CivArchive
    Lip Fullness Slider - Krea 2 - v1.0
    Preview 142003885
    Preview 142003900
    Preview 142003876
    Preview 142003862
    Preview 142003867

    A concept slider for lip thickness. Negative weights thin the lips toward a narrow line, positive

    weights fill them out. It is a direction, not a look - there is no trigger word, and nothing is

    added to your prompt. Set the weight and render.

    It works across subjects and media: men and women, young and old, photographic, painted and cartoon.

    The mouth is the only thing it is asking the model to change.

    HOW TO USE

    - No trigger word. Load it and set the weight.

    - Effective range: -4 to +4. The effect is gentle inside ±2 and does most of its work between ±2

    and ±4.

    - Start at +2 or -2, then push. 0 is the base model untouched.

    - Works on any subject with a visible mouth. A tight crop on the face shows it best.

    ONE TIP THAT MATTERS

    A slider moves what your prompt leaves open. If you want the lips and ONLY the lips to change, name

    the things that should stay put - the person's age, the hair, the lip color , Prompts that already

    say "a 45-year-old woman with fine lines", "short cropped dark hair" or "bare lips in a pale dusty

    rose tone" hold those steady across the whole range while the thickness moves freely.

    Trained on Krea 2. Rank 4. No dataset of a person, no identity baked in.

    Description

    v1.0 · rank 4 · trained on Krea-2-Raw · no trigger word

    EFFECTIVE RANGE
    -4 to +4. Inside ±2 the change is subtle; most of the visible movement happens between ±2 and ±4.
    Values beyond ±4 are not recommended - a LoRA past its range degrades the picture rather than doing more

    FAQ

    Comments (4)

    boomhildeSep 6, 2026· 1 reaction
    CivitAI

    Does what it says, and I'm happy to see someone entering the Krea2 slider arena. Nice work! Only issues I had were that it seems to square/elongate the head shape a bit at low/high weights, and somehow ears become more prominent at low weight? But I did not heavily define the face and it's more pronounced at weights outside the suggestion [-6/+6] so I can't say that it couldn't be corrected via prompt.

    shakir_abdo
    Author
    Sep 6, 2026

    I'm glad that you liked the adapter. As for the issues you mentioned, I'll try to make more experiments and perhaps update the model.

    boomhildeSep 10, 2026

    @shakir_abdo On followup, now I'm not actually sure that's what it's doing - still works great for the lips but my second test has more face alteration than the first time. But again the faces aren't heavily defined in the prompt so this may be a 'me' problem. Sliders seem to be particularly tricky to train especially with how quickly Krea2 picks up concepts & also possible that the turbo LoRA I use has an effect. I put up a comparison for the four sliders you've posted so far, & no shade meant it's just what I'm seeing. Same prompt/seed/sampler/scheduler/steps for all. Overall your sliders are great & appreciated, & I don't generally do character LoRAs so any changes are nbd. (I have not tested with character LoRAs, I just mean that the shifts I see still work for what I'm prompting.)

    shakir_abdo
    Author
    Sep 10, 2026

    @boomhilde No shade taken at all! I really appreciate you taking the time to test them so thoroughly and share the comparison, and yeah, I agree that sliders can be particularly tricky to train.

    The unwanted shifts that’s actually something I noticed while comparing different training steps and weights. There’s always a trade-off between making the slider strong enough to have a noticeable effect and keeping it from altering the subject’s identity too much. I’m basically trying to find that sweet spot in the middle. 😄 Thanks again for all the testing and detailed feedback! ❤️

    LORA
    Krea 2

    Details

    Downloads
    1,072
    Platform
    CivitAI
    Platform Status
    Available
    Created
    9/6/2026
    Updated
    10/22/2026
    Deleted
    -

    Files

    lips_fullness_slider.safetensors

    Mirrors