CivArchive
    Preview 149522758
    Preview 149522762
    Preview 149522851
    Preview 149522900
    Preview 149523025

    This model created with generated images and it boosts the image with brushstrokes that giving your images the look and feel of a painting.

    Credits:

    All credits go to artists of the LORA's used! You can find all LORA's in the "suggested resources" section below!

    About this LORA:

    Memadukan is an indonesian term meaning "mixing things".

    This LORA is NOT based on a specific artist or style movement, but is literally mixed using result images from blending LORA's together! This is not equal to merging LORA's with tinkering on weights. Images are really generated, selected and retrained into a new LORA model that hopefully create something unique! The complete dataset got created with generated images from a FLUX model and FLUX LORA's. The results are both used to train the new FLUX and SDXL LORA.

    Versions:

    [ MOSS ]

    • v 1.0: This model struggles to stay consistent. It does not work on every merged MOSS model, so I suggest to use the original SFT model or the 4-bit NVFP4 quantized checkpoint. works best if steps are kept low 8-10 since MOSS jumps into detailing alot with more steps and this will undo the impasto effect. I might create a new version in the future with longer training to see if the result will become more consistent for MOSS.

    [ KREA2 ]

    • v 1.0: Works great with shorter prompts. Just prompt subject and leave out any other styling or detailing from your prompt or add brush strokes, impasto to strengthen the effect.

    [ FLUX ]

    • v 0.1 : Nice, but not very steady on generating the painting style on all prompts. However it generates more illustration alike results. Might require extra trigger words : "Brush Strokes", "Impasto"

    • v 0.5 : Good, Specifically on landscapes. Might generate malformed faces, portraits and animals, but great smooth brush strokes!

    • v 1.0 : Final, Generates more detailed and sharp visible strokes. Results might sometimes come out more like a drawing because of the detailed representation.

    [ SDXL ]

    • v 0.3 : Nice, it's more of a drawing style on this version. Uses trigger words : "Brush Strokes", "Impasto", "in memadukan_painter art style", "memadukan_painter". it is very steady on all prompts even prompts it was not trained on.

    • v 1.0 : Surprising, the sharp edges on the strokes give a sort of mosaic feel. Might be great for story book illustrations.

    FLUX LORA was trained on text prompt description

    SDXL LORA was trained on tags generated on the result images.

    Description

    Tested on "moss_image_sft_nvfp4" works great with very low steps 8-10, strength 1, config 1. Also tested it on custom merged base models and models that speed up steps, but those are trained on realistic images so it will produce a mix of impasto and photorealistic content mostly. This Lora might not mix very well with other Lora's aswel since I think MOSS prefers photorealistic output.

    FAQ

    LORA
    SDXL 1.0

    Details

    Downloads
    16
    Platform
    CivitAI
    Platform Status
    Available
    Created
    11/19/2026
    Updated
    11/20/2026
    Deleted
    -

    Files

    mossimage_lora_memadukan_brush_strokes.safetensors