CivArchive
    SweetSpot-4x - v1.0
    Preview 143341610
    Preview 143341608

    Epoch 50 model is a little crisper without over doing it and keeping the organic vibe.

    Model Overview

    This is a custom-trained RRDBNet upscaler built specifically for a cinematic, organic aesthetic. Unlike standard commercial upscalers (like 4x-UltraSharp) that rely on aggressive edge sharpening or generic benchmark metrics, this model is custom-tailored to preserve natural micro-contrast, realistic atmospheric falloff, and true textural weight.

    What It Excel At (Best Use Cases)

    Forests & Foliage: Renders organic matter, moss, bark, and leaves with natural depth, avoiding the "deep-fried" artificial crunch and halo artifacts common in standard models.

    Architecture & Water: Preserves complex highlights, specular reflections, and smooth lighting gradients without blowing out contrast or flattening out dynamic range.

    Organic Materials: Retains tactile grain and soft atmospheric haze, giving images a cohesive, filmic weight rather than a sterile digital look.

    What It Is NOT Built For

    Synthetic / Flat Graphic Design: If you are upscaling crisp vector logos, clean UI elements, or flat anime line art, a generic sharpening model will give you sharper geometric edges.

    Aggressive Edge Crunch: It will not artificially manufacture high-contrast lines where none exist; it favors realistic micro-detail over synthetic sharpness.

    Description

    Comments (1)

    LigH_deSep 21, 2026· 1 reaction
    CivitAI

    I believe your uploads are broken if they are reported as only 16 bytes large.

    Upscaler
    Upscaler
    by onespirit5777
    Download (Beta) View on CivitAI
    architecture
    style
    fantasy
    water
    forest

    Details

    Downloads
    4
    Platform
    CivitAI
    Platform Status
    Available
    Created
    9/20/2026
    Updated
    9/23/2026
    Deleted
    -

    Files

    sweetspot4x_v10.safetensors

    Size:
    16.00 B
    SHA256:
    9bbcbf73561f6bc5d0a17ea6a2081feed2d1304e87602d8c502d9a5c4bd85576

    Mirrors

    HuggingFace (633 mirrors)
    CivitAI (18 mirrors)
    TensorFiles (1 mirrors)
    ModelScope (9 mirrors)
    ModelScope CN (7 mirrors)
    TensorArt (1 mirrors)