CivArchive
    Kurzgesagt Art Style FLUX - FLUX v1.0
    NSFW
    Preview 30310286
    Preview 30313789
    Preview 30312152
    Preview 30310064
    Preview 30315861
    Preview 30310154
    Preview 30310054
    Preview 30310170
    Preview 30310232
    Preview 30310253
    Preview 30313759
    Preview 30310261
    Preview 30310319
    Preview 30310267
    Preview 30310397
    Preview 30310564
    Preview 31357780

    Please Donate Buzz for FLUX Lora Training !

    FLUX v1.0:

    Style Lora trained on 106 images with the Kurzgesagt YouTube channel (@kurzgesagt) style, kindly provided by https://civarchive.com/user/chilon249 (Check out his SD 1.5 Kurzgesagt Lora) and captioned in detail by GPT-4 vision and curated. Trained on FLUX 1.D. Tested on FLUX 1.D [full], NF4 and fp8! Use with keywords "Kurzgesagt style", "vector graphics illustration". Use with strength 0.7-1.0 depending if you use additional Lora resources with it. Positive prompt can be something like:

    Positive prompt: Kurzgesagt style illustration, {describe the scene or character in full sentence}, <lora:Kurzgesagt_style_vector_graphics_FLUX_v1_r1-000031:0.85>

    Description

    FAQ

    Comments (8)

    RudyBagaSep 20, 2024
    CivitAI

    I find that the images are a bit more cohesive with the Lora weighted 0.8-0.9, but a terrific LoRA overall!

    deep_synthSep 20, 2024

    Found 0.8 good too

    steffangund
    Author
    Sep 20, 2024· 1 reaction

    Thanks for trying it out!

    lesjoMar 25, 2025

    I found I had to crank the weight of the LoRA up much higher than 1. Somewhere between 2 and 2.5 seems best for balancing "it didn't do anything" with "it's burning the image". There will still be examples of both in my output but there will also be some that worked well.

    It is possible that this is affected by having the Guidance setting at 3 rather than the default of 1, but I don't have to crank up the weight of other LoRAs when I dial up the Guidance. It's not a big deal regardless, just be prepared to expand the range of weights you try.

    steffangund
    Author
    Mar 25, 2025

    @lesjo If you are using FLUX in Forge it is recommended to use 1 as the guidance and slightly vary the CFG ditilled I think. So that could be the reason.

    lesjoMar 27, 2025

    @steffangund I'm using ComfyUI, and I don't even know what CFG Distilled is or where I'd add it to the workflow. I know the normal CFG is useless as it always has to be set to exactly 1 (which is also why Flux ignores negative prompts), and was told Guidance is the parameter that effectively replaces CFG. I think just cranking up the weight of the LoRA works well enough that I'm not going to go changing the whole workflow to implement something else.

    steffangund
    Author
    Mar 27, 2025

    @lesjo Fair enough as long as it works

    doctorhejjDec 23, 2025
    CivitAI

    I'm getting great results with .9. Thanks.