck, garage, loft, locker room, etc.), and highly explicit, varied descriptions of the penis, foreskin, glans, balls, sweat, saliva and body hair.

    The selected text becomes the final prompt, which is then combined with 4 reference images:

    face identity

    body mass

    thighs / lower body

    skin + body hair

    The final character is effectively created as a mix of information coming from all four references. Because several different source images are being combined into one body, occasional anatomical conflicts can still appear, especially with complex poses or strongly different reference images.

    If you notice deformations such as duplicated arms, extra limbs or malformed genitals, try using the same reference image for both Body Mass and Skin + Hair. Using one consistent body reference for these two inputs reduces conflicting structural information and may significantly improve anatomical consistency.

    Completely eliminating these defects would require a much heavier workflow with additional refinement or correction passes, which would also make generation substantially slower. This workflow is intentionally optimized for a strong balance between quality, reference control and very high generation speed.

    The workflow runs at CFG 1 and is optimized for very fast generation.

    Full image generation takes around 15 seconds on an RTX 5080.

    Ultra-fast workflow focused on high detail, strong reference influence and varied full-body results.

    Description

    1.0

    FAQ

    Comments (1)

    altoiddealerOct 7, 2026· 1 reaction
    CivitAI

    Example output appears to be only 70% of body