CivArchive
    NoobAI "Novel" Style/Quality Lora - v1.5 Alpha
    NSFW
    Preview 42703401
    Preview 42703398
    Preview 42703408
    Preview 42703415
    Preview 42703403
    Preview 42703410
    Preview 43065805
    Preview 43065803
    Preview 43065804
    Preview 43065806
    Preview 49819529
    Preview 49819638
    Preview 49819664
    Preview 49819665
    Preview 49819705
    Preview 49819706
    Preview 49819763
    Preview 49821445
    Preview 49821490
    Preview 49821504

    6/01/2025 Update

    Novel is a quality and Style Lora trained on fully synthetic self-generated high-quality images. Novels enforce a Smooth-lined Sudo 3D anime style with quality enhancement features such as larger attention to detail for anatomy, eyes, hands, and genitalia on both Furry and humanoid characters.

    This Lora is currently a work in progress and has minor baked-in bias. I'm generating a much larger dataset that has been taking a large amount of time to ensure that most, if not all, bias is removed.

    Trained on NoobAI V1.0 and tested on ChromaXL Mix "Mango" https://civarchive.com/models/907792?modelVersionId=1024571

    Version 1.5 Update (2025 Update)
    For the best results, I highly suggest using ChromaXL Mix V4 Sorbert with this Lora. These settings guidelines are intended to use the model "ChromaXL mix Sorbert"
    Lora Strength 0.3 to 1.0 any higher will start to cause style issues.
    CFG 1.7-3
    Euler A with (AYS) Align your steps or Beta as the scheduler at 12-18 steps. For those not using comfy as there main SD model, AYS11 is recommend on Forge or Re-forge
    All normal 1K res formats work fine on it. E.g. Square, landscape, portrait.

    Recommended Postive masterpiece, best quality, newest, absurdres, highres, n0ai,
    Recommended Negative worst quality, bad quality, grainy, noisy, render, film grain, text, sketch, black-and-white, group, (watermark:1.2), (text:1.2), (comic:1.2),

    After a month of testing, I can confirm this Lora will work with most character Lora's, some weights may need to be adjusted to reduce the chance of style bleed. It will also work well with other Quality enforcement Lora's.

    Currently Supports all versions of NoobAI Epsilon-Pred and all forked models using NoobAI as it's base. I have tested on V-pred models and even though it has some level of impact I would still not recommend its usage on any V-pred type models.

    Version 2 is currently under development with triple the data size of Version 1.5 and will include a huge array of new concepts to reduce bias and add fixes to flaws that NoobAi models currently such as certain body and fur types certain anatomy and other popular mediums that NoobAi struggles with. The estimated timeframe is hopefully by the end of the month.

    Description

    NAI PRO V1.5 Now with better human, Scalie, and overall Anthro Support with less baked-in bias and with improved Lora weight control, Still far from perfect but I wanted to share an improved version while I work on increasing my overall dataset which may take some time. I hope you enjoy and if you do feel free to leave me a like :3

    A In-between version, with the following improvements.

    Reduced baked-in Bias for Anthro, Feral, and Human Generations.
    Reduced network overhead to allow greater strength control now recommended Weight guidelines is from 0.1 to 0.7
    Generating duo causes fewer artifacts and is less prone to fail.
    Bleed effects from the training data are less likely to happen.
    Reduced "Female influence" on Male Solo generations.

    FAQ

    LORA
    SDXL 1.0

    Details

    Downloads
    316
    Platform
    CivitAI
    Platform Status
    Available
    Created
    11/29/2024
    Updated
    8/30/2026
    Deleted
    -
    Trigger Words:
    N0AI

    Files

    NovelstyleV1.5.safetensors

    Mirrors

    HuggingFace (1 mirrors)
    CivitAI (1 mirrors)

    Available On (1 platform)

    Same model published on other platforms. May have additional downloads or version variants.